Home

AudioSculpt

1. Analysis or LPC Analysis in the Analysis menu For the FFT a dialog will be displayed 39 AudioSculpt Users Manual Analysis Parameters Window size ve Fundamental Frequency 215 34 Window step 120 000 FFT Size lte4 9 Analysis Window Output format Factory Settings It lets you specify the parameters With the Output format button you can bring up the dialog displayed below Here you can specify the format in which results will be written to the text file It has two frames the top one says time and number the bottom one amp amp_db freq midicents real imag and phase midicents real imag phase You can select one of the fields by clicking on them a selected field is shown highlighted If you click a second time you unselect it The selected fields will be exported If time is selected the resulting file will 40 AudioSculpt Users Manual show the time of the analysis offset in the soundfile measured in seconds in the first line of the results If number is selected the first line will show the number of point of the analysis The number of points is one more than the FFT size divided by 2 The result of the analysis are given in columns If amp is selected one column will show the amplitudes of the spectrum envelope If amp_db is selected one of the columns will show the amplitudes measured in decibels With the freq field selected the frequencies will be written in a column and the midice
2. Rodet and Ph Depalle The algorithm looks for the most significant ones and ignores the other which are called spurious peaks In each short time amplitude spectrum the amplitudes of the maxima are detected The differences in am plitude between the maximum and the neighbor points must exceed a pre defined threshold All peaks that do not satisfy this constraint are eliminated Then you can ask to export only the peaks with the highest am plitudes The number of peaks that will be exported can be defined before the analysis You can apply the analysis on a part of the sound by selecting it before launching the Peak Detection com mand in the Processing menu The dialog for the peak detection is shown below It lets you set the threshold and the number of peaks to be registered Peak Detection Threshold in dE Humber of peaks Output Format Cancel Analysis Parameters Process With the Analysis Parameters button you can set the values of the analysis parameters If the window con tains a sonogram the results will be represented 42 AudioSculpt Users Manual It shows the results as breakpoint functions on the sonogram the results in each analysis window is shown as a small breakpoint on the sonogram and the breakpoints of successive analysis are connected with a line With the pointer tool in the toolbox you can grab a breakpoint and move it This way erroneous results can be modified by the user
3. Set Default Folders The Set Default Folders option is important only when you use the command line It lets you specify where AudioSculpt has to look for the sound and parameter files and where the console files generated by AudioSculpt will be saved By default AudioSculpt takes the folder where the application resides After you have specified the folders they will be taken as the default folders and will be memorized by AudioSculpt The three folders can be the same folder and can be changed at any time Page Setup The Page Setup command brings up the standard printer Page Setup dialog box It lets you set the format of the document to be printed Print The Print command starts printing the document AudioSculpt permits the printing of the sound and the sonogram image and also of the text files If the active window only displays a sound pane then the sound curve will be printed if the active window displays the sonogram then the sonogram without the sound curve will be printed To have a good reproduction it is important to mark the Colors Grey Scale option Quit The Quit command ends the AudioSculpt session Before closing the application a dialog will ask you if you want to save the last changes made to the text sound or sonogram 75 AudioSculpt Users Manual 2 The Edit menu Undo The Undo command un does the last command or key stroke executed This function is not available for sounds or s
4. The first line gives the starting time of the surface the number of frequency couples describing the surface the end time of the surface and the gain The following lines read the time of the frequency points and then two frequencies describing the lower and upper border of the surface The following figure shows another example of a surface displayed graphically this time In this figure an interpolation is done between the start time and the time of the two first points as well as between the time of the last two points and the end time frequency S000 2000 30dE 1000 61 AudioSculpt Users Manual 3 Cross synthesis With AudioSculpt new sounds can be synthesized based on the analysis of one or two existing sounds Cross synthesis is the combination of two sounds at the level of the spectral representation It is used in order to create different types of mixes between two sounds hybrid effects overlap etc AudioSculpt permits two types of cross synthesis cross synthesis source filter and generalized cross synthesis The two sounds parameters window size and window step can also differ according to the input channel If the window steps are different for each input the time scales of the spectral representations will be dif ferent as well Since during the synthesis stage the time scale of the primary input is used as a reference the time scale of the secondary input will be automatically changed giv
5. analyse taille de fen tre pas d avancement taille de la FFT type de fen tre 8 AudioSculpt Users Manual Parce que cette m thode garantit une ind pendance des parametres une grande finesse de manipulation est pos sible Toutefois comme les param tres utilis s par AudioSculpt pour analyse doivent tre choisis avec soin une documentation sp cifique sera publi e elle guidera l utilisateur dans le choix des param tres en fonction du son traiter et des r sultats vis s Notez que dans le menu d AudioSculpt les traitements Processing sont plac s avant l analyse Analysis la raison est que les traitements des sons disponibles poss dent leur propre mode d analyse tandis que les fonctions offertes par le menu Analysis permettent de recueillir leurs r sultats soit dans un fichier ex ploitable par d autres programmes soit par affichage graphique Les fonctions des deux menus sont donc in d pendantes Seul le sonogramme chappe cette r gle puisque sous le menu Analysis il permet de d finir des traitements qui seront ex cut s au moyen d une fonction du menu Processing 9 AudioSculpt Users Manual About this manual The manual is divided into 4 parts e Overview and Quick tour e Analysis e Processing e Menu reference As an introduction to AudioSculpt we will give an example of a short working session This example is meant to explain some of the basic featur
6. rca documentation e Research reports e Musical works Software AudioSculpt User s manual Second Edition April 1996 IRCAM Centre Georges Pompidou Copyright 1995 1996 Ircam All rights reserved This manual can not be copied in whole or part without the written consent of Ircam This manual was written by Peter Hanappeunder the supervision of Marie H l ne Serra and was produced under the editorial responsibility of Marc Battier Departement de la Valorisation Ircam Proofreading by Andrew Gerszo AudioSculpt Design Program Project supervision SVP Design Program Analysis modules Algorithms Documentation Coordination Manuals Philippe Depalle Chris Rogers Chris Rogers Gerhard Eckel Philippe Depalle Philippe Depalle Gilles Poirot Chris Rogers Jean Carrive Xavier Rodet Philippe Depalle Guillermo Garcia Ernst Terhardt Boris Doval Marie H l ne Serra Marc Battier Jean Carrive Peter Hanape Mario Mary Brice Pauset Marie H l ne Serra This documentation corresponds to version 1 0 or higher of AudioSculpt and to version 1 3 of SVP Apple Macintosh is a trade mark of Apple Computer Inc AudioSculpt is a trade mark of Ircam Second Edition April 1996 Ircam 1 place Igor Stravinsky F 75004 Paris Tel 01 44 78 12 33 Fax 01 44 78 15 40 E mail ircam doc ircam fr IRCAM Users group The use of this software and its documentation is restricted to mem bers o
7. In case the window only has a soundfile open and no sonogram analysis has been performed yet the results will be saved in a text file Again you can choose what results will be exported Therefore you click the Output format button and then in the dialog you can mark the desired data The formats of the results that can be chosen are e time in seconds e number the number of peaks desired e amp the amplitude of the peak e amp_db the amplitude of the peak in decibel e freq frequency of the peak e midicents frequency of the peak in midicents The peaks are ordered according to decreasing amplitude Fundamental estimation The Fundamental Estimation command looks for a fundamental frequency in every short time spectrum The algorithm was conceived by B Doval and X Rodet In every analysis window the peaks in the ampli 43 AudioSculpt Users Manual tude spectrum are detected then an estimation of the fundamental frequency is made based on these spectral peaks You can set a threshold to eliminate spurious peaks if the difference in amplitude between the high est peak and another peak in the amplitude spectrum exceeds the threshold the peak is eliminated from the analysis This reduction can be seen as a noise reduction technique To decrease the calculation time of the algorithm and to help it in its search you can specify a minimum and maximum fundamental frequency The algorithm will look only for a fund
8. ited with the internal editor of AudioSculpt for files smaller then 20000 characters or with any third party text editor for bigger files File name lengths The Macintosh operating system does not allow you to use names longer than 31 characters As AudioSculpt adds extensions to the names you choose such as out or console use short file names in any case no longer than 23 characters Additional documentation for AudioSculpt As of this writing there is currently one other manual available for AudioSculpt Cross Synthesis Hand book by Mario Mary This cross synthesis tutorial comes with a DAT cassette of examples available from Ircam s Forum Also make sure to consult the latest issue of AudioSculpt Newsletter for up to date infor mation 14 AudioSculpt Users Manual 2 quick tour So let s start When you ve installed AudioSculpt on your hard disk as explained in AudioSculpt newsletter you should find two items in the AudioSculpt Folder namely AudioSculpt 1 0 which is the application and a file called english msg Two soundfile are available separately Clarinet AIFF and Cym bal AIFF They are distributed with this manual You can also obtain them from Ircam s Forum Now launch the application by double clicking its icon The menu bar changes and should look like this i File Edit Processing Analysis Sound Windows You will find all functions and commands that AudioSculpt offers in the
9. 44 AudioSculpt Users Manual decibel freq the fundamental frequency and midicents the fundamental frequency in midicents Analysis based on perceptual criteria masking effects The Masking Effects command reduces the amplitude spectrum to a small number of frequencies which contribute to the tonal perception The algorithm is based on work by Terhardt It models the masking of spectral components which occur in our perception Both the amplitude and frequency of the components are taken into account in measuring the masking effects A spectral component can be masked and thus be come perceptually less significant because of the presence of a spectral component in the same frequency region As the human ear is not equally sensitive to all frequencies the perceptual importance of a spectral component also depends on its frequency First the maxima in the amplitude spectrum are detected using the same algorithm as in Peak Detection see above The algorithm then works on the remaining maxima Processing calculates the amplitudes of the peaks after modelling the masking effects Masked Amplitudes The effects of masking are subtracted from a peak s original amplitude in order to obtain the masked amplitude According to Terhardt s algo rithm a significant component has a positive masked amplitude in dB A component with a masked am plitude smaller than zero is considered as perceptually insignificant Due to the interaction of
10. 84 SVP 7 Synthesis window 85 T Terhardt 8 TerhardtE 2 45 Text file 37 39 Threshold in dB 44 86 87 88 time 25 Time scale modification 78 Transfer function 79 Transposition 22 78 True Spectral Pitch 88 U Undo 23 76 UNIX 82 Use Command Line 82 V Vocodeur de phase 8 W Weights 45 46 88 Window size 31 62 83 84 Window step 32 83 85 102 AudioSculpt Users Manual
11. Example Alpc50 e pic lt threshold gt lt nnumber gt peak detection with lt threshold gt threshold is given in dB and lt number gt the number of peaks to be detected Example Apic 3 n20 e mask lt list gt lt threshold gt lt nnumber gt Terhardt s masking algorithm with lt list gt of parameters masked amplitudes amp perceived frequencies weights weight or all of the previous all lt threshold gt threshold beyond which peaks are sought 25dB and lt number gt the number of peaks to be detected Example Amask amp weight 10 n20 e 0 lt fmmin gt lt fMmax gt lt FMAX gt lt snthreshold gt fundamental estimation with lt min gt minimum fundamental frequency 50 B 94 AudioSculpt Users Manual lt max gt maximum fundamental frequency 2500 lt MAX gt maximum harmonic frequency 7500 and lt threshold gt threshold is given in dB 50 Example Af0 fm100 fM1000 F2500 sn60 B lt start gt or b lt start gt Specifies the start time of the analysis within the sound file with lt start gt given as sample number integer or as time in seconds float The start time can be negative to add silence at the beginning Example B44100 B1 4 D lt value gt or lt file name gt Constant or time variable compression expansion with lt value gt the compression expansion factor and lt file name gt the name of the file describing the compression expansion factor Example D2 0 expan
12. OK to start the analysis Two new panes are added to the original window and then the sonogram is drawn gradu 17 AudioSculpt Users Manual ally Analysis Parameters Window size 1024 Fundamental Frequency 215 34 Window step 120 000 FFT Size ltde4 9 Analysis Window Synthesis window The complete sound is analyzed even if you have selected only a small part When the analysis is finished you should see something like the figure below on your screen The window now contains three separate panes e the upper pane is the sound pane as before except that it has no longer a time axis e in the lower left hand corner of the window is the sonogram pane The time axis in second is shown horizontally at the bottom of the pane The vertical axis on the right side of this pane is the frequency axis in Hertz The sound pane and the sonogram pane share the same time axis e The lower right hand pane is the short time amplitude spectrum pane It has a frequency axis which it shares with the sonogram pane and an amplitude axis on the bottom of the pane Each of these axis time frequency or amplitude can be rescaled as we ve done previously The three panes are separated by a thick grey line the pane separator Move the cursor above a pane sepa rator you ll see its shape change into two triangles Now click with the mouse key in the pane separator move it up or down and then release the mouse key the pane will res
13. V E S name soundfile Alpc 30 U Oa N4096 M4000 Whanning 12000 name outputfile Fast Fourier transform FFT analysis SVD v t s name soundrile ALit U 0a N40960 M4000 Whanning T2000 name outputfile Peak detection Syp vV t o name soundiile Apie 0 0 n5 U Oe N4096 M4000 Whan ning I2000 name outputfile Fundamental Estimation SVD EN t o name Soundrile ArO im50 TM2500 7500 S050 U 04 N4096 M4000 Whanning I2000 name outputfile 91 AudioSculpt Users Manual Masking effects SvE v t S name soundiile Amask weight Lov0 nl0 U 0a N4096 M4000 Whanning I2000 name outputfile Band filter Svp v t S name soundfile Fbande parameterfile A Z Jhamming N4096 M4000 Whanning I250 name outputfile Breakpoint filter Svp v t S name soundfile Fbreakpt parameterfile A Z Jhamming N4096 M4000 Whanning I250 name outputfile Formant filter interpolation on the parameters Svp v t S name soundfile Ffof parameterfile A Z Jhamming N4096 M4000 Whanning I250 name outputfile Formant filter interpolation on the filter responses Svp v t S name soundfile Ffifof parameterfile A Z Jhamming N4096 M4000 Whanning I250 name outputfile Surface filter Svp v t S name soundfile Fsurface parameterfile A Z Jhamming N4096 M4000 Whanning I250 name outputfile Cross synthe
14. for the result file During the processing a window shows you the progress of the computation Open the resulting file with AudioSculpt if the file is less than 20000 characters or with a text editor if the file is longer then 20000 characters The file should look like this 25 AudioSculpt Users Manual Clarinet AlFF out 0 550635 20 00ST 2e 206246 39 102013 31 540477 13 564026 31 077595 40 107166 34 137051 IT feeode 21023300 3 309496 26 051715 16 214020 11 621792 22 190704 20 40155 3 603706 213 0 000000 lee T01393 173 265961 l3 133362 0 929456 13 973314 175 546097 21 319733 349 320221 495 663269 ae 41614 soo 427124 326 371521 24 405151 Dad G55 266 830200 696 411 745 5 409177 642 804686 The first line shows the time of the analysis offset in the soundfile measured in seconds and the number of analysis points in this case the number of points is equal to the FFT size divided by 2 As of the second line the result of the analysis is given in the columns The first column shows the amplitudes of the spec trum measured in decibels and the second column shows the instantaneous frequencies of the spectral com ponents The command line Most of the AudioSculpt commands can be launched using such a command line A command line is a text line which closely resembles an UNIX like instruction In appendices A and B you find a list of all com mands and their options The folders in
