Voice Activated Wireless Control of Devices in a Room Environment
Contents
1. 0 5 seconds Analysis time 5 seconds FM Transceiver e Max bit rate chip can switch at a maximum rate of 20Hz resulting in a gap of about one second between code generation and reception e Range Tested on a 50 feet link The transmitter design is reckoned to have a range of several hundred meters but the signal strength and signal noise ratio may deteriorate over this range 1 6 DETAILED SPECIFICATIONS 11 e Error rate Depends on many factors distance proximity to other ra dio stations etc the latter has been removed by going below the FM bandwidth range Over 20 meters the error rate was virtually zero 12 CHAPTER 1 USER MANUAL Part II Theory of Operation amp System Design Chapter 2 Theory of operation This chapter gives the detailed description of the operation of the system The chapter is divided as follows Section 2 1 gives the theoretical background and the block diagram of the system Section 2 2 gives the details of the hardware utilized Section 2 3 gives the details of the firmware that went into the running the hardware with required FSMs Section 2 4 gives the details of the tests done during integration of the system 2 1 System description 2 1 1 Word Recognition The word recognition algorithm is based on Mel Cepstral coefficients of the spo ken isolated word The first step is sampling the mic input and detecting the start of the word The sampling is carried out at 8 ksp2. 10 SRAM and 64K x 16bit 150ns Atmel AT29LV1024 15 parallel FLASH The upper 32K words of the SRAM are mapped into external data space The lower 32K of SRAM can be used in lower 32K program space if the DSP is in mi croprocessor mode This is useful during program development and debugging as the program can be loaded and run from SRAM instead of burning it in internal FLASH However in microcontroller mode the lower 32K of SRAM is left unutilized The external flash is mapped into the upper 32K of external program space Thus it can be used for storing program if the code size exceeds 32K It can also be used to store permanantly any data in the form of tables which are not accessed very frequently The problems in accessing flash are it is quite slower 150ns compared to CPU clock speed 25ns and it is programmable sectorwise with sector size of 128 words The required read write algorithms are implemented as the part of firmware GPIO General Purpose Input Output The board has 1 6 LEDs for indicating various states and jumper configurations 2 10 general output LEDs controlled by 10 GPIO pins IOPF0 5 IOPA6 7 IOPC6 7 of the DSP 3 8 bit general purpose output port controlled by port IOPB0 7 of the DSP 4 8 bit general purpose input port at port D IOPDO and port E IOPE1 7 of the DSP The LEDs are driven using ULN2803 Darlington pairs array The input output ports are 5V compatible The TI Level Shifter chip
3. System Block Diagram 2 1 4 System block diagram The system block diagram is shown in figure 2 1 2 2 Hardware 2 2 1 DSP Board The DSP board is designed around Texas Instruments TI TMS320LF2407A DSP It is the member of the C2000 family of DSP controllers The prominent features of LF2407A include 1 2 3 40 MIPS performance at clock speed of 40MHz Low power 3 3V operation Havard architecture with two separate internal busses for data and pro gram for high speed operation 32K on chip FLASH memory for program 2 5K on chip fash access data RAM External memory interface EMIF with separate program space data space and IO space each with 64K addressing capability 32 bit ALU TI C2xx compatible Instruction set featuring several specialized instruc tions and addressing modes optimized for implementing standard DSP algorithms like filtering FFT These specifications make LF2407A the most advanced processor among C24x family 2 2 HARDWARE 19 Power Supply The DSP requires 3 3V for its operation and 5V for internal flash programming The two voltages are generated from DC 9V input using TI TPS76833 and TPS76850 low drop out voltage regulator chips respectively The TI supervisor chip TPS3305 33 monitors both the supply lines and triggers the reset if the voltage drops below 98 of its normal value On board Memory The onboard memory consists of 64K x 16 bit 10ns Cypress CY7C1021CV33
