Home

観測信号の分布に混合ガウス分布を、音声信号の振幅の事前分布に

1. 3900 46 0 46 0000 Re X D Im X D sa B 0 X 39 00000000000 x acos 46 V z2 y 48 y asin0 47 0 tan 49 00000000000000 0000 2000000000000000 ex 23 50 exp 2 51 D000 x y000000000 p x p x p y 52 2 9 000000 70000 53 0000000000000 p a 0 J p x y J p x p y 53 39 00000000 70000000000 dr J IPS IPS Ox Oy Ox Oy 06 cos 0 a cos 0 sind asin 0 a cos 0 sin a 54 000000 53 0000000000000 p a 0 J p x p y oie C2 Gs CE 4 a x P 202 2 55 P e 000000 9000000000 27 va 0 27 a 2 i f P xz a e 2 OMEN T 202 a a exp 56 0000000000000000000000000000000 0 00 000 0 00000000000000000000000000000000000000 00000 40 Parametric MMSE STSA 00000000000000000000000 20000
2. 512points 32ms 256points 16ms Viterbi 300 MFCC 0 97 B 2 2 X000 1 00000000 3600000 36 BX NOISY phase gx CLEAN phase 90 80 70 60 50 40 20 10 1 m oo D I 0000 36 1 www OOdB B C3 3 05dB lodb EHE 1098 www 05d8 Wak lodb Ww 2056 Z L 00dB 1 11 S 5 ww 00dB 0505 OodP aww 100 www 1048 1508 ww 02 www C 10dB ie m e 000 00000000000000000000000000000000000000 00000000 36000000000000000000000000000000 0000000000000 50000000000000000000006000 0 3 52000000000000000000000000 00000000 5 0000000000000 0 0000000000000000000000000000000000000 00000000 000000000000000000000000000 0 000000000000000000000000000000000000 0 0000000000000000000000000000000000000 0 00000000000000000000 37 c 000000000 0000000000000000000000000000000000000 0 0000000000000
3. 6 0000 0 0000000000000000000000000000000000000 Parametric MMSE STSA GM 0000000000000 Parametric MMSE STSA 0 1 MMSE STSA 10 0 0 Parametric MMSE STSA 2100000000000 0 000000000000000000000000000000 MMSE STSAD Parametric MMSE STSA IOO0000000000000 0 0000000000000000000000000000000000000 000000000000000000 8 000000000000000000 0000000000000 58 000000000000000000000000 27 00 00000000000000000000000000000000000000 0 00000000000000000 0000 1 Ephraim and D Malah Speech enhancement using minimum mean square error short time spectral amplitude estimator IEEE Trans Acoust Speech Signal Processing vol ASSP 32 pp 1109 1121 Dec 1984 2 I Andrianakis P R White Speech spectral amplitude estimators using optimally shaped Gamma and Chi priors 2008 Elsevier 3 J S Lim A V Oppenheim Enhancement and band width compression of noisy speech Proc IEEE 67 12 1586 1604 1979 4 Patric J Wolfe Simon J Godsill Simple alternatives to the ephraim and malah suppression rule for speech enhancement IEEE 2001 0 7803 7011 2 01 5 Patrick J Wolfe and Simon J Godsill Simple Alternatives to the Ephraim and Malah Suppression Rule for
4. Frequency Frequency 35000 30000 25000 20000 15000 10000 5000 16000 14000 12000 10000 8000 6000 4000 2000 Probability meme bus j Re 4 2 0 Amplitude 0000 29 2 Fase offi j Re 4 2 0 Amplitude 30 000000 00 0 4 T T 75 0 3 0 2 j Re Gaussian 0 Amplitude 0000000 31 32 Frequency Frequency Probability 35000 T r 30000 25000 20000 15000 10000 5000 i ai mm bus jim 4 2 0 Amplitude 2 32 0000 16000 14000 12000 10000 8000 6000 4000 2000 0 offi j Im 4 2 0 Amplitude 00000 33 0 4 TF white Re Gaussian 6 4 2 0 Ampli 000000 34 00000000 00000000000000000000000000000 0 00000000000