15. in the command line constant cross synthesis lt X gt factor of the amplitude spectrum of the first sound 1 0 lt x gt factor of the amplitude spectrum of the second sound 1 0 lt Y gt factor of the phase spectrum of the first sound 1 0 lt y gt factor of the phase spectrum of the second sound 1 0 lt q gt factor of comultiplication 0 0 Example Gcross CrossPara g lt gain gt lt file name gt or lt tremolo gt lt mode gt lt file name gt Performs an amplitude modulation The modulation envelopes are described in a file containing the time amplitude information The following types are available e sinus sinus wave modulation e carre Square wave modulation e triangle triangular wave modulation e scie sawtooth wave modulation Examples ggain MyFile gtremolo sinus MyFile H lt sample rate gt Determines the new sampling rate of the sound file If this option is used do not use the options A and Z Example H32000 Displays the help file containing a description of the options B 96 AudioSculpt Users Manual l lt step gt or lt method gt lt file name gt or i lt step gt or lt method gt lt file name gt Specifies analysis step in samples The default values is 1 8 of the window size lt method gt is the method for reading the parameter lt file name gt Three methods are possible e sync ordered pairs in parameter file time fundamental frequency e dep ordered pairs in paramet
16. items are activated shown in black only 1f a text file is active Each of the four filters needs specific parameters which are specified from within a text file In this para graph we will give an example of the formant filter 27 AudioSculpt Users Manual formant filter parameters 0 0 1 440 0 50 1 0 2 440 0 50 880 2 100 2 0 3 4400 50 880 2 100 1760 2 200 Copy the values given in the figure above into your text window On each line these values should be sep arated by at least one space or Tab and each line should be terminated by a carriage return use the Return key When you ve copied the values choose the Formant Filter option In the dialog box you specify to which of the open soundfiles you wish to apply the filtering choose Cymbal AIFF Be sure that the option filter response is marked and then click Process and Save Give the new soundfile a name and press OK The new sound will appear on the screen in its own window Formant Filter Process sound Cymbal AIFF oF Formant filter interpolation Cancel filter response 2 filter parameters Process in place Analysis Parameters Process and Save A formant filter is a filter that is made up of second order filters Each second order filter is defined by spec ifying its central frequency its bandwidth and its gain These parameters can vary in time The first column of the parameter file says when to apply the filter parameters measured in sec
17. reader is referred to a theoretical work on signal processing Fast Fourier transform FFT The FFT is the algorithm used to compute the Fourier transform which converts a windowed signal into a spectrum The FFT gives over a certain time interval the decomposition of the sound into a sum of narrow band signals centered around successive frequencies that are linearly spaced over the frequency range go ing from 0 Hz to SR 2 SR stands for the sampling rate of the soundfile The number of frequency compo nents in which the sound is decomposed is equal to half the FFT size 33 AudioSculpt Users Manual Realtime Analysis Useful musical range ends here 2000 4000 e000 igi 1000 12000 14000 14000 15000 20000 ze E Amplitude and phase spectrum The FFT computes a complex value for each frequency component These complex values can be represent ed as a real and imaginary part or can be transformed into an amplitude and a phase value The amplitude corresponds to the energy of the spectral component The phase deals with the instantaneous frequency that is the small deviation of the frequency of the component relative to its central frequency The series of com plex values of every frequency component is called the complex spectrum and to every complex spectrum corresponds one amplitude and one phase spectrum The short time Fourier representation makes it possible to follow the time evolution of the amplitudes and phases o
18. sound interval that 1s analyzed In other words the temporary evolution within that sound interval are averaged into one short time spectrum It makes it hard to decide at what time exactly an event took place With smaller windows on the other hand the events are better captured Because of windowing the spectrum of the sound is convolved with the spectrum of the window and peaks can be introduced or reinforced in the sound spectrum This effect mainly depends on the type of window used The effect is smaller when the Blackman window is used than when the Hanning or Hamming window 1S used What window size and window type is best for a given sound 1s a delicate matter If the sound has spectral components that are close in frequency it is better to choose the Hanning or Hamming window to guarantee a better spectral resolution than with the Blackman window On the other hand if the sound has only a few distinct spectral components it is better to use the Blackman window to avoid as much as possible the in troduction of spurious peaks This is but a mere indication If the intermediate spectral representation 1s ma nipulated in view of a synthesis the choice of the window type and the window size can also depend on the type of processing applied FFT size When using the FFT algorithm the number of points in the windowed segment must be a power of two But 35 AudioSculpt Users Manual for a pseudo periodic sound the windo
19. the cut or copy function 77 AudioSculpt Users Manual 3 The Processing menu The Processing menu groups together the signal treatment functions Time Scale Modification With the Time Scale Modification command you can expand or compress the soundfile in the active win dow The expansion factor can be introduced in the text zone next to scale factor The duration of the sound is multiplied by the scale factor If the factor is bigger then 1 the sound will be expanded if the factor is smaller then 1 the sound will be compressed The scale factor is only limited by the amount of free space on your hard disk The re scaling process uses an FFT algorithm so it s important to specify the analysis parameters A time varying time scale modification is possible if you have a parameter file open The parameters de scribe the evolution of the expansion or compression in time You can choose the soundfile on which to apply the modification in the dialog box Sample Rate Conversion With the Sample Rate Conversion command you can modify the current sampling rate of the sound In the text zone next to desired sampling rate you can fill in the new sampling rate The conversion does not alter the length in time nor the pitch of the original sound Only the number of samples changes The size of the interpolation function can be set using the Resampling Parameters button Transposition The Transposition function transposes th
20. the structure of sound as well as to users who wish to study and change the properties of sounds The applications to musical com position are manifold because it is possible to change independently the amplitude frequency and time evolution of musical sounds whether they be natural instrumental or vocal AudioSculpt also offers ad vanced analysis results in the form of text files that can be interpreted by other programs Finally the user can operate directly on the graphical representation a procedure which simplifies complex sound manipu lations The operations of AudioSculpt The operations that AudioSculpt offers can be divided into two main groups analysis and processing Analysis The most significant difference of AudioSculpt with respect to SVP is the spectrographic analysis the sono gram mode The spectrographic analysis facilitates the processing because the properties of the sound can be observed visually The user can work directly in the window that contains the sonogram Further possible analyses are e the Fourier transform e the calculation of the spectrum envelope with the linear prediction technique e peak detection in short time spectra e an estimation of the fundamental frequency e the evaluation of the perceptual pitches and weights Terhardt s algorithm These analyses form the analysis modules they do not allow the modification of the sound The results are recordedeither in a text file o
21. trum corresponds to a large section in the soundfile If the amplitudes of the frequency components change over this section the change will not be reflected in the short time spectrum If the sound evolves rapidly a short window size is needed to accurately register the changes So the window size conditions the tem poral resolution of the analysis The frequency resolution also depends on the window size If a detailed spectral analysis is desired a large sound segment should be used and thus a large window If the window is small it is hard to decide precisely what the frequency components of the sound are This is reflected in a less precise spectral analysis There fore a compromise should be made Window step The next window or analysis frame is obtained by shifting the window in time by an amount called the window step In general and for reasons that come from the theory of spectral analysis successive windows should overlap that is the window step is always smaller than the window size and should not be bigger then 1 4 of the window size In AudioSculpt the window step is equal to 1 8 of the window size by default The temporal resolution of spectral evolution also depends on the window step In other words the details of the temporal evolution of a sound are best preserved when the window step is small It is possible to control the window size and window step separately so as to adapt the window size to the fre
22. type of filter amplitude response amp itude fiter am plitude response Spectrum LoTR HN Pre Were e band filtering preserves or removes the energy in a specified band of frequencies e breakpoint filtering is used to create a response curve that is more general than the band filtering type the curve is made up of linear segments of variable gain e formant filtering is useful for creating an envelope with a formant like structure One typical use of for mant filtering is to apply a succession of formant envelopes as in articulate speech to an instrumental sound This type of filtering is also useful for creating spectral resonances that can be controlled by their center frequencies and bandwidths This figure shows these three types of filters in AudioSculpt EALES band filtering feguercy THESS break point filter ng n wlll lures Len en RARE The formmart filtering All eee ANA He oe e surface filtering is used to modify the gain of a surface defined in the frequency time plane This figure shows an example of surface filtering in AudioSculpt 52 AudioSculpt Users Manual Pre Wer this polygon is called a surface S000 siete 30dE 1000 The type of filtering can not change during processing but the parameters that control these various filters cut off frequency gain bandwidth etc can All four types of filters can be specified in AudioSculpt using a text file AudioSculpt
23. 2 Lu on Play sound 17 89 number 25 Number of peaks 87 88 Fins pen oe Pointer tool 44 83 O PoirotG 2 Poirot G 7 Open 15 75 Poles 36 38 63 Option Power Macintosh 19 A 94 Preferences 77 B 95 Print 75 D 95 Process andsave 22 23 E 95 Process in place 22 23 F 95 Processing menu 21 78 G 96 ProTools 13 g 96 H 96 Q h 96 97 Quit 75 J 97 M 97 R N 97 real 25 nn 97 Realtime analysis 84 nS 97 Realtime sonogram scaling 85 O 97 Resampling parameters 23 73 R 98 Rodet X 2 resM 98 Rogers G 2 11 resW 98 Rogers C 7 S 98 Run Script 82 T 98 98 S trans 98 Sample rate conversion 73 78 v 98 Save 75 W 98 Save As 75 Z 99 Select All 76 Options 94 Selection tool 83 Output format 24 Serra M H 2 11 Ouvrir 15 Serra M H 7 Set Default Folders 75 P Set default folders 13 Page setup 75 Short term spectrum 32 Pass pencil 54 84 Show Clipboard 77 Paste 76 Show console 89 PatchWork 39 85 86 Sine play tool 39 84 PausetB 2 Sonogram 37 38 46 61 PCI 13 Sonogram analysis 17 82 83 Peak Detection 8 42 Sonogram filter 21 53 61 Peak detection 45 86 Sonogram filtering 81 Pen minus 84 Sound Designer Il 14 89 Pen plus 84 Sound folder 13 Perceptual pitch 45 46 88 Sound Manager 13 89 Phase 25 Sound menu 88 101 AudioSculpt Users Manual Sound output 89 Spectral envelope 36 59 Spectrogram filter 82 Stop processing 82 Stop sonogram analysis 83 Surface filtering 60 80 Surface tool 54
24. also offers a sono gram filter which is a surface filter that can be specified graphically One essential feature of AudioSculpt is that it permits temporal control of filtering Once the type of filter has been designated the user can specify any number of filter coefficients that are to take effect at a given moment in the sound AudioSculpt passes from one such configuration to the next by interpolation linear interpolation in time The sound s timbre is modified as a function of the evolution of the filters This dy namic control of filters can be used not only to bring about timbral evolution but also to create rhythmic effects by alternating different filter configurations For example the rapid alternation between low pass and high pass filters on a noisy input sound imposes on the sound a rhythm determined by the speed of the alternation The sonogram filter The sonogram filter is a surface filter but with AudioSculpt these surfaces can be drawn directly on the sonogram instead of working with text files Once you have the sonogram displayed you can take the sur face tool or a pen in the toolbox and draw the regions that will be filtered on the sonogram The surface tool is used in a different way than the pens With the surface tool you draw a surface and then give it a gain The gain defines how much a region in the sonogram will be amplified or attenuated With the pens on the other hand you can color a region This region w
25. am will change its shape into a pipette If you click in the sonogram an inspector window is shown on the screen This window has three fields e Time tells you the time of the clicked point in the pane e Frequency tells you its frequency e Amplitude indicates the amplitude of the point in the sonogram in decibel If you move the cursor vertically while holding down the mouse key you ll see that the curve in the short time spectrum pane is being traced The displayed curve is actually one slice of the sonogram pane at a par ticular time The amplitudes that are shown in shades of grey in the sonogram pane are shown here as an amplitude frequency curve Carine tA FF a ___ E Time 2 070 secs z Frequency 4918 4 Hz Amplitude 41 5 dB i The other tools of the toolbox are discussed in Part 2 and 3 of this manual 20 AudioSculpt Users Manual The sonogram filter Take the surface tool and draw a line freely on the sonogram while holding down the mouse key When you release the button the line will be automatically closed The surface that has been chosen is shown in grey Here is how to draw a surface Here is another way to use the surface tool Click once in the sonogram with the surface tool one end of a line is fixed the other end of the line can be moved with the mouse Click a second time the endpoint of the first line is fixed and a second line is drawn Continuing this way you gradual
26. amental frequency within the specified frequency region Also you can set a maximum harmonic frequency Spectral peaks with a frequency higher than this maximum fre quency will not be taken in account The analysis can be applied to a selected part of the sound Mark a region in the sound pane before choosing the Fundamental Estimation command in the Analysis menu In the dialog window the analysis parame ters can be set Minimum fundamental Maximum fundamental Maximum harmonic and Threshold in dB Fundamental Frequency Estimation Minimum fundamental Maximum fundamental S00 Maximum Harmonic 7500 Threshold in dE T Output Format Cancel Analysis Parameters Process The default values are 50 Hz and 2500 Hz for the minimum and maximum fundamental the default value for the maximum harmonic is 7500 Hz The ratio Maximum harmonic Minimum fundamental is limited for reasons of memory storage and calculation time The parameters for the spectral analysis can be set using the Analysis Parameters button If the sonogram 1s displayed the results will be shown on it as a breakpoint function The breakpoints can be edited with the Pointer tool If the sonogram is not displayed the results are registered in a text file You can then set the formats of the result using the Output Format button time in seconds number the num ber of peaks will be 1 amp the amplitude of the peak fundamental frequency amp_db the amplitude in
27. an important role in the recognition of vowels in speech The spectral envelope can also be used to filter another sound We will come back to this option in the section on cross synthesis source filter in the Processing chapter The LPC technique calculates a filter that has a frequency response which closely resembles the spectral envelope of the sound The transfer function that represents the filter 1s defined by a set of poles which go in pairs Each pair of conjugate poles represents a formant in the spectral envelope The number of poles equal to twice the number of formants controls the smoothness of the curve The number of poles in the LPC analysis can be specified by the user so as to increase the precision of the spectral envelope Thirty poles are used by default The higher the order of the analysis the more detailed the envelope is described and the more individual spectral components are modelled A detailed description is not always necessary one might be interested in the global structure of the envelope and only notice the most significant formants in case of across synthesis for example Once the poles of the filter are known the filter response or spec tral envelope can be calculated using the FFT algorithm So you can specify the number of points in which the spectral envelope is shown by setting the FFT size Here is a figure of the FFT spectrum and its spectral envelope 36 AudioSculpt Users Manual amp