4. Recognition This mode is valid only when system has undergone full training In this mode the system continuously samples the audio signal and detects the start of the word This is indicated by blinking of state LED On capturing the word it extracts the parameters and compares them with the previously stored template On identifying the match the word number flashed on the word LEDs and the transmission of corresponding device code begins The word LEDs remain unchanged till the transmission continues On detecting the mismatch the red state LED is flashed The recognition mode continues till the system is forced into training mode using DIP switches Note This product must always be connected to the power supply It is activated only when the command is spoken and is in the idle or waiting state at other times 1 5 Safety precautions 1 Do not speak into mike when you do not wish to operate the device be cause it might respond to words similar to the ones specified and lead to unwanted switching 2 Do not touch the inside components on the printed circuit board 3 If attached to the wheelchair make sure that the assembly is firm 4 This product has been designed for indoor use only 1 6 Detailed Specifications Word Recognizer e Supply Voltage 9V DC e Supply current 150mA when idle and 300mA when in operation e Optimum Mike distance 2 inches e Mike electret e Voice sampling rate 8K Hz e Maximum word length
5. during the trans mission after the command is recognized correctly It indicates correct reception of the codes If it is not lighting up or is flickering then there is some problem in the transmission and the reception of the codes The A first solution would be to align the antenna of the system more strategi cally Also the distance between transmitter and receiver can be decreased to improve the reception If still the transmission is unreliable the trans mitter has to be tuned correctly by adjusting the trimcap Refer to the detailed description of the transmitter In case the receiver is still unable to receive the correct code check the following points Testing points VCC and GRND of all circuits The two tx modules should have a common ground and similarly for the two receiver modules ENCODER ALE pin of the microcontroller Should give a waveform resem bling a noisy square wave Pin 16 of the DTMF chip Should show different tone pairs when in use FM TRANSMITTER Can only check the points in the pre amplifier stage since beyond that high frequency oscilloscope will be required If that is available the waveforms across the local oscillator can be measured They should be at a frequency of about 70 MHz Else the transistors might have become faulty The rest are all passive devices and don t really get spoilt with usage 31 32 CHAPTER 3 SERVICE MANUAL FM RECEIVER Check the local oscillators you could do this by checkin
6. in the range of a few milli volts whereas the O P tone of the DTMF chip is of the order of 2V This incompatibility was removed by using a simple resistor bridge Decoder The DTMF decoder circuit was built and it was fed a direct I P from the DTMF generator circuit The circuit worked reliably outputting the appropriate binary codes The StD pin which signals the detection of valid tone pair at the input was found to be high each time FM receiver The FM receiver was built and tested by transmitting over our transmitter tuned to below the FM commercial band After various changes to the capacitor values in the local oscillator the right combina tion was hit upon The inaccuracy in calculation of the capacitor values arose due to the usage of hand wound inductors The receiver was able to clearly catch the words spoken into the transmitter when in a close range The link was thereafter also tested over a longer range and was found to be working reliably The link was then tested by transmitting different bit sequences from the DSP board The Std pin at the DTMF decoder that signals the reception of a valid tone was checked to be high each time The LEDs on the pcb displayed the tone number received Thus the link worked reliably for transmission and reception The microcontroller at the receiver was coded to receive the 8 bit device code serially and then compile it together to regenerate the transmitted sequence It would t
7. interface was tested by running the dummy com munication cycles 2 3 FIRMWARE 23 FM Transceiver Bit rate The link was checked by transmitting different codes continuously The time lag between the switching of different lines was measured The time lag was found to be about 1 second meeting specifications Range The link was tested for the range of approximately 50 meters The link worked reliably Error rate Virtually error free when tuned properly 2 3 Firmware The firmware code is developed using the Code Composer v4 1 It is an Inte grated Development Environment for programming C2000 series DSPs It has a C compiler an assembler and linker It also provides the debugging features through the JTAG emulation interface We used the XDS510PP PLUS JTAG pod for downloading and running the codes on our board The code is written mainly in C except for assembly code for some parts relating to the configuration of the DSP and the peripherals The basic support files were written for the register address definitions memory map of the target processor vector tables initial configuration of var ious control and status registers initialization of various peripheral modules like timers counters GPIO ADC module This involved writing mainly the MACROs and small functions assembly c 2 3 1 Routines and FSMs FSMs The receiver FSM is shown in figure 2 2 Routines The routines involved are classified as follows 1 Initial