5. 3000000 30 Frequency Frequency Frequency 2500 T T 2000 1500 1000 500 T white Im 0 Amplitude 18000 T 16000 14000 12000 10000 8000 6000 4000 2000 T rest jim t 3 0000 24 0 Amplitude 16000 T 14000 12000 10000 8000 6000 4000 2000 j Im U 25 0000 0 Amplitude 31 Frequency Frequency Frequency 2500 2000 1500 1000 500 white Re 16000 14000 12000 10000 8000 6000 4000 2000 16000 14000 12000 10000 8000 6000 4000 2000 26 000 Amplitude 00 0 T T il rest j Re 3 4 2 Amplitude U 27 0000 00 j Re ne E sect 4 2 28 0000 0 Amplitude
6. Hu hh TERETE MMSE STSA 0000000000 AE HATET EME 000000000 0000000 1130067 0 00 00 250 10 300 0000000000000000000000000000000000000 000000000000000000000000000000000006000 0 00000000000000000000000000000000000000 000000000000000000000000000000000006000 0 000000000000000000000000000000000006000 0 000000000000000000000000000000000006000 00000000000000 0 00000000000000000000000000000000000000 0 00000000000000000000000000 Parametric MMSE STSA 0000000000000000000000000000000000000 000 20000000000000000000000000000000000 OO0O000000000 000000000 Parametric MMSE STSA GM MMSE STSA Parametric MMSE STSA 2 0 00000000000000000000000000000 500000000000000000000000000000000000000 00000000 MMSE STSAQ Parametric MMSE STSAQG 000000 000000000 Parametric MMSE STSA 00000 MMSE STSA 915 Parametric MMSE 0000 00000000000000000000000000000000000000 0 00000000000000000 5
7. lt 76 Parametric MMSE 5 45 000000000 00000000000 000000 29 0000000 OOOO 000 0000000 0 00000000000000000 0 0000000000000 2000000000 f p pD AD f p Ne Aq 77 0000 78 00000000000000 p D r N 0 234 1 N 0 wa 78 0000000 2000000000 78 000000000000000000 E DP f fiot r N o Da r N 0 Vuda aD r f pe 0 Da aD Q 7 f IDP x o RE EE 79 D uuu U UU 79 0000000000000 2 Aq 1 80 O00r000000000000000000 1 r 81 2 000000 29000000000 46
8. bus restaurant 5dB OdB 508 10dB 5dB 008 5dB 1048 clean 0 20 00000 0000000000006 25 NOISY m MMSE STSA B P MMSE STSA B GM init multi B GM init fixed m CLEAN mNOISY m MMSE STSA mP MMSE STSA B GM init multi MGM init fixed m CLEAN zn 70 E 60 mNOISY 50 BMMSE STSA xx 40 P MMSE STSA 96 30 MGM init multi 20 MGM init fixed 10 MCLEAN u ail 5dB 008 5dB 10dB 5dB 008 5dB 10dB office shop clean 0 21 00000 0000000000000 File Edit Transform View DG E SR 2 BB amp 00 000 off j 50 4 9 Configuration Demonstration _ WaveBar 00 000 000 5 0 22 010 0000000000 00 00000000000 Wavesurfer 150 00000 26 5 7 3 OOO00000000000000000000000 Parametric MMSE STSA UUOUOO0O0O00O0000000 0000000000 Parametric MMSE 5 000000000000000 OO00000000000000D00000000000000000000000 00000000000000000000000000000000000000 0000000000000000000000000000000 40000000 000000000000000000000000000000000006000 0000000000 22000000000000000000000000000 0 000000000000000000000000000000000 00000000 MMSE 5 000000000000000000 15 00000000000000000000000000000 0 000000000000000000000000000000000 00000000000000000000000000000000000000 0 000000000000000000000000000000000000000 E 0 0 0 0