28. ant the central frequency the gain and the bandwidth are given The following figure shows the evolution of the envelope applied to the sound s spectrum The formants are represented individually before they are added amp itude GC S20 1760 3920 tme t 0 0 sec INu DAAN fquerncs amplitude TBO 3920 tmet 1 0 sec Wan A ot eel 1760 Jai NA time t 20 sec J OX Frequency When you ask the Formant Filter in the Processing menu the dialog will ask you for the soundfile to amplitude which you want to apply the processing Surface filtering The surface filter modifies the amplitude of a surface defined in the amplitude frequency plane Every sur 60 AudioSculpt Users Manual face filter modifies the sound according to a fixed gain In the re synthesis the spectral components that fall in the regions defined by the surface will be amplified or attenuated according to the gain attributed to the surface The surfaces can overlap in that case their gains will be added The command Sonogram Filter see above offers a very similar filter that can be operated graphically The format of the parameter file is as follows e First line start time number of points end time and gain e Following lines time lower and upper border of the surface The following example describes a surface filtering surface filter parameters 0 0 4 3 0 40 1 0 1000 2000 2 0 1000 2000 3 0 2000 S000 4 0 4000 3000
29. cal interface The command lines can be saved in a text files so you can create a personal data base of command lines The lines can then be copied from the text file into the dialog box The Process button will start the analysis The command line window can stay on the screen inactivated while you are working with other files The text in the dialog box can be copied and cut into a text document such as a script see Run Script below The commands for the spectrogram filter and the sonogram analysis are not available through the command line In the annexes A and B you will find a complete list of the commands and their options A command can also be sent from within a text file Type the command in a AudioSculpt text file or open a saved text file and press the Enter key Run Script The Run Script lets you launch a list of processing tasks using the principle of the command line Different kinds of processing are possible and several soundfiles can be manipulated with one script Type the com mand line in a text file an AudioSculpt text file choose New Text Document in the File menu Each line must be separated by an empty line Choosing the Run Script command will start every process one after another If you choose Stop in the dialog box that appears on the screen the current process will be termi nated and AudioSculpt will jump to the next command The Stop Script button will stop all commands You can also run the script by pre
30. cond case the vocal impulse passes through the resonating instrument body The vocal timbre is determined by the instrument s colora tion Cross Synthesis Source Filter Filtered Sound Clarinet amp lFF Filtering Sound Cymbal AlFF Analysis Parameters Process and Save The Cross Synthesis Source Filter command in the processing menu brings up the dialog displayed below Filtered sound is the sound which will be filtered with the spectral envelopes of the filtering sound In this example the spectral envelope of the Cymbal AIFF is applied to the Clarinet AIFF To set the parameters of the analysis click the Analysis Parameters button The general parameters for both channels can be set window size window step analysis window as well as the number of poles in the LPC analysis the size of the FFT and the resynthesis window Remember that the higher the number of poles the more detailed the spectral envelope You may want to use a few poles when you only want to model the most important formants of the spectrum envelope 63 AudioSculpt Users Manual Analysis Parameters Channel 1 Channel 2 Window size re 1044 Fund Frequency 215 3 215 4 Window step 126 000 126 000 analysis Window Ipe order FFT Size 1024 9 synthesis Window Generalized cross synthesis Generalized cross synthesis is an extension of the previous cross synthesis The spectra of the two sounds initially represented by their r
31. e Stop button in the progress window or by selecting Stop Sonogram Analysis in the Analysis menu After the analysis a toolbox is brought on the screen Here follows a short description of the tools Selection Measure tool Sine play tool Pen plus The Selection tool The Pointer tool The Measure tool Pointer Harmonic measure tool Surface tool Pen minus With the Selection tool also called linear selection tool you can select a region in the soundfile With the pointer tool also called arrow selection tool you can select and drag graphical objects drawn on the sonogram With the pointer tool break points of a breakpoint function can be moved The measure tool gives you the coordinates of the cursor position in the sonogram The Harmonic measure tool The Harmonic measure tool shows the first 18 harmonic partials of a fundamental frequency in the form of horizontal lines on the screen 83 AudioSculpt Users Manual The Sine play tool The Surface tool The Pen plus The Pen minus Realtime Analysis The sine play tool plays a sine tone with a frequency and amplitude corre sponding to the clicked point in the sonogram This tool works only on Power Macintosh With the surface tool you can draw surfaces on the sonogram pane see also the Sonogram Filter These surfaces specify the regions of the sonogram that will be filtered With the pen plus also called Pass pencil you can draw fr
32. e X x Y y q for that time are given In the above example the amplitude and phase spectra of the output vary linearly between the two sounds between 0 and 1 second 65 AudioSculpt Users Manual Between and 2 seconds the output returns to the first sound and after 2 seconds the resulting sound is a combination of the two amplitude and phase spectra and of the multiplication spectrum Depending on the input generalized cross synthesis can create transitions that are quite complicated in terms of timbre Timbral interpolation between two different and contrasting sounds like for example a pitched sound and a noise sound is especially interesting 66 AudioSculpt Users Manual 4 Time scale modification and transposition Time scale modification Sound expansion and compression are techniques used in order to change the rate at which events occur in order to change the duration of a sound while retaining its character or to change both speed and duration simultaneously This expansion compression allows the modification of the time scale of events without changing the spectral information they contain Changing the speed at which a sound evolves makes it easier to perceive the details of the sound For ex ample the expansion of a piano sound allows the striking of the hammer on the strings and the resonance of the vibrating string to be isolated In the same way slowing down a recording of speech makes
33. e scale is multiplied by 1 5 and the resulting soundfile is thus 50 longer than the original Constant Timescale Modification scale factor EE Cancel Process in place Analysis Parameters Process and Save Similarly to reduce the total duration of the sound type for example 0 5 The resulting sound will then be half as long as the original one Time varying time scale modification It can be useful to have varying expansion or compression For example the attack of a percussive sound is left un modified while the rest of the sound the stable part is stretched The scaling parameters must be typed in a text file he format of the parameter file is as follows e First column time at which to apply the parameters e Second column scale factor On each line these values should be separated by at least one space or Tab Each line should be terminated by a carriage return use the Return key To expand a sound without modifying the attack portion type time scale modification haba D Co 0 0 0 4 1 5 3 1 In this example time stretching will begin at 0 4 seconds just after the end of the attack The scale factor varies from to 2 between 0 4 and 0 8 seconds and stays equal to 2 after that Between 0 and 0 5 seconds 68 AudioSculpt Users Manual the coefficient is equal to 1 and hence there is no stretching over this segment When the scaling parameter changes too abruptly the resul
34. e sound to a higher or lower pitch In the text zone you can specify the transposition in cents If the time correction option is marked AudioSculpt will adjust the length of the new soundfile to make just as long as the original soundfile The time scale factor will be calculated auto matically by AudioSculpt In case of a time correction it is necessary to set the analysis parameters 78 AudioSculpt Users Manual A time varying pitch transposition is also possible if the active window is a parameter file describing the evolution of the transposition In the menu besides Process sound you can specify the sound on which to apply the transposition The option time correction is also valid in case of a time varying transposition Breakpoint Filter This functions is only accessible if a parameter file is active The Breakpoint Filter is a filter which has a transfer function that is a correction of connected linear seg ments Three types of transfer functions can be defined amplitude phase or amplitude and phase transfer function Each type of transfer function defines a mode Mode Transfer function 0 amplitude 1 phase 2 amplitude and phase According to the mode the points of the breakpoint function are defined as frequency amplitude pairs as frequency phase pairs or as frequency amplitude phase triplets The amplitudes are measured in dB the phases in radians and the frequencies in Hertz The format of the parameter f
35. eal and imaginary parts are decomposed into amplitude and phase spectra which can be manipulated independently of one another This independent control over amplitude and phase gives a considerable variety of sound combinations One particularly attractive application is making hybrid sounds by applying the phase spectrum of one sound to the amplitude spectrum of another sound or crossfading that is the progressive transformation of the amplitudes and phases of one sound into the am plitudes and phases of another Here is a technical description of the operation For each analysis window the complex spectrum is broken up into an amplitude spectrum and a phase spectrum The pha se values are transformed into instantaneous frequencies If Al and 1 are the amplitude and phase spectra for a particular window and A2 and 2 are amplitude and phase spectra for a second sound at the same location then the amplitude and instantaneous phase for the synthesized frame are obtained using the following formula A XAI xA2 qAIA2 o Yol yp2 64 AudioSculpt Users Manual The formula basically states that the amplitude spectrum for the output signal is a linear combination of the amplitude spectra of the input signals augmented by a co multiplicative factor scaled by the coefficient q Given independent control over amplitude and instantaneous frequency it is possible to alter the frequencies of a sound s partials without changing
36. ed on algorithms that were previously developed at Ircam These analysis modules do not change the soundfile but bring out certain characteristics of the sound To give an example of such an analysis we will execute a short FFT analysis Select a small part of the sound file then choose the FFT Analysis command in the Analysis menu You should see the following dialog Analysis Parameters Window size ve Fundamental Frequency 215 3 Window step 120 000 FFT Size 1024 Analysis Window The analysis here is the same as the FFT analysis of the sonogram except that the results of the analysis are saved in a text file These text files can be very long and are meant to be read by other applications You can define what data will be saved and in what format Click the Output Format button and a window pops up 24 AudioSculpt Users Manual number amp amp db midicents real Imag phase It has two frames the top one says time and number the bottom one amp amp_ db freq midicents real imag and phase You can select one of the fields by clicking on them If you click a second time you unse lect it The selected fields will be exported to the text file In this example we ll select time number freq frequency and amp_db amplitude measured in dB Click OK Select the default settings for the analysis parameters and launch the analysis A dialog box will pop up and let you choose a home folder and a name
37. ediction LPC 36 Analysis display sonogram or text file 37 The sonogram 37 The toolbox 38 Text file output 39 Detection of partials peak detection and fundamental estimation 41 Peak detection 42 Fundamental estimation 43 Analysis based on perceptual criteria masking effects 45 Processing overview 47 Filtering 51 General principles 51 The sonogram filter 53 The surface tool 54 The pen plus and pen minus 54 Band filtering 55 Breakpoint filtering 56 Formant filtering 58 Surface filtering 60 Cross synthesis 62 The cross synthesis source filter 62 Generalized cross synthesis 64 Time scale modification and transposition 67 Time scale modification 67 Constant time scale modification 68 Time varying time scale modification 68 Improving the quality of a time stretched sound 70 Transposition 70 Cents scale 70 Constant transposition 70 Time varying transposition 71 Direct soundfile processing operations 73 Sample rate conversion 73 The File menu 74 New 74 Open 75 Close 75 Save Save As 75 Set Default Folders 75 Page Setup 75 Print 75 Quit 75 The Edit menu 76 Undo 76 Cut 76 Copy 76 Paste 76 Clear 76 Select All 76 Preferences 77 Show Clipboard 77 The Processing menu 78 Time Scale Modification 78 Sample Rate Conversion 78 Transposition 78 Breakpoint Filter 79 Band Filter 79 Surface Filter 80 Formant Filter 80 Sonogram Filter 81 Cross Synthesis Source Filte
38. eely on the sono gram to mark the regions that will be passed by the sonogram filter see Sonogram Filter With the pen minus also called Cut pencil you can draw freely on the sonogram to mark the regions that will be eliminated by the sonogram filter see Sonogram Filter This command works only on Power Macintosh The Realtime Analysis command lets you analyze sound in realtime You must have a microphone or any sound device connected to your computers sound input The incoming sound will be analyzed FFT analysis and the spectrum will be displayed in a new window You can set the number of points of the analysis in the menu on the left just below the title bar In the menu on the right just below the title bar you can change the way the spectrum is displayed as a curve as a series of lines or as a sonogram The frequency axis is shown below on the right you find the amplitude axis in case the spectrum is displayed as a curve or as lines Both axis can be rescaled Realtime Analysis me y i A 200 400 oO LE 1000 1200 1400 1600 1600 2000 2200 2400 2600 2500 E Analysis Settings The Analysis Settings dialog lets you specify the parameters for the FFT analysis and the synthesis Window size the window sizes determines the number of sound samples in each analysis Fundamental frequency allows to compute the window size automatically If you specify the win dow size the corresponding fundamental frequency wi
39. envelopes 62 AudioSculpt Users Manual The sequences are combined by multiplication of the FFT spectra by the LPC spectral envelopes This cross synthesis is called source filter because it is similar to a source filter synthesis system in which the output signal is calculated by passing the source signal through a time varying linear filter Such a sys tem models the voice glottis and vocal tract and certain instrumental sounds In this voice model the source signal corresponds to an excitation and the linear filter to a resonator It can either be a periodic im pulse train or a noise source The resonator is characterized by its resonant frequencies also called formants The resulting sound spectrum is the product of the spectrum of the source signal harmonic or noise spec trum with the frequency response curve of the resonator curve corresponding to the set of formants In AudioSculpt the signal source is a sound and the resonator is another sound For this operation only the spectral envelope for the filtering sound is evaluated The applications of source filter cross synthesis are numerous and are in general quite interesting One example now commonplace is the filtering of an instrumental sound by speech or the inverse In the first case one hears the instrumental sound filtered by the vocal tract shaped as it were by the phonemes of speech The instrument s timbre is deformed by the vocal formants In the se
40. er file time step value e pos sample position in the sound file is given for each window Examples 256 Isync StepFile J lt window type gt Specifies synthesis window type There are two possible windows e hamming default e hanning M lt window size gt or lt method gt lt file name gt or m lt window size gt or lt method gt lt file name gt Specifies analysis window size with lt window size gt expressed in samples 1024 or lt method gt the method for reading the parameter file lt file name gt The following modes are possible e sync ordered pairs in parameter file time fundamental frequency e dep ordered pairs in parameter file time window size Examples M2000 Mdep MyFile N lt FFT size gt or n lt FFT size gt Specifies FFT size in samples 1024 The FFT size must be greater than window size see M Examples N4096 N8192 nn Do not normalize output sound file default is normalize Spectral normalization for analysis output default is no normalization O lt mode gt lt list gt or lt mode gt lt file name gt Specifies the output format with lt mode gt is the output type Possible modes types are e a for ASCII e b for binary default lt list gt is a list of strings separated by number time amplitude frequency phase etc lt file gt is a text file where output format is specified Examples Oa number amplitude phase Oa foo format R lt sample ra