8. switching on the DIP switch 1 before power on In this case the system erases the previously stored template and starts with word no 1 same as in initial blank state This is indicated by binary code 1 flashed along with the state LED The training sequence goes as follows 1 Pronounce the device name to be switched clearly in your normal voice at normal speed The system will flash the current word number as an acknowledgment that it has captured the features of the spoken word 2 Pronounce the same word again clearly The system checks whether the two pronunciations of the same word are close enough On close enough match the system will store the template of the word and flash the word number twice indicating that the training for the word has been successful If system doesn t find acceptable match betwen the two instances it will indicate an error by flashing the state LED The system will go back to the same word number and the steps 1 and 2 will have to be repeated for that word 3 On success the system goes into training of next word 4 The steps 1 3 repeat until all the 5 words have been successfully recorded 5 The system will automatically switch to recognition mode NOTE In case of training initiated using DIP switch turn the switch off before next power on otherwise the recorded template will be destroyed switches provided Refer to the section on training for using DIP switches 10 CHAPTER 1 USER MANUAL
9. used the TDA7000 chip for the same The IC has an FLL Frequency Locked Loop system with an intermediate frequency of 70 kHz The I F selec tivity is obtained by active RC filters The only function which needs alignment is the resonant circuit for the oscillator thus selecting the reception frequency Spurious reception is avoided by means of a mute circuit which also eliminates too noisy input signals Special precautions are taken to meet the radiation requirements The TDA7000 includes the following functions 1 R F input stage Hand wound inductors 7 turns of 22 gauge copper wire 130nH 2 Mixer 3 Local oscillator Hand wound inductors 5 turns of 20 gauge copper wire 56nH 4 I F amplifier limiter 70 KHz 5 Phase demodulator 6 Mute detector 7 Mute switch 2 2 3 Circuit diagrams The circuit diagrams are given in the appendix at the end 2 2 4 Block tests Word recognition 1 Mike and amplifier were tested with speech i p and waveforms as expected were observed 2 The LF2407A DSP itself was tested for program loading and running through JTAG 3 The memory interface and input output ports were tested by writing and reading specific patterns 4 The internal ADC of the LF2407A was tested firstly with waveforms us ing signal generator and then the actual speech The sampled data was dumped into the memory and the graph was viewed using the Code Com poser s graph utility 5 The audio codec serial
10. word spoken will be taken as the first word to be trained Please refer to training section for interpreting the response of the system NOTE Speaking into the system in training mode might cause inadvertent storage of the word Thus speaking frequently into the blank system should be avoided The user must carry out proper training sequence lif incorrect template gets stored in the system it can be erased completely using the DIP 1 4 OPERATION 9 The accuracy and range of the RF link can be tested using the RF Link Test which checks whether the codes are being transmitted properly over the link 1 4 Operation There are two basic modes of operation of the system e Training e Recognition The basic steps involved in the two operations are described below Training To ensure accurate recognition the training is essential for any new user intend ing to use the device The training sequence involves the storage of the user s template in the system memory The system stores template for only one user Thus device cannot be used reliably by more than one users simultaneously The template remains stored even after power off On power on the system fetches the templates stored in it If the number of words stored in the template is less than 5 the system goes into training mode from next word onwards This is indicated by flashing of state LED and word number on the word LEDs The system can be forced to go to training mode by