9. Spectral Subtruction 3 000 550 00000000 50 ETT ONE ET EET AER VE ET EE VET META Ed 0 Ea Er EJ Minimum Mean Square Error 00000 MMSE Short Time Spectral Amplitude 1 0 0 0 MMSE STSA 00000000 00 000000000000 MMSE STSA HOO0000000000000000000000000000 OOOOO00000000000000000000000000000 Parametric O00000000000000000000000 O00O0O00000000000000000000 00000000000000000 Parametric 00000000000000000000000 0 0000000000000000000000000 MSE STSA GMO000000000000000000000 000000000 2 00000 OO0000000000000000000 0000000 000000000 0000000000000 21 0000 OOOO00 0000000000 w00 1 000000000 00000 4 000000000 0 000000000000000000 90 00000000000000 22 000000000000000000000000000000000000 N 000000000000000 00 2 0000000000 00 000 0000 2 00
10. Xy 79 44 1 1 0006000 600000 04 0000 SNR 0dB 512points 32ms 256points 16ms Viterbi 3 0 97 1 2 0000000000000 35000000000000000000 CLEAN 0 D 000000000 500000 34 100 9 8 70 aM Hi MM i 6 5 4 30 20 1 M 0 EM ME m CLEAN white airp j bus j rest j offi j shop j 1 SNR OdB 35 00000000000000000000000000000000000 o O EXT IIT oe O 5 00000000000000 CLEAN white airpj busj restj offij shop 11 1 97 90 94 48 94 62 95 02 95 37 95 37 94 38 CLEAN 00 0 00 3 42 3 28 2 88 2 53 2 53 3 52 35 B 2 00 2 0 45 00000000000000000000000000000000000 000000000000000000000000000000000006000 00000000 00000000000000000 Xi 4 45 2 1 000 200000000000000000000000 600000 000 X X 4 79 0000000 Ooo X X 0000000 0 6 01000 ATR set A white airp j bus_j rest_j offi j shopj 1 SNR 001 5dB 10dB 15dB Parametric MMSE STSA 2 0 3
11. 0 a Ey ta PRG 1 0 lt a lt 1 37 B8 00000000000000 3 P 38 0 otherwise 54 0000 SNROOOOOSTSADOOOOOOOOOOOOOGDOOOOOOOOOD 0 000000000000000 0000000000000000000000000000000 03000000 OOO 4 0000000000000000000000000000000000 D uui 5 4 1 0 00000000000000000000000000000000000000 00000 39 000000 12 T Yea ifi lt Tra 39 1 1 2 22 0 Yt otherwise 7 frames D D D 15 5 4 2 0 0000000000000000000000000000000000 1 000000 0000 SNROD 40 000000 Yk w 40 2 0 4 000000000000 2 00000000 Wk 00 4200 000000 1100000006000 24 W W 41 1 if Y t n 9 1 4 0 elseif k gt Tz 42 eld 52 otherwise 72 7 0 000000000000000000000 LE 1 0000 3 000000 0 2 000000 1 2000000000000 000000000000000000 2 0 000000000 2 0000 0 000000000 0000000000000000 0 000000000000000 2000000
12. 2 1F n 0 5 1 0 0 2 68 68 65 0m xs 2 E CR P g o 1 7 1 2 2 E exp i EAE Jo 2i 0 A n w 2 n fr R 1 1 i 1 Bil Aar m z 2 5 A n 2 27 1F n 0 5 1 2 2 exp R n J Nn 2 w waj 2 2 2 esc GO en C nimm D 18 10 5 1 me n 5 ee 3 69 a n w 0 670000000000000000000000000000000 1 R 00 zx GE Lew 4 rep ane un tiet on s 2 ep SE i Gti 70 44 57 0000000000000 7 wa 2 3 n 7 2 m E A n 2 exp C iF 1 A n 2 oo 3 n z nex 1 2 2 n 1 n w A n exp amp 1 7 1 0 n w T 9 m T n b AGI 2 exp g 1F 5 n 15 2 E A n 2 exp amp FAC 7 1 v n 2 A n w 2 exp amp 1F 72 Au exp S 801 15000 72 000000 00000000 amp 00000000000 73 0000000 fr 2 0 5 1 r fi w 0 5 r 2 0 1 r filw 0 2 fel B I PB n als 9 em Ri 73 A amp Vk 1 n 1