41. es of AudioSculpt and its graphical interface If you are already familiar with AudioSculpt you can skip this chapter In Part 2 you will find a more detailed description of the analysis In Part 3 the processing functions are explained In Part 4 you will find a list of all the commands in the menus 10 AudioSculpt Users Manual Part 1 Overview Introduction to AudioSculpt AudioSculpt is a sound processing software package that was designed and developed at Ircam It is based on SVP Super Vocodeur de Phase a program created by Gilles Poirot and Philippe Depalle of the Anal ysis Synthesis team Under the supervision of Gerhard Eckel and with the coordination of Marie H l ne Serra AudioSculpt has been revised and optimized by Chris Rogers of the Interface and Sound Represen tation team and other extensions were made by Jean Carrive Based upon SVP 1 3 for Macintosh AudioSculpt is a graphical interface for representing sounds as well as for the control of transformations time expansion or compression transposition filtering cross synthesis etc It contains all of the commands previously available in SVP 1 3 for Macintosh and adds some new possibilities Its main new features are the spectrographic analysis sonogram and the filtering by drawing polygon surfaces directly on the sonogram 11 AudioSculpt Users Manual AudioSculpt is an analysis synthesis program useful to researchers interested in
42. esign Card it can be used to play the soundfile See the AudioSculpt Newsletter to find out more about playing back soundfiles The Windows menu In the Window menu you ll find the list of all the soundfiles that are actually open The active soundfile is marked If you choose another soundfile in this menu its window will come to the front and will be made the active window Show Console The Show Console command displays the console window on the screen Annex A Command lines All the commands of AudioSculpt that are available through the menu bar can be launched using the command line The only exceptions are the sonogram analysis and the sonogram filter Certain pro cessing options are available only through the command line and are not available as through the com mands in the menus This annex gives the equivalent command line of the commands in the menus The command line always starts with the identifier svp followed by a number of options An op tions is preceded by a minus sign and can be followed by a set of parameters and values The op tions can be given in any order though the last parameter should be the name of the output file Since AudioSculpt can treat two files at once the options concerning the first sound file are written in cap ital letters and the those for the second file in small letters The options are explained in more details in annex B Linear prediction analysis LPC analysis Svp
43. etal lic or flanged character the problem may be the lack of resolution in the spectral analysis In this case the window size which is 1024 samples by default or 23 2 milliseconds at 44100 Hertz should be increased as much as possible Suppose that the input sound is periodic with a fundamental of 220 Hz and a sampling rate of 44100 Hz The window size should be equal to 4 or 5 times the period of the sound or 4 44100 220 samples approximately 800 samples In AudioSculpt five times the period length is selected as the window length see Part 2 Analysis FFT Transposition When creating new sounds one might want to shift a recorded sound to a higher or lower frequency The whole sound can be transposed with a constant interval or the interval can vary in time This way it also possible to adjust slight frequency deviations of the sound according to the desired frequency AudioSculpt offers a constant transposition and a time varying transposition Both types of processing specify the trans position interval in cents The cent unit stands closer to musical notation then Hertz The next paragraph tells more about it Cents scale When we divide the interval of an octave into 1200 equal steps footnote equal here means according to the frequency ratios the distance between f3 and f2 is equal to the distance between f2 and f1 1f f3 f2 equal to f2 f1 one step corresponds to one cent If two frequencies are one cent apart their freque
44. exception is the sonogram filter for which the sonogram analysis should be run first The function of the analysis modules is to provide a better understanding of the sound The analysis results can help for further processing or can be used in other applications such as PatchWork a Lisp based com puter aided composition package also available from Ircam s Forum Playing the soundfiles If you use a Macintosh equipped with a 16 bit audio system like the Power Macintosh series you can use your internal Digital to Analog conversion hardware which will play 16 bit stereo soundfiles AudioSculpt will play your soundfiles directly through this hardware with CD quality Make sure that the Sound Man ager in your Macintosh system is version 3 or higher Version 3 comes with System 7 5 If you use an older system you will still able to install Apple Sound Manager 3 0 If your machine is equipped with 8 bit audio hardware you should use a sound processing board such as Digidesign s Audiomedia I and II Sound Tools I and IT ProTools or any other 16 bit board AudioSculpt will still be able to play your soundfiles through the Macintosh hardware but sound quality will be poor Newer Macintoshes with PCI bus may play directly through the Sound Manager Third party boards for the PCI bus such as Audiomedia IT or ProTools by Digidesign should be available in 1996 About soundfiles and data files When you first use AudioSculpt you should set
45. f the Ircam software users group For any supplementary infor mation contact D partement de la Valorisation Ircam Place Stravinsky F 75004 Paris Tel 01 44 78 49 62 Fax 01 44 78 48 15 40 E mail bousac ircam fr Send comments or suggestions to the editor E mail bam ircam fr Mail Marc Battier Ircam D partement de la Valorisation Place Stravinsky F 75004 Paris http www ircam fr musinfo To see the table of contents of this manual click on the Bookmark Button located in the Viewing section of the Adobe Acrobat Reader toolbar Contenu R sum 7 Les op rations de AudioSculpt 8 Analyse 8 Traitement 8 About this manual 10 Introduction to AudioSculpt 11 The operations of AudioSculpt 12 Analysis 12 Processing 12 Playing the soundfiles 13 About soundfiles and data files 13 Soundfiles 14 Parameter files 14 File name lengths 14 Additional documentation for AudioSculpt 14 A quick tour 15 Opening a soundfile 15 Sonogram analysis 17 The sonogram filter 21 Process in place or Process and save Transposition 22 The analysis modules 24 The command line 26 Processing and parameter files formant filter 27 What s next 29 Analysis 30 General principles of the short time spectral analysis 30 Window size 31 Window step 32 Choice of window 33 Spectral analysis 33 Fast Fourier transform FFT 33 Amplitude and phase spectrum 34 Windowing effects 34 FFT size 35 Linear Pr
46. f the frequency components of a sound Most filters are designed to manipulate the amplitude spec trum The only exceptions are the breakpoint filter and the generalized cross synthesis as they actually ma nipulate the phase spectrum Changing the way the phase spectrum evolves in time will result in a modification of the frequencies of the components Windowing effects Windowing has two kinds of side effects on the FFT spectrum These side effects depend on the window size and on the type of window The first side effect is in the frequency resolution and the second side effect is that windowing can introduce spectral components where there were nt any previously Because of the windowing the sound spectrum is convolved with the spectral representation of the window signal This has an effect on the spectral resolution Two spectral components that have only a small differ ence in frequency between them can be merged into one spectral peak in the short time spectrum and it be comes impossible to detect them separately The resolution depends on both the window size and the window type The spectral components of a sound are best isolated when the window size 1s large The num ber of points in the window should be greater than or equal to a multiple of the number of points in a period 34 AudioSculpt Users Manual This multiple depends on the type of window For a Hamming and Hanning window it is equal to 4 The multiple for
47. he spectra provided by the analysis the rate at which the spectrum evolves is thus changed A sound can thus be expanded or compressed without altering its frequency content A frequential modification al ters the composition of a spectrum at a given moment Frequential modifications serve to filter the sound by multiplying the spectrum by the frequency response curve of a filter or to transpose the sound to a higher or lower frequency The filter is used to modify the distribution of energy in the spectrum and thus the timbre of the input sound As in the example the filter ing can vary during the processing AudioSculpt also allows the possibility of processing two sounds at once and of combining their spectral representations This combination involves the generated spectra and thus makes cross synthesis possible This option which can be used in order to generate fairly complex combinations of two sounds is particu larly useful from a musical standpoint This illustration shows thge two types of sound modification temporal and frequential fiterning amp itude at time scale modification je frequence Once the series of spectra possibly modified 1s obtained the resulting sound signal can be synthesized using the inverse Fourier transform From the intermediate spectral representation the sound is synthesized and written to a file As was the case in the forward Fourier algorithm the inverse Fourier algo
48. hich the spectrum is displayed is always one more then the FFT size divided by 2 in the analysis window menu you can specify the window type used in the analysis Three types of window are proposed Hanning Hamming and Blackman If you use the LPC analysis the number of poles have to be set The number of poles should be between 10 and 200 by default 30 poles are used The Factory Settings button sets the parameters to the default value The sonogram analysis concerns the whole soundfile even if you have selected but a small part On launching the analysis two new panes are added to the window beneath the existing sound pane the left pane is the sonogram pane the right pane 1s the short time spectrum pane You can stop the analysis with the Stop button in the progress window or by selecting Sonogram Analysis in the Analysis menu The toolbox Selection he a Pointer Measure tool gt AE lt a Harmonic measure tool Sine play tool gt AS After the analysis is done a toolbox is brought on the screen Some of the tools help you inspecting the sono gram others can be used to manipulate the sonogram to create a new sound and will be discussed in the Processing chapter Here follows a short description of the analysis tools The Measure tool gives you the coordinates of the cursor position in the sonogram Clicking with the 38 AudioSculpt Users Manual mouse in the son
49. ift key and select a region in the time axis with the mouse Release the mouse key and you will see that the selected region is stretched to fit in the window size Double clicking the axis brings it back to its initial size 16 AudioSculpt Users Manual Clarinet AlFF The signal window behaves like any standard window you can resize it move it on the screen and the scroll bar lets you display the hidden part of the sound Choosing Play Sound in the Sound menu will let you hear the sound A quicker way to do this is to press the space bar Pressing it again will stop the sound from playing Let s select a part of the sound To do so click in the sound pane and mark the region with the mouse while holding the mouse key down The selected region is now highlighted in black Clarinet AlFF Press the space bar and yov ll notice that you only hear the selected sound Pressing the command key the clover apple key and the spacebar at the same time will let you hear the complete sound again Sonogram analysis Now that you can get around with soundfiles let us analyze them In general AudioSculpt works with the sound in the active window You know the window is active when its title bar has its normal appearance Choose Sonogram Analysis in the Analysis menu A dialog box will appear and ask for the analysis pa rameters The default settings will do for now so click the button Factory Settings and then click
50. ile is as follows e First column time at which to apply the parameters e Second column the mode of the filter 0 1 or 2 e Third column the number of breakpoints e Following columns the coordinates of each breakpoint frequency amplitude for mode 0 frequency phase for mode 1 frequency amplitude phase for mode 2 The dialog box allows you to choose the sound on which to apply the filtering Band Filter This functions is only accessible if a parameter file is active The Band Filter eliminates the spectral components that fall into a stop band and does not affect the com ponents that fall into a pass band These bands are defined by the frequencies that border them and whether the first one is a pass band set the flag to 0 or a stop band set the flag to 1 The nature of the following bands are specified automatically since pass and the stop bands alternate The values of the border frequen cies can vary in time but the number of bands in the filter and their mode pass or stop cannot change 79 AudioSculpt Users Manual The format of the parameter file is as follows e First column time at which to apply the parameters e Second column the number of border frequencies e Third column flag specifying if the first band is a stop or pass band e Following columns the values of the border frequencies The dialog box allows you to choose the sound to which to apply the filtering Surface Filter This function is
51. ill either be completely cut out of the sonogram in case of the pen minus or in case of the pen plus this region will be left untouched but the rest of the sonogram will be eliminated These kinds of tools are different which explains why their surfaces are not shown si multaneously on the sonogram Shifting from a surface tool to a pen will hide the surface tool s regions they are kept in memory though and display the previously drawn pen traces Shifting back to the surface 53 AudioSculpt Users Manual tool will display its surfaces again and hide the pen s surfaces Here s a description of how to manipulate these tools The surface tool Pen plus With the Surface tool you can draw surfaces on the sonogram pane see also the Quick tour at the begin ning of this manual These surfaces specify the zones of the sonogram that will be filtered The surface tool offers two ways of drawing surfaces e You can freely draw a line with the mouse holding the mouse key down When you release the mouse key the surface will be automatically enclosed e You can specify the surface as a polygon Every click with the mouse defines one angle point and one side of the polygon To close the polygon you must double click You can select and move a surface with the pointer tool A selected surface has a flashing borderline To delete a surface you select it and then press the option and delete key simultaneously With the pointer t
52. ing a time expansion compression For example if sound 1 has a window step of 256 samples then the time interval between the short term spectra is equally 256 samples If sound 2 has a window step of 128 samples then the interval between the spectra is 128 samples In the cross synthesis the first short term spectrum of sound 1 is combined with the short term first spectrum of sound 2 the second spectrum is combined with the second spectrum and so on Since the spectra of sound 2 has a time interval that was initially half the time interval of sound 1 the spectral data of sound 2 will be stretched by a factor 2 This corresponds to a dilation by a factor 2 The cross synthesis source filter Cross synthesis source filter consists of multiplying the FFT spectrum from the primary input by a spectral envelope resulting from linear predictive analysis on the secondary input The first sound 1s thus filtered by the spectral envelope of the second This type of operation is very close to the source filter synthesis model in AudioSculpt the source is the first sound and the filter is the spectral envelope of the second A common application is the filtering of an instrumental sound by the voice so as to give it a speaking instrument effect One of the sounds is analyzed by the FFT and the other by linear prediction LPC Two parallel spectral representations are obtained one being a series of FFT spectra the other a series of LPC spectral