11. 10 011 11 00 Device 4 TV 010 100 10 01 Device 5 AC 010 101 11 00 Note As switching is in toggle mode the value of the on off bit has no effect Thus any one of the two 2 bit codes in the last two columns can be used The 8 bit code is transmitted serially using DTMF tone signalling with a unique start and stop tones Thus one communication cycle contains 10 tones interleaved with null tones 2 1 3 Alternative designs considered RF was preferred over IR due to the line of sight problems associated with the latter This enables the user to use the product from any position in the room Other modulation techniques such as QPSK were also considered but FM was finally chosen due to its ease of implementation and compatibility with the readily implementable tone signaling In word recognition the features are extracted for the whole word A sophis ticated technique involves detection and recognition of individual phonemes of the word This is essential for improving the resolution between similar sound ing words But this involves unrealizably high complexity in terms of time space and coding effort Pees Tecn ML fitta banke Calculate iremen log of fterbank output 18 CHAPTER 2 THEORY OF OPERATION Audio i p DSP w h built in ADC DTMF and 7 analog amplifiers Transmission windowin preamp AE 8 un Warping Figure 2 1
12. FFT algorithm is applied on each frame Fea ture extraction is done using Mel Frequency Cepstral coefficients MFCC The 15 16 CHAPTER 2 THEORY OF OPERATION 50 ED Mel Filter banks coefficients are those extracted from a triangular scaled filter bank This filter bank tries to simulate the human ear frequency response and the process is called perceptual weighting The Mel filter banks are applied to the FFT of the signal Cepstral smoothening is done by taking Log of the output of the filter bank The Mel Cepstrum coefficients are then extracted by taking IDCT Inverse Discrete Cosine Transform In our case for real signals and if we consider only the real part of the IDCT the IDCT operation is equivalent to taking a IFFT There are 15 coefficients per frame and thus 300 coefficients for a single word During the training sequence these coefficients form the stored templates During recognition the extracted coefficients are compared with each of the words in the stored template The distance between words is calculated using the DTW Dynamic Time Warping algorithm The minimum distance is compared against experimentally determined threshold value Mini mum distance less than threshold qualifies a match else mismatch is flagged In training mode the user has to speak the same word twice The distance between coefficients of the two instances is calculated similar to recognition The word is accepted only if the two instanc
13. SN74ALVC164245 is used to convert 3 3V level of DSP to 5V level and vice versa Audio Codec The board has TI TLV320A1C320 DSP codec for the mic and speaker interface The codec has 16 bit sigma delta ADC and DAC mic preamplifier antialiasing filter decimation filter interpolation filter It supports glueless serial interface to the TI DSPs For LF2407A the interface is through SPI Serial Periph eral Interface module The codec is suitable for audio applications and has maximum sampling rate of 22ksps 20 CHAPTER 2 THEORY OF OPERATION Analog Part Analog circuit consists of an microphone preamplifier It is the cascade of two inverting amplifiers using OP27 opamps OP27 is chosen for its better noise performance The each of the stages gives gain of about 10 thus making the maximum total gain of 100 Gain can be adjusted by changing the potentiome ter The optimum gain is adjusted such that the normal speech signal gets amplified to full scale voltage but without clipping The opamp circuit is de signed for using unipolar 5V supply available on the board using 7805 voltage regulator The output is obtained unipolar between 1 2 3 3V It is biased at 2 5V The output of the preamplifier is connected to the channel 0 of the inter nal ADC of LF2407A Zenner protection diode is used at the output to avoid damage to the ADC in case of output rising above 3 3V JTAG The board also has JTAG port for real time Scan Based Emulation IEEE S