13. 8000000000000000000 8 000000000000000000000000 0000000000000 00000 1 0000 2 00000 2 1 2 2 2 3 2 4 0000 MMSE S TSA eee E 3 2 Parametric MMSE STSA 4 Parametric MMSE STSA GM 000 4 1 42 00000000000000000 4 2 1 4 2 2 5 0000 Meat cas 5 1 5 2 5 3 5 4 5 5 5 6 INNE 200000000 2 10 10 11 11 12 5 6 2 5 6 3 57 5 7 1 5 7 2 5 7 3 00000 A 1 1 2 00 A3 A4 5 DDD D Uu BA 001 1 1 1 2 B2 D02 2 1 2 2 B3 00000 031 c 2 000 c3 1 0 000000000000000000 a SR I HERE D uuu LM ROTE MEE Sota treer Kast FE 1 A 27 28 30 1 11 000 D Parametric MMSE STSA 000000000 00000000000 Hi 39 41 46 SESS HEEL EE u EI 1 qu ES 0 000000000000000000000000000000
14. 7 UL 62 1 A 7 1 mg 3 ou ve 59 60 I 61 000000 58 00 0000000000000 000 27 ED 0 z d _ 27 s Hee R ro 20 m afit a 2 64 42 0 000000 00000000000 64 00000 57 000000000000000000000 VEM fr f gt 1 2 R a gt 2 00 SENC z uf a exp xd exp ae exp 2 9 da da gt 27 pi J a exp 122 exp 59 0 da da w Jo WAq 0 1 ys Jr f gt n a v m2 2 TAI n Hi Hle cu NL ME 2 251 oo 2 2 2 exp 53 2 n UR a 10 0 Ax A n w 27 1 n gt v mz 2i Tal n E A N na 1 R ve 2n v mw n exp exp Le 5 da Am esa n Sha a CC mm Css As 94 a et dis 7 ne 4 A n w 65 000000 17 6 6310 a 2 _ P T sv 3p 3 v p l g x exp az mm 0 1 Y 1 66 0 5 270000000 BY n 5 p 2 DE _ 2 LE a exp aa da 243 iF n 0 5 1 2 _ T n 2 Sie JUD a n 051 67 43 000000000000 o sx 29000000 7 5 p exp aa da P 1F1 0 5 1 1 E E 5 A n z
15. Speech Enhancement IEEE 2001 6 Y Ephraim D Malah Speech Enhancement Using a Minimum Mean Square Error Log Spectral Amplitude Estimator IEEE Transactions on ACOUSTICS speech and signal processing vol ASSP 33 No2 APril 1985 7 ChangHuai You SooNgee Koh Susanto Rahardja Adaptive 8 Order mmse stsa estimation for speech enhancement IEEE 2003 0 7803 7663 3 03 8 Rainer Martin Speech Enhancement Based on Minimum Mean Square Error Estimation and Supergaussian Priors IEEE Transactions on speech and audio processing Vol 13 No 5 September 2005 9 007000000000000000000000 0000 IEICE Tech Report SP2010 103 2011 01 28 19000000 0000000000 http sugiyama www cs titech ac jp sugi 2007 Canon MachineLearning28 jp pdf 2012 01 3 11 Ambient Noise Database for Telephonom etry 1996 0 0000 1996 12 Yutaka TSUBOI Takehiro IHARA Kazuyuki TAKAGLKazuhiko OZEKI The Use of Overlapped Sub Bands Multi Band Multi SNR Multi Path Recogni tion of Noisy Word Utterances IEICE TRANSACTIONS on Information and Systems Vol E91 D No 6 pp 1774 1782 13 Rainer Martin Spectral Subtraction Based on Minimum Statistics Proc EU SIPCO 94 pp 1182 85 Sept 1994 0 000 0000 0000 000000000 MMSE STSATI D D DI C LI U 0 0 00000000000 A Vol J87 A No 7 pp 851 860 July 2004 14 15 Centre for Spe