53. int Filter Process sound J Cancel Process in place Analysis Parameters Process and Save In this type of filtering the frequency response curve for a filter is defined as a set of points connected by lines There are three possible modes of operation for this type of filter amplitude response phase response 56 AudioSculpt Users Manual or amplitude and phase response Mode Operation 0 amplitude response i phase response 2 amplitude and phase response Depending on the mode the points will be ordered pairs of the form frequency amplitude frequency phase or triplets of the form frequency amplitude phase Amplitudes are given in decibels phases in radians and frequencies in Hertz The following parameter file specifies breakpoint filtering that passes from a configuration in which the ex treme regions of the spectrum fundamental and high harmonics are attenuated to the inverse configuration at 1 5 sec and then comes back to the initial filter at 3 sec Only the amplitude response is specified in this filter mode 0 and the filter is defined by amplitude frequency pairs The format of the parameter file is as follows e First column time at which to apply the parameters e Second column the mode of the filter 0 1 or 2 e Third column the number of breakpoints e Following columns the coordinates of each breakpoint frequency amplitude for mode 0 frequency phase for mode 1 frequency a
54. it easier to examine the articulation of the phonemes This type of processing provides a useful effect by deforming the time scale of the micro events which make up a sound Changing the duration of a sound is often necessary when attempting to synchronize digital sounds or when one needs to insert a sound of a pre determined duration into a piece of music Since it 1s impractical and sometimes impossible to sample a sound of exactly the desired duration whether the sampling is done from a recording or from a live player one must be able to change this duration arbitrarily without degrading the quality of the sound too much AudioSculpt offers a constant time scale modification the argument is then a coefficient referring to a con stant stretch factor You can also obtain a time varying time scale modification when you use a file of co efficients In both cases a coefficient bigger than 1 increases the length of the sound whereas a coefficient between 0 and 1 reduces it the duration of the sound is multiplied with the scaling factor To obtain a good sound quality it is important to set the analysis parameters properly 67 AudioSculpt Users Manual Constant time scale modification You can modify the length of a soundfile with a constant factor when choosing the Time Scale Modification command in the Processing menu In the dialog box below next to scale factor you can type the factor to be used In this example the tim
55. itude spectral envelope FFT spectrum reg were Analysis display sonogram or text file AudioSculpt offers two ways of inspecting the results of the spectral analyses The results can be displayed as an image on the screen in what is called the sonogram see also the Introduction A quick tour The re sults can also be displayed in ASCII format and saved in a text file The sonogram In the Analysis menu you find the command Sonogram analysis It brings a a dialog on the screen Analysis Parameters Window size 1024 Fundamental Frequency 415 3 Window step 120 000 FFT Size 1024 Analysis Window Synthesis window 37 AudioSculpt Users Manual The dialog lets you specify the analysis parameters and the type of spectral analysis Here is a brief survey of the parameters Window size the window sizes determines the number of sound samples in each analysis Fundamental frequency allows you to compute the window size automatically If you specify the Window step FFT Size Analysis window window size the fundamental frequency that is calculated will be dis played The window size is set to five times the length of the period that corresponds to the fundamental frequency the window step determines the time interval between two succesive analy ses The time interval is measured in number of samples the FFT size determines the number of points of the analysis The number of points in w
56. ive J 2 11 Carrive J 7 Clear 76 Close 75 Command line 82 91 94 Constant transposition 23 Copy 76 Cross synthesis generalized 62 source filter 62 63 81 Cross synthesis source filter 81 Cross synthesis generalized 81 Cut 76 Cut pencil 84 D Depalle Ph 2 11 36 Depalle Ph 7 Desired sampling rate 73 Digidesign 13 89 DovalB 2 E Eckel G 2 11 Eckel G 7 Edit color palette 85 Edit menu 76 Excel 39 85 86 F Factory settings 38 83 Fast Fourier transform 33 FFT 8 33 analysis 23 24 85 size 26 33 35 83 85 Fichier 15 File 15 74 Filter band 51 52 55 79 breakpoint 51 52 56 79 fof 59 formant 52 58 80 pass band 51 sonogram 61 81 stop band 51 surface 51 52 60 80 Filter response 28 Fof 59 Fondamental 8 Formant filter 28 58 Formant filtering 80 freq 25 Fundamental Estimation 8 87 Fundamental estimation 43 Fundamental frequency 83 84 G GarciaG 2 Generalized cross synthesis 81 Get sound info 89 H Hamming 35 38 85 Hanning 35 38 85 Harmonic measure tool 39 83 l imag 25 L Linear prediction LPC 36 Linear selection tool 83 LPC 8 33 86 M Mary M 14 Masked amplitudes 45 46 88 Masking Effects 45 88 Masking effects 45 Max 39 85 86 100 AudioSculpt Users Manual Maximum fundamental 44 87 Maximum harmonic 44 87 Measure tool 38 83 midicents 25 Minimum fundamental 44 87 N Phase spectrum 51 64 Pics 8 NE 8
57. ize accordingly This way you can resize each of the panes and even hide one of the panes when you move the pane separator against the win dow border 18 AudioSculpt Users Manual Clarinet AlFF Sound pane Short time amplitude spectrum pane sonogram pane The scroll bar at the bottom of the window lets you scroll the sonogram Grab the handle in the scroll bar with the mouse and move the sonogram directly with the mouse in either direction On a Power Macintosh the sonogram will follow in real time on slower computers refreshing the sonogram image might be slower The sonogram shows the frequency contents of the sound in time The sound is decomposed into frequency components and the sonogram displays the evolution of the amplitudes of these components in time The grey scale corresponds to the amplitude white means a low amplitude black means a high amplitude when using the default white to black scale the color scales are set with the color palette 19 AudioSculpt Users Manual You should also find a toolbox palette on your screen Here it is with the names of the tools Selection M gt if he lt at Pointer Z lt a Harmonic measure tool ar a _ Surface tool Measure tool gt Sine play tool gt Pen plus gt As an example let s see what the Measure tool does Select the Measure tool by clicking its icon Moving the cursor above the sonogr
58. ll be displayed The 84 AudioSculpt Users Manual window size is set to five times the length of the period that corre sponds to the fundamental frequency Window step the window step determines the time interval between two analy ses The time interval is measured in number of samples FFT Size the FFT size determines the number of points of the analysis The number of points in which the spectrum is displayed is always the FFT size divided by 2 Analysis window in the analysis window menu you can specify the window type used in the analysis Three types of windows are proposed Han ning Hamming and Blackman Synthesis window in the synthesis window menu you can specify the window type used in the synthesis Two types of windows are proposed Han ning and Hamming The Factory Settings button sets the parameters to the default values Realtime Sonogram Scaling If the Realtime Sonogram Scaling command is marked the sonogram will be re drawn while you re scale the time or frequency axis If the command is unmarked the sonogram will temporarily disappear while you rescale This allows a faster rescaling on slower Macintoshes To mark the command you click it once to unmark it you click it a second time Edit Color Palette The Edit Color Palette command brings the color palette onto the screen With the color palette you can change the contrast of the sonogram You can choose between four types of black and white or c
59. ly draw a polygon Double clicking will close the surface ee nee Wet AI era gi Now that you can draw surfaces we ll continue with what it s designed for filtering Start selecting a sur face with the pointer tool and type a number 10 for example on the keyboard The number will be written in the label attached to the surface The number in the label specifies a gain expressed in dB All the fre quencies that are covered by the surface will be amplified according to this gain when the sonogram filter is executed You can type a negative gain also type 10 for example and the area that is covered will be attenuated Give a gain to every surface you ve drawn and then choose Sonogram Filter in the Processing menu You ll be asked to give a name to the new file in the dialog box and then click the Save button A process window comes on the screen and shows you the progress of the computation The resulting sound will be displayed in a new window 21 AudioSculpt Users Manual Audiasculpt 1 0 Donna ipii Bureau Cancel sare soundfile Normalize Just one last word before we finish this tour You can save the sonogram by choosing Save in the File menu The name of the new file will be asked in a dialog box The analysis file will be saved separately from the soundfile If you open the analysis file AudioSculpt will automatically open the soundfile with it AudioSculpt is able to find back the so
60. mplitude phase for mode 2 mode here O amplitude response number of pairs frequencyamplitude pairs time breakpoint filter 0 0 0 3 0 OF 100 60 1000 o fi 1 5 0 4 0 60 100 UO 1000 BO0 20000 80 3 0 0 3 0 OF 100 60 1000 o In this type of filtering operation the number of pairs defining the frequency response curve for the filter can change in time They change linearly from one line to the next The mode of operation however am plitude phase or amplitude and phase cannot change 57 AudioSculpt Users Manual O 100 1000 f O 100 1000 20000 f A O dB O dB t 3 0 60 dB O 100 1000 f Formant filtering Formant Filter Process sound Son origi nalr Formant filter interpolation type fe filter response Cancel C3 filter parameters Process in place Analysis Parameters Process and Save In this mode a filter is defined as the sum of a certain number of formants These are technically called 2 58 AudioSculpt Users Manual pole recursive filters each of which is described by a central frequency in Hertz a gain in decibels and a bandwidth in Hertz The gain corresponds to the height of the maximum for the curve The resonant fre quency is the frequency at which the curve attains its maximum and the bandwidth corresponds to the half width of the curve when the amplitude is 3 dB below the peak value This figure formant filter response curve 2 pole
61. n It models the masking effects of the filtering in the internal ear In the dialog four different analysis can be chosen Weights Masked amplitudes Perceptual pitches All Further parameters are Threshold in dB Number of Peaks the weights measure the spectral importance of the spectral com ponents The weight is always larger then zero the amplitudes of the spectral components after the masking effects have been taken in account The amplitudes of the signifi cant spectral components have a positive value gives an estimation of the True Spectral Pitch the pitch of the spectral components as they are perceived by the listener exports the weights the masked amplitudes and the perceptual pitches the difference in amplitude between a peak at the 1 th index and the amplitudes at the neighboring points at indexes 1 3 1 2 1 2 1 3 must exceed the threshold Peaks that do not satisfy the constraint are neglected The default value is 25 dB You can specify how many peaks you want to be registered Only the m peaks with the highest amplitude will given as result By default the 5 highest peaks are given If the sonogram is displayed the Perceptual Pitch will be shown on the screen as a breakpoint func tion if the sonogram 1s not displayed the results will be saved in a text file In that case the formats of the result can be set using the Output Format button time in seconds number the number of s
62. n a ten percent margin 22000 Hz without degrading the sound quality Moreover sampling at a lower rate can save you space on the hard disk You can change the sampling rate of the soundfile with the Sample Rate Conversion command in the Pro cessing menu You type the new sampling rate in the text zone beside desired sampling rate Using the Resampling Parameters button you can set the length of the interpolation function measured in samples Sample Rate Conversion current sampling rate 44100 0 desired sampling rate 32000 Process in place Resampling Parameters Process and Save 73 AudioSculpt Users Manual Part 4 Menu reference 1 The File menu The File menu is composed of the standard file management functions New The New command creates a new text document 74 AudioSculpt Users Manual Open The Open command lets you open a soundfile a sonogram or a text file The sound formats that AudioSculpt recognizes are the AudioSculpt AIFF and Sound Designer format The size of text files are limited to 20 000 characters If you open a sonogram analysis the corresponding soundfile will be opened as well Close The Close command closes the active window text sound or sonogram window If the document has un saved changes AudioSculpt asks you if you want the changes to be saved Save Save As The Save and Save As commands lets you save a soundfile a sonogram or a text file
63. ncies have a ratio of 21 1200 A distance of 100 cents corresponds to a half step 200 cents corresponds to 1 whole step and so forth Constant transposition For a transposition with a constant interval choose the Transposition function in the processing menu A dialog box is displayed In it you can specify the transposition in cents In this example the sound is trans posed by 700 cents or a perfect fifth 70 AudioSculpt Users Manual Constant Transposition number of cents i e time correction Cancel Resampling Parameters Process in place Analysis Parameters Process and Save If the time correction option is marked AudioSculpt will adjust the length of the new soundfile to make it just as long as the original soundfile The time adjustment requires a time scale modification AudioSculpt will automatically calculate the time scale factor In case of a time correction it is necessary to set the anal ysis parameters Time varying transposition A time varying pitch transposition is possible if the active window is a parameter file describing the evolu tion of the transposition The format of the parameter file is as follows e First column time at which to apply the parameters e Second column interval of the transposition in cents Here is an example of such a parameter file time varying transposition HE T 1 z2 So Q To run the transposition choose the Transposition comma
64. nd A dialog box will appear In the menu besides Process sound you can specify the sound on which to apply the transposition 71 AudioSculpt Users Manual Time varying Transposition Process sound Cymbal AlFF x time correction Cancel Resampling Parameters Process in place Analysis Parameters Process and Save If you check the time correction checkbox as in the illustration above AudioSculpt will ajust the duration of the output sound so that it matches the original sound duration You ll see that when you select time correction the Analysis Parameters dialog box becomes available It s a good idea to open it and set the parameters according to your sound s characteristics 72 AudioSculpt Users Manual 5 Direct soundfile processing operations Sample rate conversion Most often sound is sampled at 44100 Hz This is the sampling rate used for Compact Discs for example Other sampling rate standards are 32000 Hz and 48000 Hz The sampling rate is mainly determined by the highest frequency in the sound The half of the sampling rate is called the Nyquist frequency A sampled sound should not contain any frequency components with a fre quency higher then the Nyquist frequency So if your sound is sampled at 44100 Hz the highest frequency component will be 22500 Hz In other words if the sound has no components above let s say 10000 Hz then you can sample the sound at 20000 Hz or give
65. nt dans la zone affichage du sonogramme obtenu Les autres op rations d analyse sont e la Transform e de Fourier FFT e l analyse par pr diction lin aire LPC e la recherche de pics dans le spectre a court terme Peak Detection e la d tection de la fondamentale Fundamental Estimation e l valuation de la ou des hauteurs virtuelles algorithme de Terhardt Elles constituent le module d analyse Le module d analyse ne permet pas de modifier les sons les r sultats d analyse sont recueillis dans un fichier texte ou affich s en superposition sur le sonogramme Un effort particulier a t fait pour la pr sentation des r sultats d analyse qui sont ais ment mis en forme par I util isateur Traitement Les commandes de traitement permettent de d finir des modifications a effectuer sur le son Elles reposent sur le principe du vocodeur de phase dans lequel le son est d abord soumis a une analyse de son volution spectrale dans le temps le traitement sp cifi alt re les r sultats de l analyse et le son r sultant est obtenu par resynth se La plupart des traitements de son sont effectu s a partir d une analyse synthese par vocodeur de phase Seuls la transposition sans correction et le r chantillonnage n utilisent pas analyse synthese Dans tous les traitements a base d analyse synth se pour obtenir un r sultat satisfaisant 11 convient de choisir les param tres de l