14. Voice Activated Wireless Control of Devices in a Room Environment GROUP 16 Siddharth Verma 99D07006 Rhishikesh Limaye 99D07025 Sujay S Phadke 99D07027 Kavitha Subramanian 99D07041 Electronic Design Laboratory Department of Electrical Engineering Indian Institute of Technology Bombay Fall 2002 December 2002 Contents I User Manual 1 User Manual 151 Introduction zz 2 2 2 222222 1 2 Getting started w sa 2a mean a la a Ee 1 22 list oras Sasha 1 2 2 Specifications 5 ea ae EA Lisi initial testin Lisa 2 5 Seiser ee 1 Operation nn rara a 1 5 Safety precautions 1 6 Detailed Specifications II Theory of Operation amp System Design 2 Theory of operation 2 1 System description 2 1 1 Word Recognition 2 1 2 Transmission 2 1 3 Alternative designs considered 2 1 4 System block diagram 2 2 Hardware 2564 8 a en e A n 252 1 DSP Boards gs fiume p in O 2 2 2 SEM Transceiver eta ae a dost dock 242 3 Oireuitdiagrams used IS 2 22 23 Block tests 2 os 259 Firmw re sra Rak aa dd 2 31 Routinesand ESMS 23 27 ROUTE t
15. art and stop bits Some tones are used for checking the RF link Once the spoken command has been given and the microcontroller at the front end has extracted the 8 bit binary device code this device code is serially transmitted over the link The codes are designated as follows Each time a code has to be transmitted a 8 bit combination from the first 6 columns above and either of the last two columns above as appropriate is selected The 8 bit code is sent to the DTMF tone generator serially The tones corresponding to each bit is then sent to the input stage of the transmitter A start bit is sent at the beginning of each 8 bit code and a stop bit is sent at the end of it bringing the total to 10 tones per ID The DTMF tone generator produces a combination of two frequencies one from a high range and one from the lower range as a tone This makes the transmission robust and minimizes interference from room noise The DTMF generator is also used for signaling in telephone lines The input to the DTMF generator is a binary combination selecting the tone to be generated There is a matrix of 4 rows and 3 columns to choose the tone from and the tone is selected by grounding the appropriate row and column as is on the telephone dial pad The DTMF O P is a tone combination at 2V and has to be stepped down to a few milli volts for the transmitter to transmit it without distortion This is performed simply using a resistor bridge at the transmi
16. as a voltage of at 9V Also you may switch on the main switches and check if the power on LED turns on Place the transmitter section at a suitable location such as on the wheelchair where the LEDs are visible 1 2 2 Specifications Word Recognizer The recognition system is capable of recognizing upto 10 isolated distinct words spoken clearly and at normal speed In the device the number of words is set at 7 8 CHAPTER 1 USER MANUAL 5 words Thus 5 devices can be controlled using the system The recognition is speaker dependent and each user has to undergo a training sequence initially to store the templates of his her speech The system is restricted to single user as multiple user templates are not stored There is a delay of about 5 secs between speaking of the word and its recognition or rejection FM Transceiver The transceiver has been designed to work below the commercial FM frequency band This avoids interference and minimizes the chances of jamming and hence can work in an electrically noisy environment The transmission of device code is done using DTMF tones It takes about a second to transmit the code after the spoken word is recognized correctly The receiver applies error checks on the received code and in case of error it cancels the operation i e switching is not performed In such case the command has to be given again If correct device code is received the corresponding device is toggled For reliable receptio
17. es match within the threshold The average of the coefficients of the two instances is stored in the template The algorithms are implemented on a TI LF2407A DSP The details of hard ware and firmware are described in subsequent sections im 2 1 SYSTEM DESCRIPTION 17 Detect start of Kraiz the 0 5 utterance and y pre emphasis accom temple epectrum o gach abore neat 0 5 Alter J0 Tramea iti recognition lee Fa to training findibhecomect determane cos recip raith for pach word e trasnina Sore the 19520 a Tor each word ina ASR Block diagram 2 1 2 Transmission Protocol The device code transmitted using the FM transceiver is 8 bit code having the following format 3 bit switchboard id 3 bit device id Reserved on off 1 bit parity The switch board id uniquely identifies one of possible 8 switch boards in the room The device id identifies the particular device among possible 8 devices on the selected switchboard Currently the switching takes place in toggle mode thus the on off bit is kept reserved The last parity bit makes number of 1 s in the code odd Thus it is used at the receiver end for error checking The codes we are currently using are Device Switchboard ID Device ID on parity off parity Device 1 light 010 001 10 01 Device 2 fan 010 010 10 01 Device 3 tube 0