16. 00 000000000 000 22 00000000000000000000000 000000000000000000000000000000000006000 0 000000000000000000000000000000000000000 000000000000000000000000000000000000000 OOO00000000000000000000000000000000000 0000000000000000000000000000000000000 UOUO0O000000000000000000000000000000000 A 5 000 1 cosh z 00000000000000000000000000000 1 200000000000000000000000 00000000000000 20000000000000000000000 33 B 1000000000000 00000000000000000000000000000000000000 0 0000000000000000000000000 1 5 00000000000 BA 001 44 00000000000000000000000000000000000 000000000000000000000 00000000000000000
17. 000 250000 40000 ag 63 w as 25 L 11 23 0 60000 2966 2965 6 00000000000 26 27 28 0 000000 Ep sa 26 NE x P x Q z 1 03 27 wA r a wx ns 26 27 28 000000 24 00000000000000000000000 000000000 4 2 2 200000000 0000000 Q 0 0 20000000000000000000 0 OOO 29 000000000 FOO O 1 29 29 OOOO G390000000000000000000000000000 300000000 05 z w 30 9 62 oz000000000000000000000000000000 31 0000 940000 9g 02 230 amp 0000 20 2902 00000000 32 33 0000000 12 EXE e C14 0 3 Oz z 23A 2 e a 1 w e gt 1 2 2 gt P x Q z 1 02 dz 32 2 2 22 f4 _ 1 u 2 d ipe o 99 62 ay 1 w z as mw TWA 2 2 2 e TE w g 1 P x Q z 1 02 dz 33 32 33 000000 240000000000000000000000000 0000000 13 5 0 00000000000000000000000000000 MMSE STSA 1 0 Parametric MMSE STSA 2 0 0000000000000000000000 Parametric MMSE 5 0 000000000000000 000000 000000000000000000000000000000000000000 51 00000 0000000000000 0
18. 0000 C 1 O0O000000000000000000000000000000000000 0000000 900000000000000000000000000006000 9000000000000000000000000000000006000 0 00000000000000 C 2 0000 000000 700000 07 0000 ATR B fknsda01 mhosda01 512points 32ms 256points 16ms 00000 fknsda 0 00000000000000 3700 25034 01 00000000000 3800000 fknsda01 1 6e 06 T 250000 mhosda01 1 4e 06 200000 F 4 1 2 06 god 1e 06 7 a v 2 d 2 t E 3 150000 NO Ae uet amp 800000 ftit YN jar E he Ty p 6 il E ma c pp 25 ei 4 8 600000 m 1 1000 NTN E JET A 4 TN te s ra P ER p eU e FA hat e IH e EUM 400000 iod usc AS F E grat He an Sabe d 50000 er E zr Ts did 200000 E Vs A a 0 0 pi pi 2 0 pi 2 pi pi pi 2 0 pi 2 pi 37 fknsda01 g HU uuu C 4 O00 phase 38 mhosda01 00000000 0 0000000000000000000000000000000000000 Ee 000
19. 0000 DDOO00000000w7000U0000000UUUUUUUUUOUO 0o00000000000000000000000000000 O00000000000000000000000000 OO000000000000000000000000000 Parametric MMSE STSA 00000000000 0000000000000 0000000000 00000 0 0000000000000000000000000000 0000000000000 9 30000000000 19 5 6 0000 2 000000 Parametric MMSE STSA lt 0000000000000000 D uuu 5 6 1 000 40000000000000 SNRQ 5dB 5dB 1008 000000 0 0000000000000000000000000000 700 5 500000 000000 0 3000000000000000 10000000 1 2params init multi 21000000000000000000000000000000 2 04 1 00 0 10 w 1 00 3 00 0 2000000000000000 72000 000000000000000000 2 3params init multi 300000000000000000 0 r 0 10 1 00 0 10 2 0 10 1 00 0 10 w 1 00 3 00 0 2000000000000000 70000000000 00000000000 3 2params init fixed 200000000000000000 0000000000000 1 0 2 00000 4 3params init fixed 310000000000000000 200000 1 0 2 0 5 5 6 2 400000000 1500 1700000 20 100 SHB 0 288 i 8 8 Wl2params init