66. nts field gives these frequencies in midicents In case of the FFT the instantaneous frequencies are exported That is the deviation of spectral components from the central frequencies are tak en in account For the LPC analysis the frequencies of the analysis points the central frequencies are ex ported The parameters real and imag give respectively the real and the imaginary component of the complex spectrum If phase is selected the phase spectrum will be exported The dialog for the LPC analysis looks like this LPC Analysis LPC order Output Format Cancel Analysis Parameters Process You can set the order of the LPC and with the Analysis button the parameter values for the FFT algorithm can be given as well The Output Format shows most of the options that were discussed for the FFT anal ysis Detection of partials peak detection and fundamental estimation AudioSculpt offers two analysis modules that inspect the structure of the amplitude spectrum The first one looks for peaks in the spectrum the second one concludes on the fundamental frequency based on the peaks in the spectrum Both analyses save the results in a text file if the sound has no sonogram open if a sonogram is shown the results will be presented as a breakpoint curve on the screen 41 AudioSculpt Users Manual Peak detection The Peak Detection command looks for the maxima in the spectra It is based on research by G Garcia X
67. of interpolation can be defined in the dialog box If you choose the filter response mode it is possible to vary the number of formants in time In the filter parameter mode it is impossible to change the number of formants 80 AudioSculpt Users Manual The format of the parameter file is as follows e First column time at which to apply the parameters e Second column the number of formants e Following columns the central frequency the gain and the bandwidth of each formant The dialog box allows you to choose the sound on which to apply the filtering Sonogram Filter With the Sonogram Filter you can filter the sound in a graphical manner The spectral surface of the sound must be calculated with the Sonogram Analysis command in the Analysis menu see also Part 2 Analysis and the introduction The surface tool and the pencil plus and minus allow you to draw a surface on the sonogram If the surface is drawn with a surface tool a gain can be given to it by selecting the surface and typing the value of the gain on the keyboard A positive number corresponds to an amplification a negative number corresponds to an attenuation If the surface is drawn with the one of the pencils the colored area will be cut out in case of the pen minus or nothing but the colored area will be kept in case of the pen plus The filter will be applied to the actual soundfile and the resulting sound is saved in a new soundfile This command is no
68. ogram shows you the exact time frequency and amplitude of the clicked point The short time spectrum pane displays the amplitude frequency curve at the selected time the displayed short time spectrum is a scan of the sonogram along the frequency axis When you activate this tool the inspector window is displayed Clarinet ALFF SEEN Time Frequency Amplitude i The Harmonic measure tool shows the first 18 harmonic partials of a fundamental frequency in the form of horizontal lines on the screen The position of the cursor defines the fundamental frequency This tool helps you scan for the exact position of the partials in the sonogram by comparing them with the harmonic spectrum of the partial measure tool The frequency scale on the right of the sonogram can help you in mea suring the frequencies The Sine play tool plays a pure sine tone with a frequency and amplitude corresponding the clicked point in the sonogram The amplitude depends on the grey scale of the clicked point You can adjust the volume with the color palette Caution this tool works only on Power Macintosh Text file output The results of the FFT and LPC analyses can also be saved in a text file These text files are mainly meant to be read by other applications like PatchWork Max or Excel since these files can be quite voluminous The analysis can be applied to a small region of the sound To do so select the region before selecting FFT
69. olor contrasts in the menu in the top of the palette The minimum and maximum display threshold corresponding to the minimum and maximum of the grey scale can be changed with the help of the two triangular markers in the bottom of the palette The thresholds are measured in dB The curve in the frame in the centre of the palette defines the dB to grey scale conversion to be applied between the two thresholds The default curve is linear but a new curve can be drawn with the mouse The stripe above the frame displays the resulting color conversion The changes in the color palette effect all the sonogram windows on the desktop FFT Analysis The FFT Analysis command launches a short term spectral analysis using the FFT algorithm If a part of the sound is selected only this part will be analyzed 1f not the whole sound will be ana lyzed The results are saved as a text file which can then be read by other applications like Max PatchWork Excel or a text editor The parameters for the analysis can be set the window size or fundamental frequency the window step the FFT size and the analysis window The results that will be exported and their output format can be defined with the Output format button The output formats proposed for the FFT Analysis are e time in seconds e number 1 plus the FFT size divided by 2 e freq instantaneous frequency in Hertz e midicents instantaneous frequency in midicents e amp the amplitude spect
70. onds that are written on the same line The second column says how many second order filters are used On the same line you then find the parameters for each second order filter the central frequency the gain and the bandwidth in that order and for each filter To move from one parameter value to the parameter value on the next line a linear in terpolation is done So between 0 seconds and 1 0 seconds a linear interpolation is done between a filter with formant at 440 Hz and a filter with 2 formants at 440 Hz and 880 Hz You can use parameter files for the filters mentioned above but also for time varying transposition for time varying time scale modification and for time varying general cross synthesis The parameters that you need to define are discussed in Part 3 28 AudioSculpt Users Manual What s next Part 2 of the manual discusses the analysis modules In Part 3 the processing functions of Audio Sculpt are detailed And in Part 4 the menus are displayed and a brief description of the commands is given 29 AudioSculpt Users Manual Part 2 Analysis 1 Analysis General principles of the short time spectral analysis Short time spectral analysis is a class of several sound analysis techniques that have some features in com mon The analysis consists of dividing the sound into small segments and looking at the frequency contents of each segment by creating a spectral representation of it This spec
71. only accessible if a parameter file is active The Surface Filter modifies the amplitude of the spectral components that fall into a surface defined in the amplitude frequency plane Every surface filter has a fixed gain The surfaces can overlap in that case their gains will be added The command Spectrogram Filter offers a very similar filter that can be operated graph ically The format of the parameter file is as follows e First line start time number of points end time and gain e Following lines time lower and upper border of the surface An interpolation is done between the start time and the time of the two first points as well as between the time of the last two points and the end time The dialog box allows you to choose the sound on which to apply the filtering Formant Filter This function is only accessible if a parameter file is active The Formant Filter is a filter composed of second order filters Each filter is defined by three parameters the central frequency in Hertz the gain in dB and the band width in Hertz Every second order filter corresponds to a formant It 1s possible to choose the kind of interpolation that is used in calculating the temporal evolution of the filter The first type interpolates the filter responses the filter responses are cal culated by adding the second order filter responses The second type interpolates the filter parameters cen tral frequency amplitude bandwidth The type
72. onograms Cut The Cut command clears the selected text and saves it on the clipboard This function is not available for sounds or sonograms Copy The Copy command saves the selected text on the clipboard without clearing it This function is not avail able for sounds or sonograms Paste The Paste Command inserts the text copied with one of the previous functions This function is not avail able for sounds or sonograms Clear The Clear command clears the selected text without saving it This function is not available for sounds or sonograms Select All The Select All command selects all of the text or sound in the active window 76 AudioSculpt Users Manual Preferences Under Preferences you can set certain options e Display splash screen at start up if the option is checked a window describing the application will appear on the screen when you launch AudioSculpt e Automatically save consoles when running scripts when the option is checked the console windows will be automatically saved in the default console folder see the Set Default Folders command e Hide console window unless using command if this option is checked AudioSculpt will not display the console window when a process is activated in the Processing or Analysis menu The console window is always shown when the command is launched from the command line Show Clipboard The Show Clipboard displays the text that was copied last with
73. onstructed from a combination of second order filters The spectrum of a sound is a series of complex numbers whose magnitudes correspond to the amplitude spectrum and whose phases correspond to a phase spectrum Filtering is the operation of multiplying the amplitude spectrum by an amplitude frequency curve and of adding a phase frequency curve to the phase spectrum With breakpoint filtering it is possible to change both the amplitude and the phase spectrum of the input signal By varying within bounds the instantaneous frequency of the components measured in each anal ysis channel it is possible to produce spectral distortion in the signal like for example limited transposi tion of a sound But this does not allow arbitrary transposition to be performed because this operation works only with instantaneous frequencies The multiplication of an amplitude spectrum by a curve serves to change the amplitudes that were provided for each analysis channel The amplitude for each analysis channel is multiplied by a coefficient if this co efficient is different from 1 the amount of energy in this channel is modified As with an equalizer it is possible to attenuate or reinforce certain frequency bands reinforce or attenuate certain harmonics or entire zones in the spectrum giving a formant like effect etc In AudioSculpt several filtering modes are possible 51 AudioSculpt Users Manual each characterized by a specific
74. ool you can also adjust the form of a polygon or rectangular surface by moving its angle points The label at tached to the surface displays the gain that will be applied to that region in the sonogram filter measured in dB To change the gain you select the region and type the number on the keyboard A positive number means the region will be amplified a negative number means the zone will be attenuated In case several surfaces overlap their gains will not be added only the gain of the surface on top will be applied Note that this differs from the Surface Filter see below The pen plus and pen minus With the pen plus also called pass pencil you can draw freely on the sonogram to mark a region The regions that are marked with the pen plus are submitted to a band pass filter which means that only these regions are left untouched That part of the sonogram that is left unmarked is eliminated You could for example retain a couple of partials of a complex sound The regions that are marked with the pen minus will be eliminated the rest of the sonogram is left un 54 AudioSculpt Users Manual changed You could for example cut out disturbing elements such as noisy clicks Band filtering In this mode a filter is defined as a set of alternating stop and pass bands The frequencies delimiting these regions can change with time the number of frequencies used to define the bands and the state of each band can not var
75. oundfile will be overwritten with the resulting soundfile This means the original soundfile will be lost unless you choose Undo in the File menu The Process and save button equally start the processing but the resulting soundfile will be saved in a new document The original soundfile will not be modified The name and the format of the new document is asked in a new dialog If the name of the new document already exists AudioSculpt asks if this new document should re place the already existing one The right side of window shows two buttons Resampling Parameters and Analysis parameters The resampling uses an interpolation function to generate the transposed sound The length of this function can be set with the Resampling Parameters button Further explanation can be found in Part 3 Most of the processing and analysis command have an Analysis parameters button In case of the transpo sition the button is only accessible if you mark the time correction option The button brings up the Analysis Parameters dialog In this dialog you can set the parameters for the FFT analysis that are used in the pro cessing A description of the parameters of the FFT analysis is given in Part 2 If the time correction option is marked AudioSculpt will expand or compress the resulting soundfile so it has the same length as the original sound 23 AudioSculpt Users Manual The analysis modules AudioSculpt offers several analysis modules bas
76. pectrum peaks amp the amplitude of the spectrum peaks amp_db the amplitude of the spectrum peaks in decibel freq the frequency of the spectrum peaks midicents the frequency of the spectrum peaks in midicents and depending on the requested analysis weights masked amplitude or perceptual pitch of the peaks The parameters for the spectral analysis can be set using the Analysis Parameter button The Sound menu Play Sound The Play Sound command plays the complete soundfile of the active window You can also press the option key and the space bar simultaneously Play Selection The Play Selection command only plays the sound that has previously been selected the selected sound is highlighted in black Pressing the space bar has the same function Get Sound Info The Get Sound Info command displays a window with the information about the soundfile At the top of the window you will find the name of the file and its format AudioSculpt recognizes and uses AIFF AudioSculpt and Sound Designer II formats The data format is 16 bits integer or float ing point in the case of an AudioSculpt soundfile The information window also shows the sam pling rate the length and the number of samples of the soundfile Sound Output Be sure that your Macintosh is equipped to reproduce sound with the proper quality The soundfile can be played using the Apple Sound Manager version 3 If your computer is equipped with the Digid
77. phique sonogramme ainsi que la possibil it de filtrer le son en dessinant directement sur le sonogramme D autres types de traitements contr l s graphiquement seront prochainement ajout s cet ensemble tel que l dition d une courbe de compression dila tation temporelle ou de transposition AudioSculpt est utile la fois pour les chercheurs qui travaillent sur la structure des sons et pour les musiciens qui veulent modifier des sons Les applications sont nombreuses car il est possible avec AudioSculpt de modi fier ind pendamment l amplitude la fr quence le spectre et la dur e des sons quelque soit leur origine AudioSculpt fournit par ailleurs des r sultats d analyse hautement sophistiqu es sous la forme de fichiers utilis ables par d autres programmes Enfin l utilisateur peut directement intervenir sur les zones d affichage graphique ce qui rend ais la manipulation d op rations complexes 1 SVP Super vocodeur de phase 7 AudioSculpt Users Manual Les operations de AudioSculpt AudioSculpt offre deux grandes familles d op rations analyse et traitement Analyse AudioSculpt reprend toutes les fonctionnalit s pr c demment disponibles dans SVP 1 3a en ajoutant une fonc tion nouvelle l analyse spectrographique le mode sonogramme L analyse spectrographique permet le traitement des sons grace son interface graphique L utilisateur intervient directeme
78. quency composition of the sound while at the same time having control over the temporal evolution For example it is possible to use a large window size for a low frequency sound a low frequency has a long period so the window should be long but set a small window step in order to resolve transient components 32 AudioSculpt Users Manual Choice of window There are different kinds of windows The choice of a particular type is determined by its effect on the spec trum of the signal The obtained short time spectrum is not the exact spectrum of the signal but the convo lution of the spectrum of the signal with the window s spectrum Spectral analysis There are different short time spectral analyses that allow the frequency contents of a sound to be evaluated including the Fourier transform linear predictive analysis discrete cepstrum etc In this version of AudioSculpt you ll find the Fast Fourier transform and the linear prediction analysis Future versions of AudioSculpt will include the discrete cepstrum analysis The parameters used in the analysis modules are among others the type of spectral analysis FFT LPC the size of the FFT the window size and the window Step In the next paragraphs we will give a brief description of the FFT and LPC analysis This description is merely an introduction to explain the basic features of these analysis techniques and their parameters For a more detailed description the