18. ests sa s nie 2 4 System integration and test III Service Manual 3 Service manual 3 1 Problem identification 13 15 15 15 17 17 18 18 18 20 22 22 23 23 25 25 29 31 4 CONTENTS Appendices 33 2 2 Circuit Diagrams ae e ae 35 Part I User Manual Chapter 1 User Manual 1 1 Introduction This device will enable you to wirelessly control the switching of various devices in the room environment merely using the spoken command It is aimed at the handicapped user who cannot push buttons on a remote but also has universal appeal as an easy and comfortable way to switch devices without interrupting the work at hand This manual gives the information you would need to use the device or to service it Please read the manual carefully before operating your unit 1 2 Getting started 1 2 1 Initial check list Your voice activated wireless controller should contain the following units e Word recognizer unit with LED display and mic e Transmitter unit with 4 wire connector to connect to word recognizer e Receiver to be mounted on the switch board e 9V alkaline battery for receiver All the units are preconnected so no assembly is required on your part Please attach the two 9 volt batteries in the LED display unit and the switch board unit in the space provided You may use a multimeter to verify that each battery h
19. g the frequency across the inductor of the L O else replace the capacitors or inductors appropriately The final O P of the receiver should be a tone pair similar to the one generated at the transmitter Else try replacing the active components DECODER If the tone is received properly at the input to the decoder but isn t being recognized then check the DTMF detector If the StD pin is high at each correct tone transmission check the microcontroller Either the microcontroller has malfunctioned or there is some bug in the code The microcontroller can be checked by checking the ALE pin 31 Part IV Appendices Appendix I 3 2 Circuit Diagrams The circuit diagrams for mic preamplifier DTMF encoder FM transmitter FM receiver and DTMF decoder are provided The circuit diagram of the DSP board is too complex to be given in the report Please refer to the schematic and board files edl16dsp sch and edl16dsp brd in Eagle 4 1 35 36 MIC Pre Amplifier _ To olp port of DSP through level shifter Figure 3 2 encoder 3 2 CIRCUIT DIAGRAMS 37 MIC PREAMPLIFIER EDI SCHEMATIC DIAGRAM FOR ES 17 Figure 3 3 FM Transmitter 2 TONE IN Figure 3 4 DTMF Decoder 38 Figure 3 5 Receiver TDA7000
20. gnal whose output is known or can be calculated The observed results were compared to the expected ones Then several routines were combined together and their behavior was tested using standard input as well as the realistic input The test criteria include the accuracy amount of memory space consumed execution time 2 4 System integration and test DSP board 1 Firstly the voltage levels at various nodes on the DSP board were mea sured The system clock CLKOUT of DSP was observed for its fre quency The SRAM was tested using pattern test The program loading and smooth execution of the code was tested through the Code Composer Analog Part The amplifier was tested first with regular input in the form of sinusoidal waveform and then the actual speech signal was applied The output waveform was analyzed so as to adjust the gain of the amplifier such that there is near full scale amplification with no clipping The ADC module of the DSP was tested with regular sine wave at the input of the preamplifier The mean level and maximum minimum were observed The mic was connected and the noise threshold levels were calculated Then the actual speech signal was captured and plotted in the graph utility of the Code Composer The time domain waveforms of the known words like one two etc were found to be satisfactory as compared to the waveforms obtained in simulations The start detect algorithm was tested for its correc