20. 00000000 700000 7 0000000000000 14 00000000000000 000000 0 2 00000000000000 7 0 9 dB To 7dB Tini frames Tz 20 frames 16 5 5 00001 Parametric MMSE STSA 000000000000000000000000 0000000000000 5 5 1 Parametric MMSE STSA J00 00000000000 70 n 0 1 0 3 0 5 0 7 0 9 1 0 1 5 0 00000000000000000000000000 00000 0 lt 7 lt 10 DUOUO0OU0000000007000000000000000000007 150 0000000000 SNRQ 0dB 10dB 154 DDOUO00000000000U 10000000 5 5 2 0000000600000 0 0000000000000000000000 12 13 14 00000 1088 1598 10dB 15dB white inso 12 00000 000000000 100 amp amp 10 Bl n 0 1 El n 0 3 Bl 0 5 n 0 6 n 0 8 n 1 5 SEE o i 8 8 8 17 EP CIT 8 8 8 EEE 8 8 8 100 100 0 0dB 5dB 10dB 15dB 0dB 5dB 10dB 15dB bus restaurant 0 13 00000 0000000000006 0dB 5dB 10dB 15dB 0dB 5dB 10dB 15dB office shop 0 14 00000 0000000000 18 n 1 0 Bl n 0 1 n 0 3 Bl n 0 5 n 0 6 8 n 1 5 1 0 Bl n 0 1 El 0 3 Bl n 0 5 n 0 6 n 0 8 n 1 5 90 3000000000000000000000 gy Ed IE 70 ost 050000000
21. 0000000000 000000000000000000000 Parametric MMSE STSA GM do k ak p Y plak dar dak ig p Y plak dek dak 000 FM 0000 zu dis p Y da da ET mee exp r N X 1 r N X da d Et 7 1 A I 2r TG 2 exp La rN X ZAq 1 r N X wha da da 1 7 27 5 arier Ze oo nox wra da d T 0 0 SEN 7 2 R 2Ra 9 a mA b s Da 59 27 2 2 ace gi a2 exp ze exp R a Faces 2 a da w 0 x 0 Ad EA 7 2 rg 2Ra mAr mn 2 d z d 0 z d 1 9 Beg 7 27 E n exp exp 2600 280 9 da da 0 WAq 0 57 58 000000 17 eq 8 431 5 In 2 ge eos 0n esp zcos 8 48 000000000000 smesso 00000000000 7 eq 8 406 3 Eq us 59 A1 00000000 z Jo iz 60 0 0000000 000000000000000 000000000000000 R2 x 1 2 ax 1 2 2202 N iX R AJM IA 4 2 Aq 21 2 E og Eten 2 LX 1 TNE v n 2 61 Xn 2 000 a amp
22. 0000000000000000 00000 00000 00000000000000000000000 11000000 500000000000000000000000000000000006000 16 00000000000000000000 1000000000 1 ATR 281 30 faf ffs fym mau mht mtk 11 5 512points 32ms 256points 16ms 16kHz Viterbi 12 300 4 HMM 12 0 97 34 0000000000000 80000000000000 000000000000000 8000 35 000000000000000000 00000000 SNRU 98 0000000000 Yn Zn dn 34 1 ico 35 TENE 200000000000000000000000000000000000 00000000000 14 52 0000 36 0000000000000000 _ 00000 5009 19 36 00000 5 3 0 Decision Directed 00000000000000000000000 5800000000 000000 58 0000000000000 Deciion Direted 0 1 00000 0000000 80000 9000000 i
23. 000000000000000000000000 4 000000 900000 13 0000000 00 B00 0 0 Xp 79 13 1 1 1 0 1 2 pi pi 3 2 pi 2 pi a O05 000000 DU 1 U 6 0000 1 000 00 700000 3 2 Parametric MMSE STSA Parametric MMSE STSAN OOO 2 00000000000000 03 100 MMSE STSAQDOOOOOU00U000000000000000000000000000 000 800000000000000000000000000000000000 00000000 14 0000000000000000 2 1 uu dau eeu 14 000 00000000000000000000 n A R 00 O lt n lt 10 000 14 10 11 00 8 080 00000000000 00000000000 15 000000 APMMSESTSA _ V rk P n 0 5 1F1 0 5 n 3 6 Ye PQl 1 0 Rx 15 D uuu 16 0000 0000000000 x103 800 e e LO CD KouenbaJJ 200 Amplitude 8 00000000000 ATR 0 500 Hz 2 kHz 1000000000000000 1 0 3 0 7 m OO O c O O O 0 0 9 00000000 00 1 4 Parametric MMSE STSA GM 000 O0O00000000000000000000000000000000000 OO00000000000000000000000000000 Parametric MMSE STSA 4 1 OO airport fh G