79. r 81 Generalized Cross Synthesis 81 Use Command Line 82 Run Script 82 Stop Processing 82 Sonogram Analysis 83 Realtime Analysis 84 Analysis Settings 84 Realtime Sonogram Scaling 85 Edit Color Palette 85 FFT Analysis 85 LPC Analysis 86 Peak Detection 86 Fundamental Estimation 87 Masking Effects 88 Play Sound 89 Play Selection 89 Get Sound Info 89 Sound Output 89 Show Console 89 Annex oo cccccececcecececcecsceccecececeececeeeess Commandlines 91 Annex B 065 List of options of the command line 94 Index 100 Resume Ce manuel pr sente le logiciel AudioSculpt pour le Macintosh AudioSculpt est un programme qui a t con u et d velopp a l Ircam Il est bas sur SVP programme cr par Gilles Poirot et Philippe Depalle du groupe Anal yse synthese Sous la direction de Gerhard Eckel et avec la coordination de Marie H l ne Serra SVP fut ensuite r vis et optimis par Chris Rogers avec des extensions programm es par Jean Carrive Con u partir de la version 1 3 de SVP pour le Macintosh AudioSculpt est une interface graphique de repr sen tation des sons ainsi que de contr le d un ensemble de transformations compression dilatation temporelle trans position filtrage synth se crois e etc La version 1 0 d AudioSculpt int gre toutes les commandes d j disponibles dans SVP 1 3A du Macintosh et ajoute de nouvelles possibilit s essentiellement l analyse spectrogra
80. r shown on the sonogram Processing The processing commands offer the possibility of defining the modifications to be applied to the sound Most of these modifications are based on the principle of the phase vocoder in which the sound is submitted to an analysis describing the spectral evolution of the sound in time Then the spectral representation is al tered according to the requested processing and the final sound is obtained by re synthesis Most of the pro cessing functions perform an analysis synthesis using the phase vocoder Only the transposition without time correction and the re sampling function do not perform an analysis synthesis Note for every processing task that performs an analysis synthesis it is important to choose the appropriate analysis parameters window size window step FFT size window type in order to obtain the best results 12 AudioSculpt Users Manual Because the values of the parameters of the analysis can be set independently it is possible to tune the pro cessing precisely To obtain the best results great care should be taken when specifying these parameter values Note the functions in the processing menu are independent of the functions in the analysis menu In other words it is not necessary to perform an analysis in the Analysis menu in order to be able to perform pro cessing on the processing menu The processing operation will perform the analysis by itself Caution the sole
81. recursive filter arn plitude A madim un arn plit y de al 306 i hand th resonart frequency frequency There are two types of formant filtering one is called Filter response the other is called Filter parame ters Filter response the number of formants can vary with time This mode is called fifof filter ing in SVP Filter parameters the number of formants is fixed This mode is called fof filtering in SVP When you launch the command Formant filtering the dialog box allows you to choose between the two types of formant filtering interpolation You write into a text file one or several lines in which you specify the time the number of formants and for each formant its frequency amplitude and bandwidth in Hz as described hereafter e First column time at which to apply the parameters e Second column the number of formants e Following columns the central frequency the gain and the bandwidth of each formant The following file defines a formant filtering that applies a spectral envelope with an increasing number of formants i e it can only be used in response mode 1 fof stands for Fonction d onde formantique This is an adaptation from a work by X Rodet 59 AudioSculpt Users Manual 0 0 1 440 0 50 1 0 2 440 0 50 880 2 100 2 0 3 440 0 50 680 2 100 1760 2 200 The first column describes the time the second column the number of formants and then for each form
82. rithm uses a window to rebuild a segment of the signal Each of the operations of filtering cross synthesis expansion compression and transposition can be param etrized and controlled as a function of time This means that blocks of signal can be read and processed sep arately over time for example the frequency response of the filter the expansion compression factor and 49 AudioSculpt Users Manual the coefficients that describe the mixing between two sounds are parameters that the user can specify and vary temporally The part on processing 1s completed with the Sampling Rate Conversion function 50 AudioSculpt Users Manual 2 Filtering General principles The filtering deals with the spectral information of the sound The spectral representation that comes from the analysis module is multiplied by the frequency response of a filter There are several types of filters that is there are several ways to define the frequency response curve for a filter In AudioSculpt you can specify the filter in a graphical way by drawing on the sonogram Besides the sonogram filter AudioSculpt has four kinds of filtering which operate based on data from a text file band filtering alternating stop and pass band filters breakpoint filtering filter response curve made up of straight line segments surface filtering filter that is specified as a time frequency surface and formant filtering where the filter envelope is c
83. rum e amp_db amplitude spectrum in decibels e real the real part of the spectrum e imag the imaginary part of the spectrum e phase the phase spectrum In the text file you ll find the time or number in the first line and underneath the requested data in columns separated by a tab LPC Analysis The LPC Analysis command launches spectral analysis using linear prediction It is used to calcu late the spectral envelope If a part of the sound 1s selected only this part will be analyzed if not the whole sound will be analyzed The results are saved as a text file which can then be read by other applications like Max PatchWork Excel or a text editor You can define the order of the LPC Analysis as well as the analysis parameters with the Analysis Parameter button The results that will be exported and their output format can be defined with the Output format button The output formats proposed for the LPC Analysis are e time in seconds e number 1 plus the FFT size divided by 2 e freq frequency scale e midicents frequency scale in midicents e amp the amplitude of the spectrum envelope e amp_db amplitude of the spectrum envelope in decibels In the text file you ll find the time or number in the first line and underneath the requested data in columns separated by a tab Peak Detection The Peak Detection command looks for the maxima in the spectra in each consecutive short term amplitude spectrum Opening
84. rum see Part 2 Analysis The short time spectrum is a spectral representa tion of the segment of signal selected by the window and gives over a certain time interval the decompo sition of the sound into a sum of sinusoids each of which is characterized by its amplitude and phase The fundamental characteristic of the Fourier transform is that it gives an intermediate representation of a signal in which frequential and temporal information are separated The intermediate representation in the form of a series of Fourier transforms makes apparent the distribution of the energy of the signal by fre quency as well as the way in which this distribution changes with time Before the signal is reconstructed the user can work directly on this intermediate representation so as to modify the original signal Since this representation contains two types of information temporal information and frequential informa tion two different kinds of modifications must be distinguished temporal modifications and frequential modifications Here is an illustration of the analysis synthesis procedure for the phase vocoder aioe VN aa IN ADR paf a af de RAM AT IEUA RAR i tme shat irne Fourier ransom amp itude fquenc Rp Anal a a 1 1 E pd q ne temporal modification consists of dilating or contracting at the synthesis stage the time interval separat tme amp itude 48 AudioSculpt Users Manual ing t
85. se menus If the name of a com mand is shown in grey the command is deactivated Opening a soundfile To open a soundfile choose Open Ouvrir in French operating system in the File Fichier in French menu and you will see the familiar file dialog 1 All standard Macintosh dialog boxes appear in French in this manual as all examples were made on a French Macintosh system 15 AudioSculpt Users Manual AudioSculpt 1 0 Donna B Clarinet AIFF Cymbal AIFF english msg Open the soundfile Clarinet AIFF by selecting it and then clicking the Open button Ouvrir in French The soundfile will appear in a new window on the screen It is represented as a time amplitude envelope Curve e the horizontal axis shows the time in seconds e the vertical axis shows the amplitude envelope of the sound Note that AudioSculpt displays the positive amplitude envelope Clarinet AlFF Click with the mouse in the time axis bar hold down the mouse key and move the mouse to the right The time axis re scales while you re moving and the sound pane stretches simultaneously Keep stretching the time axis until the envelope curve disappears and changes into the actual sampled sound curve Depending on the scale you will see the amplitude envelope or the sampled sound curve Now double click on the time axis to bring it back to its original scale There is also another way of changing the time scale Hold down the sh
86. several simultaneous spectral components the frequencies that are perceived are not necessarily the same as the frequencies that were detected These small frequency shifts of the spectral components are calculated in order to obtain the Perceptual Pitches The third result Weights accounts not only for the masking effects but also for the spectral importance of the components A weight bigger than zero is given to significant components a zero weight is given to the insignificant components Choosing the Masking Effects command in the Analysis menu will bring up a dialog 45 AudioSculpt Users Manual Masking Effects Weights Masked amplitudes Perceptual pitches All Threshold in dB Humber of peaks ES Output Format Analysis Parameters Process If a sonogram analysis has been performed the spectral component with the highest weight is traced on the sonogram If no sonogram is displayed the results will be registered in a text file The dialog lets you determine what analysis will be performed Weights Masked amplitudes Perceptual pitches or All of the preceding The threshold for the peak detection can be set and you can specify how many peaks you want to be saved Only the peaks with the highest amplitude will be given as a result By default the highest peak is given The formats of the results can be set using the Output Format button time in seconds number the num ber of peaks desired The next fo
87. sion with factor 2 sound file will be twice as long In case a transposition with time correction is wanted this option should be included in the command line No value for the compression expansion factor should be given since AudioSculpt will calculate it automatically E lt end gt or e lt end gt Specifies the end time of the analysis within the sound file with lt end gt given as sample number integer or as time in seconds float The end time can be big ger then the length of the sound file to add silence at the end Example B88200 B5 0 F lt filter type gt lt file name gt or f lt filter type gt lt file name gt Determines the filter type The following types are available e breakpt breakpoint filter e bande band filter e surface surface filter e fifof formant filter interpolation of the filter response e fof formant filter interpolation of the filter parameters lt file name gt contains the temporal evolution of the parameters B 95 AudioSculpt Users Manual Example Fbande ParaFile G lt cross synthesis type gt lt file name gt or lt cross synthesis type gt lt Xvalue gt lt xvalue gt lt Yvalue gt lt yvalue gt lt qvalue gt The two types of cross synthesis that can be used are e cross general cross synthesis e mul source filter cross synthesis The parameters can be given in a file lt file name gt contains the parameters for a variable cross synthesis or directly
88. sis source filter SVD SV t s iiltered sound s tiltering sound GmMmuUL A A Jhamming N4096 M4000 Whanning I2000 alpc 30 whamming m1024 1128 name outputfile General cross synthesis SVP v m S filtered sound s iraliering sound Geross xXlw0 xL lt 0 x 0 yl 0 q0 0 A a Jhamming N4096 M4000 whanning L2000 a whamming m1024 1128 name outputfile Sample rate conversion Svp v t S name soundfile H22050 resWhanning resM21 name out putfile 92 AudioSculpt Users Manual Time compression expansion SvE v G So name Sound ile DZ A 2 Jhamming N4096 M4000 Whan ning 1250 name outputfile Transposition without time correction Svp v t S name soundfile trans 1200 resWhanning resM21 name outputfile Transposition with time correction Svp y t 5 name soundriile D trans 1200 A 2 Jh mmine N4096 M4000 Whanning I2000 resWhanning resM21 name outputfile 93 AudioSculpt Users Manual Annex B List of options of the command line A value given between brackets shows the default value of a parameter Additional examples of com mand lines can be found in annex A A lt analysis type gt or a lt analysis type gt Specifies the analysis type THE Following types can be used e fft Fast Fourier Transform Example Afft e Ipc lt order gt linear prediction analysis with lt order gt the number of poles 30
89. sis will not seek the fundamental frequency beneath the threshold specified by the minimum fundamental the default threshold is 50 Hz Maximum fundamental the analysis will not seek the fundamental frequency above the threshold specified by the maximum fundamental the default threshold is 2500 Hz Maximum harmonic the analysis will not consider harmonics above the threshold speci fied by the maximum harmonic the default threshold is 7500 Hz Threshold in dB specifies a noise level if the difference in amplitude between a peak and the highest peak is bigger then the threshold the peak will be neglected The ratio Maximum harmonic Minimum fundamental is limited for reasons of memory storage and calculation time If the sonogram is displayed the result will be shown on the screen as a breakpoint function If the sonogram is not displayed the result will be saved in a text file In that case the format of the result can be set using the Output Format button time in seconds number the number of peaks will be 1 amp the amplitude of the fundamental amp_db the amplitude of the fundamental in deci bel freq the fundamental frequency midicents the fundamental frequency in midicents The parameters for the spectral analysis can be set using the Analysis Parameter button Masking Effects The Masking Effects command reduces the amplitude spectrum to a small number of peaks which contribute to the tonal perceptio
90. ssing the enter button when the text window is active You should give different names to result files of every treatment Stop Processing The Stop Processing command stops the active process The Analysis menu 82 AudioSculpt Users Manual Sonogram Analysis The Sonogram Analysis command performs a spectral analysis of the soundfile A dialog lets you specify the analysis parameters Window size Fundamental frequency Window step FFT Size Analysis window Analysis type the window sizes determines the number of sound samples in each analysis allows to compute the window size automatically If you specify the win dow size the fundamental frequency that is calculated will be displayed The window size is set to five times the length of the period that corre sponds to the fundamental frequency the window step determines the time interval between two succesive analy ses the FFT size determines the number of points of the analysis in the analysis window menu you can specify the window type used in the analysis two types of spectral analysis techniques are proposed fast Fourier trans form and linear prediction The Factory Settings button sets the parameters to the default value On launching the analysis two new panes are added to the window beneath the existing sound pane the left pane is the sonogram pane the right pane is the short term spectrum pane You can stop the analysis with th
91. t available with the command line Cross Synthesis Source Filter In the Cross Synthesis Source Filter the sound marked next to Filtered Sound will be filtered with the spectral envelope of the sound selected under Filtering Sound The spectral envelope of the Filtering Sound is calculated using the linear prediction analysis The parameters of the FFT analysis and of the LPC analysis can be set with the Analysis Parameters option Generalized Cross Synthesis The Generalized Cross Synthesis is a synthesis technique that for example permits a gradual spectral tran sition between two sounds The short term spectra of both sounds are decomposed in an amplitude spectrum and a phase spectrum Both spectrums can be manipulated separately A gradual spectral transition is pos sible if you use a parameter file describing the temporal evolution of the parameters The initial soundfiles should be both stereo or mono the result file has the same configuration as the initial files The format of the parameter file is as follows e First column time at which to apply the parameters e 2nd 3th 4th 5th and 6th column the values of X x Y y and q 81 AudioSculpt Users Manual Use Command Line With the Use Command Line function you have access to a box that lets you enter a UNIX style command line With this command line you can launch analysis or processing and change its parameters without hav ing to manipulate the graphi