21. hen write 1 to the output line corresponding to the appropriate device This link was tested using all the blocks The trans mitter was coded to transmit all the codes in sequence twice Thus at the receiver each device was switched on first in turn and then switched off This was displayed by the LEDs at the O P port of the microcontroller Thereafter the link was tested for speed of switching Each time the delay introduced by the delay routine at the transmitter was reduced It was found that the link worked reliably with switching delays of around 2 4 SYSTEM INTEGRATION AND TEST 27 100ms Thus allowing transfer of one byte within one second as per the specification 28 CHAPTER 2 THEORY OF OPERATION Part III Service Manual Chapter 3 Service manual 3 1 Problem identification 1 Check whether both the Power On LEDs are shining on the DSP board If not then check whether the power supply wires are connected securely and polarities are correct Check the voltage using multimeter 2 Check whether the state LED glows everytime you speak into the mic If not then decrease the mike distance from mouth and try again If not still then press the reset switch and try again If it is still not responding then adjust the potentiometer in the preamp circuit to increase the gain of the amplifier If possible the output waveforms can be checked at the test point provided 3 Check whether the red LED at the receiver lights up
22. ization routines initDSP initBoard initTimersA initADC These configure the CPU and peripheral registers for desired operating mode of the hardware 2 External FLASH access routines flash_clear flash _BLKread flash_BLKwrite 3 Interrupt Service Routines bad_trap nmi_trap and c_intl which catches the ADC end of conversion interrupt and analyses the sample data for start detection 4 FFT routines fourl from Numerical Recipes which works with float data and intFFT provided by TI for integer FFT on C2xx DSPs 5 Core routines processData which forms the core of the speech recogni tion algorithm It starts with stored word and calls FFT routine extracts the coefficients and passes them to the DTW routine recognitionTrain ing and minTraining 24 CHAPTER 2 THEORY OF OPERATION If less than 9 bits recvd or stop bit not recvd invalid valid Figure 2 2 Receiver FSM 2 4 7 SYSTEM INTEGRATION AND TEST 25 RF link interface the routine rf link test carries out RF link test by transmitting all the tones that are used during normal transmission The routine xmitbyte takes a byte and transmits it serially with interleaving null tones with suitable delay main as always the starting point of the program execution 2 3 2 Routine tests Each of the routines was tested individually The test procedure involved ap plying the routine under test to standard input si
23. n the distance between transmitter and receiver should not be more than 50 meters Front Panel The front panel consists of two red power LEDs in the lower right corner The main display LEDs include three green word LEDs and one red state LED The green LEDs denote binary coded integer between 1 to 5 which indicates the index number corresponding to the recognized word The red state LED denotes various states like detection of start of word mismatch in recognition There is also one red RESET LED The system also has array of 8 DIP switches which can be used for configuring the system for initial use or in case of some problems 1 3 Initial tests After switching on the device the both the power LEDs should light up It should be followed blinking of RESET LED indicating that the system has started execution The four display LEDs should flash 3 times indicating that the system initializations are over and system is ready for input In a blank system the training is required before it can be used Thus the system automatically goes into training mode This is indicated by flashing of state LED and binary code 1 on the word LEDs This denotes that user should start training from word no 1 For detailed operation of the training sequence refer to the next section To test whether the system is working smoothly user can speak any word into the mic The state LED should light up indicating start of word But as the training is not done the
24. s using 10 bit ADC This sampling rate is sufficient for speech signal which has the bandwidth of about 4 KHz The start detection is done by comparing the signal energy and zero crossings with the background noise thresholds The noise thresholds are calculated at system startup The noise thresholds are checked for their cor rectness against the expected range of values This ensures that the calculation faithfully represents only the background noise and not the actual signal Once the start of word has been detected the system starts storing the incoming signal as spoken word The end of the word should also be detected but this involves quite elaborate techniques as compared to the start detect and it often gives incorrect results e g if a word has a silent portion then end detect may detect it as end of word thus losing the rest of the word On the other hand a simplistic model stores the speech for fixed amount of time after start detect We chose the latter model with word length fixed at 0 5 secs This gives opti mum combination of feature extraction distinction system memory resources and time for analysis The captured speech undergoes various pre conditioning operations like pre emphasis filter framing applying hamming windows These operations are executed real time as the samples are being stored After entire 0 5s duration word has been stored it is divided into 20 frames of 256 samples each with an overlap of 56 samples The