24. 006000 0 000000000000000000000000000000000000 Fig 18 000000000000000000000000000000000006000 0000000 2000000000000000000000000000000 0 0000000000000 3000000000000000000000000 0 000000000000000000000000000000000 5 7000 linit fixed 0 init mullti 0000000000000000 23 5 7 Parametric MMSE STSA MMSE STSA 10 Parmetric MMSE STSA 2 5 7 1 000 600000000000000 SNRU 5dB 0dB 5dB 1008 00000 OOOOO0000000000000000830S0000000000000000m 00550000000000003800000000045S000000000000 O0O0O000000083000000000000000000000 10000000 OO 1 000000000 2 MMSE STSA MMSE STSA 0 00000000000000 3 P MMSE STSA Parmetric MMSE STSA 2 000000000000 4 GM init multi 0000000000000000000 0 10 2 0 10 1 00 0 10 1 00 3 00 020000000000000002 000 0 10 1 00 0 00000000000000000 5 GM init fixed 000000000000000000 0000000000000000000 r 0 5 1 0 0 200000 6 CLEAN 0000000000000 L1 5 7 2 0600000000 1900 2100000 24 100 90 80 70 1 60 8 50 40 30 20 10 0 hall 5dB 008 5dB 10dB 5dB 008 5dB 10dB white airport clean 0 19 00000 000000000
25. 3 0000000 N 1 Xy an Wn 0 k N 2 n 0 1 N 1 E Wn n Xk e NT 0 n lt N 3 k 0 000000000000 00000000000 w 00000000 23 Yk Xk 5 000000 1 000000000 4 0000000000000 40000000 10000000 Y Xy Dk 4 OOO00000000006G 000000000 Yk Rei X Apel 5 D Nye 5 00000 00 0000 00 0000 20 00 0000 700000 0 00000000000000000000000000000000000000 23 000 00000000000000000000000000000000 0 000000000000000000000000000000000000 0 000000000000000000000000000000000000 000000000 200000000000000 3000000000000 31 00000000000000000000000000000000000 0 0000000000000 64B 0t J000000000000 0000000000000000000 2 4 117 400000000000000000000000 Short Term Fourier Transform 0000000000000000000000000 OOOO frame period 00000 frame length 0 50 00000000000 OverlapaddJ 000000000000000000000000 1 1 Fas Fourier Transform 112 1 20 9 40 80 100 0 100 200 300 400 500 600 Time ms Frequency Hz 2 0000 3 0000000000 000000 0 00000 0000000 0 00000 fra
26. aussian_ airport GM e 0 15 0 1 Probability Frequency Probability Frequency 0 5 Amplitude Amplitude 10 0000 Dk 500 Hz 2 kHz 1000000000 000000 0 0 00000 0 000 00000000 0 000 100000000 2 00000000000 25 000000 A 00000600 00000000000000000000000000 2 1 2 nai E E 000 r 0 lt r lt 1 0000002 0 lt z lt 1 ww gt 1 0000000000000 00000000000009 2 2 70000 0 0000000000000000000000000 p Y ape Aa k 18 18 00 14 00 800000000000000000 19 00000 OU E 10 00 0 _ fx z 0 5 1 r fi w 0 5 2 0 1 0 2 fel B I T G n Quen 97 em Ri OTR 20 19 lt Aq k a Sk Dk vk n S S ns Ek 21 1 A 1 0 lt Aa k 22 42 00000000000000000 0000000000 20000 1000000000 00000000000 0 0800000 POON 17 00000000 2 1 0 00 2200 10 9 8 gt P x Q z 10 23 00000000000000 9000000000 0000 00000000 0 0000000000000000000000000000000000000 90000000000000000000000000000 940000 0 24 000000000000000000000 g 0 ad 24 OOO 000000000 00000000000 60000000000 40 0 0000000000000000 4 2 1 300000000 0000000000 2 000000000000000000