92. te description of the signal over the chosen portion a y REA PAT amp itude F spectrum of the window ed signal wf is ee ee ns f AA frequency ea tme Window size The number of samples in the windowed segment is named the window size This windowing operation is performed by multiplying the signal by a fixed function of time at a certain time index within the sound The role of this operation is to pick out a small segment of the signal while at the same time it guarantees sufficient spectral and temporal resolution The short time spectrum gives an accurate description of the characteristics of the signal as long as the sig 31 AudioSculpt Users Manual nal is stationary over the length of the segment delimited by the window In general this is the case for short durations during which the characteristics of the amplitude time curve for the sound amplitude envelope length of the period shape of the wave form etc do not change appreciably If the segment has a stable frequency content the frequency information of the sound segment is contained in the short time spectrum and the temporal information 1s reflected by the changes in the frequency com position over successive analyses If the short time spectrum is calculated over a portion of signal that is non stationary the changes in time will not be accurately registered in the successive analyses If the window size is large the short time spec
93. te gt or r lt sample rate gt Specifies sampling rate Used when performing resynthesis on analysis data file resM lt value gt Specifies resampling window size with lt value gt the resampling convolution window size in samples 31 resW lt value gt Specifies resampling window type with lt name gt the resampling window type hamming S lt file name gt or s lt file name gt Specifies input file name read from the default sound folder Example Sflute T Display the current position of the synthesis window during the calculation t Display the current position of the analysis window during the calculation trans lt cents gt Transposes by the given number of cents Time correction is performed if used in conjunction with the flags A Z D Example trans 1200 will transpose one octave up V Prints out details about the AudioSculpt patch structure and parameters W lt window type gt or w lt window type gt Specifies analysis window type There are three possible windows e hamming default e hanning e blackman Example Whanning Z Performs a resynthesis inverse fft and overlap add Index A AIFF 14 89 amp 25 amp db 25 Amplitude spectrum 51 64 Analysis menu 82 Analysis parameters 23 Analysis settings 84 Analysis type 83 Analysis window 83 85 Apple 2 Audiomedia II 13 B Band filtering 55 79 Battier M 2 Blackman 35 38 85 Breakpoint filtering 56 79 C Carr
94. the Blackman window is bigger As a general rule in order to obtain an acceptable analysis for pseudo periodic sounds it is recommended that a window size of perhaps 4 or 5 times the period of the sound be used AudioSculpt takes a window length of 5 times the period length If the sound is not periodic the window size is determined by the distance between the closest frequency components that one wishes to separate For example if the closest spectral components have a difference between them of 100 Hz and the sound is sampled at 44100 samples per second then the window size should be at about 2205 samples long Indeed 100 Hz corresponds to a period of 10 msec which on its turn corresponds to 441 samples Multiplying by five gives the final result Your goal in using AudioSculpt will usually be the modification of the sound s spectrum Most of the pro cessing techniques offered by the program involve the conversion of the sound into spectrum a procedure that is transparent for the user In this process however it is desirable to have the least possible spectral distortion When the window is enlarged the window effects are diminished and the spectral components are better resolved The window size thus has a direct influence on the quality of the spectral data obtained On the other hand the spectrum must not be so large that the resolution of the time evolution of the spectrum is compromised The bigger the window size the longer is the
95. the Default folders In the File menu you will find an item called Set Default Folders In this item choose Sound Folder The dialog that comes up lets you walk through the folder structure of your computer The Default Sound Folder is the folder where AudioSculpt looks for the soundfile when you use the command line It should be set to the folder where your soundfiles reside In this example make sure that you are in the AudioSculpt 1 0 folder and then click the Set Default Folder button Do the same for the parameter folder 13 AudioSculpt Users Manual Soundfiles AudioSculpt processes 16 bit soundfiles and the 32 bit AudioSculpt format If you inadvertently use an 8 bit soundfile AudioSculpt will not proceed AudioSculpt is capable of processing stereo files Be aware that stereo file processing is slower and takes up more memory than mono soundfile processing Three sound formats are recognized by AudioSculpt AIFF Audio Interchangeable File Format the Sound Designer II format and its own AudioSculpt format Most digital audio applications let you save the sound file in one of these formats The AudioSculpt format saves the samples as a floating point number using 32 bits per sample Parameter files Parameter files are the files that control the parameters of a time varying processing such as transposition etc All parameter files should be edited and saved as text files 1 e plain ASCII files The files can be ed
96. their amplitudes and vice versa We can obtain a hybrid sound with the combination X 1 x 0 Y 0 y 1 q 0 Here the instantaneous frequency spectrum of the secondary input is applied to the primary input sound With the combination X 0 x 1 Y 1 y 0 q 0 the amplitude spectrum of the secondary input is applied to the primary input sound The multiplication of the two amplitude spectra is a rather delicate matter When this operation is used alone X x are zero the resultant spectrum may be zero if the two spectra are complementary that is when their spectra do not overlap at all If the coefficients X x Y y and q are constant they can be specified in the dialog when choosing the General Cross Synthesis command if not they can be provided in table form from a text file The following example creates a non time varying combination the phase spectrum of Cymbal AIFF sec ondary input is applied to the amplitude spectrum of Clarinet AIFF Generalized Constant Cross Synthesis amplitude scale factors phase scale factors Sound 1 Clarinet AlFF Sound 2 Cymbal AlFr oF amplitude cofactor oo Process in place Analysis Parameters Process and Save In the example below the parameters are specified in a text file g 0 0 1 0 0 0 1 0 0 0 0 0 En 10 0 0 1 0 0 0 1 0 0 0 2 0 1 0 0 0 1 0 0 0 0 0 3 0 0 5 0 5 0 5 0 5 0 5 X X Y y q The first column specifies the time of the parameters and on each lin
97. this module gives access to a dialog in which the parameters can be given Threshold in dB the difference in amplitude between a peak at the 1 th index and the amplitudes at the neighboring points at indexes 1 3 1 2 1 2 1 3 must exceed the threshold Peaks that do not satisfy the constraint are neglected The default value is 25 dB Number of Peaks You can specify how many peaks you want to be noticed Only the m peaks with the highest amplitude will given as result By default the 5 highest peaks are given With the Analysis Parameters button you can set the values of the analysis parameters If the sonogram is displayed the results will be shown on the screen as a breakpoint function if the sonogram is not dis played the results will be registered in a text file In that case the formats of the results that can be chosen using the Output Format button are e time in seconds e number the number of peaks desired e amp the amplitude of the peak e amp_db the amplitude of the peak in decibel e freq frequency of the peak e midicents frequency of the peak in midicents The peaks are ordered according to decreasing amplitude Fundamental Estimation The Fundamental Estimation command looks for the fundamental frequency based on the spectral analysis In every analysis window an estimation of the fundamental frequency is made In the di alog window the analysis parameters can be set Minimum fundamental the analy
98. ting soundfile might be corrupted To compress the attack and still expand the rest of the sound this time by a factor that increases with time type time varying modification Whe oo aI Woo oom on AWM BM ee oOo D J e J ooo on AudioSculpt performs a linear interpolation in time between the coefficients given on each line For exam ple between 1 0 and 1 5 seconds the stretching factor changes linearly from 1 5 to 2 7 an increase of 1 2 Between 1 5 and 2 0 seconds the coefficient changes linearly from 2 7 to 4 1 an increase of 1 4 From 2 0 to 2 5 the coefficient increases by 1 6 etc It is apparent that the stretching factor increases more and more rapidly with time and thus the sound is slowed down by more and more Here is a figure of the stretching coefficient as a function of time stretching coeicent Jr 4 1 ef 1 5 1 0 0 5 0 0 020A 1 0 1 5 2 0 2 3 0 tine 69 AudioSculpt Users Manual Improving the quality of a time stretched sound If the spectral analysis is lacking in resolution artifacts can appear simply because the time stretching op eration done by AudioSculpt uses the spectral representation of the sound If the short term Fourier trans form is not very selective the phase measurements will not be reliable and distortions will occur when the sound is resynthesized If the quality of a time stretched sound is not sufficient a bad time stretched sound sometimes has a m
99. tral representation can be obtained us ing one of several spectral analysis techniques Each segment is obtained by multiplying the signal by a win dow in the time domain To obtain the next segment the windowing operation is repeated after having shifted the window a certain distance along the time axis Each of the segments obtained in this way is pro cessed by an algorithm which calculates its short term spectral representation One of the short time spectral representations is the Fourier transform which gives the amplitude and phase 30 AudioSculpt Users Manual of the signal as a function of frequency that is the energy distribution of the signal frequency like the sono gram The short term spectral representation is a useful representation for two reasons on the one hand it is an intuitive representation for the user because it functions much in the same way as the human auditory sys tem which detects the frequencies that make up a sound at any given instant On the other hand because of the decomposition of sound along two variables time and frequency it separates the temporal and frequen tial information of the sound This allows the temporal information to be manipulated independently of the frequential information Because in many cases the goal of such an intermediate representation is to be able to modify the signal s properties it is important to know to what degree the short time spectrum gives an accura
100. undfile even if you have renamed it or moved it into another folder Through several examples the next paragraphs will explain some features of the AudioSculpt interface Process in place or Process and save Transposition Open a soundfile and choose the Transposition command in the Processing menu This command lets you shift the sound to a higher or a lower frequency A dialog like the one in the figure below appears on the screen The interval with which the sound has to be transposed can be written in the text zone besides num ber of cents 100 cents corresponds to the interval of a half tone Type for example 700 cents which corre sponds to the interval of seven half steps 1 e a perfect fifth Now let s take a look at the options in the dialog box since most of the dialog boxes of the processing commands have them in common 22 AudioSculpt Users Manual Constant Transposition number of cents EANIRI time correction Cancel Resampling Parameters Process in place sis Parameters Process and Save The name of the selected processing technique is marked at the top of the window In this case it should say Constant Transposition The right side of window shows three buttons one above the other Cancel Pro cess in place and Process and save The Cancel button cancels the command The dialog disappears and no processing takes place The Process in place button starts the processing The original s
101. ur describe the peaks in the amplitude spectrum amp the amplitude of the spectral peaks amp_db the amplitude of the peaks in decibel freq the frequency of the peaks midicents the frequency of the peaks in midicents According to the requested analyses the weights the masked amplitudes the perceptual pitches or all of these can be saved The parameters for the spectral anal ysis can be set using the Analysis Parameters button 46 AudioSculpt Users Manual Part 3 Processing Processing overview Most of the sound processing modules use an intermediate short time spectral representation of the sound This means that the first step of the processing consists of an analysis that generates this intermediate spec tral representation The second step consists of modifying this representation and finally the resulting soundfile is reconstructed using the inverse transformation of the analysis step re synthesis In order to perform the spectral analysis the sound signal is windowed This windowing operation is performed by multiplying the signal by a fixed function of time at a certain index within the sound The role of this oper 47 AudioSculpt Users Manual ation is to pick out a small segment of the signal The series of samples in the windowed segment is trans formed into a spectrum by applying a Fourier transform As in the figure below for each window of signal there is a corresponding spect
102. w size is not necessarily equal to a power of two If the window is enlarged to a power of two the resolution in time of the short time spectrum is diminished To resolve this problem a window size is initially chosen 5 times the length of the period for example and then is padded with enough samples with value zero so that the next highest power of two is attained Adding zeros to a windowed signal has no effect on frequency resolution separation of the sound s frequen cy components The spectral resolution however is improved because the spectrum has been more finely sampled Increasing the FFT size makes the spectrum easier to inspect The size of the FFT determines the spectral resolution which is the distance between two frequency bins in the spectrum Spectral resolution is given by the ratio of sampling rate to FFT size Linear Prediction LPC The FFT and LPC are not equivalent because they compute a spectral representation that does not contain the same information The FFT contains all the spectral information while in the LPC spectrum only the spectral envelope is retained The implementation of LPC is based on a program by Ph Depalle The spec tral envelope concerns the contour of the amplitude spectrum and not so much the spectral components The knowledge of the spectral envelope allows the determination of the maxima of the envelope which are called the formants The position of these formants plays for example
103. which the parameter and console files are looked for or saved can be set with the Set Default Folders command Now choose Use Command Line in the Processing menu A window will be displayed Type the follow ing text in the text zone of the command line window svp v t SClarinet AIFF B0 0 E0 5 Afft U Oa Whanning N4096 M4000 11024 Clarinet AIFF out Click Process The console window is brought onto the screen showing the sound format and the parameters of the analysis When the analysis is completed the text file should look like this 26 AudioSculpt Users Manual Clarinet AlFF out 18 302334 0 000000 9 522903 0 4q7659 O edezel 0 000226 0 099571 0 021576 0 000090 O 07 S246 0 537098 0 710259 0 213065 1 609715 2 073066 4 01ele4 3 956949 5 169414 1 873504 z 30059q 0 250644 0 665036 0 125053 0 692931 O 014475 O 276 e1 0 202166 8 O 067 o41 GO mi1idse You ll find the values of the complex spectrum in columns see Part 2 to know more about complex spec trum You can copy and save a command line in a text file that you can create from AudioSculpt s command File gt New and thus create a library of command lines Processing and parameter files formant filter Open a soundfile and then open a new text document in the File menu In the Processing menu you ll find four filter types e breakpoint filter e band filter e surface filter e formant filter Caution these
104. y during processing The filter parameters are defined from within a text file You should create or open a text file and enter the Band filter data also called a parameter table The format of the parameter file is as follows e First column time at which to apply the parameters e Second column the number of border frequencies e Third column flag specifying if the first band is a stop or pass band e Following columns the values of the border frequencies Take for example the following parameter table number of points in which the filter changes time stop pass frequencies The lines are separated by carriage returns Return key and the numbers on each line are separated by blank spaces or tab strokes The first column gives the time describing the filter s time evolution The second column gives the number of points in which the filter changes and the third column gives the state of the first band where 0 Stop band l Pass band Successive bands will alternate between O and In this example there is one stop band The stop band lies between 50 Hz and 200 Hz at 0 sec and because AudioSculpt always interpolate linearly the data from one line to the next one this band moves linearly to the frequency region between 500 Hz and 1000 Hz which 55 AudioSculpt Users Manual is attained at 1 5 sec and then returns to its initial state at 3 sec 0 200 ajdi Breakpoint filtering Breakpo

AudioSculpt

Contents

Download Pdf Manuals

Related Search

Related Contents