25. tan dard 1149 1 which can be used for in system loading and debugging of programs The board also has provision for RS232 interface using Maxim MAX3221 chip It can also be used for program loading after enabling the boot ROM inside the DSP 2 2 2 FM Transceiver The wireless link of the product is designed modularly The functional require ment of the wireless link is that it should be able to reliably transmit and receive the codes corresponding to the different devices to be switched and switch the appropriate line high This wireless transmitter takes in the device code of the device to be switched on from the word recognizer front end and transmits it over the channel The receiver then receives the code decodes it and switches on the appropriate line The modulation technique used is frequency modulation The transmitter side is divided into two modules one to obtain device codes and convert them into appropriate format for transmission over the link and the other to actually perform the transmission at high frequency Similarly the receiver side has two modules one to receive and demodulate the high frequency signal and the other to identify the device from the signals obtained 2 2 HARDWARE 21 Transmitter side Encoder This project uses tone signaling for transmitting bits We use a DTMF Dual Tone Multiple Frequency tone generator for the purpose Two tones are set aside for the two bits 0 and 1 and two tones for the st
26. tness and reliability The thresholds were adjusted so that only a well spoken word is captured The speech recognition algorithm was tested on the TI 6211 DSK DSP Starter Kit On our board the testing was carried out by dividing the code into logically separate processing steps and the data after every pro cessing step was observed in the graph utility The entire speech recognition algorithm was combined and tested for recognition accuracy and time for analysis 26 CHAPTER 2 THEORY OF OPERATION FM TRANSCEIVER 1 Testing the encoder Different tone codes were given as I P to the DTMF generator through its input pins The O P tone pairs generated at pin 16 of the DTMF chip were observed and were as expected Thereafter different bit sequences were dumped onto the input pins of the 5089 and the O P tone pairs were confirmed The FM transmitter was built and its range and tenability tested It worked reliably with clear reception using normal FM radio over range of 50 meters Voice input was used to test the link Also the settings of the trimcap was disturbed a few times to test how easy tuning was It was found to be easy enough to tune the transmitter to the receiving station Thereafter the DTMF tone generator O P was fed as an I P to the trans mitter However the received tone checked using a jack was highly dis torted After a lot of testing it was identified that the transmitter takes I P signals only
27. tter I P FM Transmitter Blocks First amplification stage This is a standard common emitter amplifier A capacitor isolates the microphone from the base voltage of the transistor and only allows alternating current signals to pass Oscillator stage The tank circuit the transistor and the feedback capacitor are the oscillator circuit here An input signal is not needed to sustain the oscillation The feedback signal makes the base emitter current of the transistor vary at the resonant frequency This causes the emitter collector current to vary at the same frequency This signal fed to the aerial and radiated as radio waves Note that the tank circuit does not oscillate just by having a DC potential put across it Positive feedback must be provided The green trim cap will enable you to tune the Tx A lower value of the capacitor will move the frequency up towards the other end of the FM band Final Amplification Stage This RF stage adds amplification to the RF signal It needs an RF transistor to do this efficiently We use a 1018 L2 an RFC radio frequency choke and the capacitor in parallel with it are designed to reduce harmonics from the circuit A small coupling capacitor on the aerial is optional to minimize the effect of the aerial capacitance on the final stage LC circuit 22 CHAPTER 2 THEORY OF OPERATION Receiver side FM receiver The receiver has to perform the demodulation of the received high frequency signal We
Download Pdf Manuals
Related Search