27. ech Technology at KTH TMH KTH WaveSurfer http www speech kth se wavesurfer 0 O 2013 1 28 16 Scott W optimal and data based histograms Biometrika 66 3 605 610 1979 17 LS Gradshteyn and Ryzhik incorporating the 4th ed prepared by Yu V Geronimus M Yu Tseytlin translated from the Russian by Scripta Technica inc and edited by Alan Jeffrey Corr and enl ed prepared by Alan Jeffrey San Diego Tokyo Academic Press 1980 29 O00000000000000000000000 Al 00000000000 500000000000000000000000000 0 0000000000000 A 1 1 00000000 000 2000 43 0 Scott s Choice 16 00000000 _ 3 50 43 000 900000000000000 0000000000 A2 OD 0000000000000000000000000000000000000 300000 0 3 0000 white airp_j bus j rest j offi j shop j 1 3min 512points 32ms 256points 16ms 160 64 3 230
28. me length frame period 4 0000 MMSE STSA Minimum Mean Square Error Short Time Spectral Aamplitude 1 O00000000000000000000000000000000000000 0 0000000000000000 MMSE STSA 5 00 MMSE LSA 6 Order MMSE STSA 70 Laplacian MMSE STSA 8000000000000 0000000000 MMSE 5 0000000 10000 Parametric MMSE 2 9 000000000 31 MMSE STSA MMSE STSAQ O00 Ephraim 0 000000000000000 0 0000000000000000000 MMSE STSAQ 00000 J E Ag Ax Ye 6 00000000000000000 4 000 8 000000000 r E Ar Yi 7 SS akplak ok p Yk dok dag E Joe plak ou p Yy dak MMSE 5 00000000000000000000000000000 OO00000000000000000000000000000000000000 00000000000 00 200 000000000000 11 08000600 0 000000000000000000 19 8000000000000 267 Q2 5 9 1 P ara zq ET amp X 2 Da 9 00000000000 10000000 5000 10 0000000 6000 11 0000000000 00000 10000000 7000000000 00000000000000000 plak 0 0000000 0 U 000000 9 10 11 00 8 00000000000 00000000 12 000000 0 5 1 va 12 k Du 1 amp a Ri TRU Alk 0000 mep 0O0OO0O0O0000000000 amp 000 SNRDHDDUU vU SNRDDHUDUD 0000000
29. multi Bl3params init multi 2params init fixed init fixed 10dB 10dB 5dB 82 5dB white airport all Ju 5dB 5dB 000000000 15 00000 9dB 8 508 85 bus restaurant 16 00000 0000000000006 2 init multi 3params init multi El2params init fixed il Bl3params init fixed JH OdB 5dB 10dB 5dB 21 70 TE 60 lBl2params init multi s eed x M H2params init fixed 96 30 3params init fixed 20 10 o 5dB 0dB 10dB 5dB OdB 5dB 10dB office shop 17700000 0000000000000 5 6 3 OO0O0000000000 1 3 2params 200000000000r000 aw 00000000000 2 0000000000000 0000000 0 00 lt lt 10000000000000000 2 4 Bparams 0 0000000000 2 0000000000000 00000000000 001 2 0000000000000000 0000 3 4 init fixed 0000000000000000000000000 00000000000000000000000000 2000000000 3 4 init fixed 0000000000000000000000000000 0 00000000000000000000000000000000000000 OO0O00000000 22 0 35 4 2 0 2 4 Amplitude 0 18 0000000000000 00 00000000 02500000000 400000 0000000000000000000000000000000

観測信号の分布に混合ガウス分布を、音声信号の振幅の事前分布に

Contents

Download Pdf Manuals

Related Search

Related Contents

観測信号の分布に混合ガウス分布を、 音声信号の振幅の事前分布に

Contents

Download Pdf Manuals

Related Search

Related Contents

観測信号の分布に混合ガウス分布を、音声信号の振幅の事前分布に