フリーソフトでつくる音声認識システム ( 第 2 版 ) サンプルページ この本の定価 判型などは, 以下の URL からご覧いただけます. http://www.morikita.co.jp/books/mid/084712 このサンプルページの内容は, 第 2 版 1 刷発行時のものです.
i 2007 10 1 Scilab 2 2017 2 1 2 1
ii 2 web 2007 9
iii 1 1 1 3 1.1 3 1.2 4 1.3 5 1.4 6 1.5 7 1.5.1 7 1.5.2 8 12 2 13 2.1 13 2.1.1 13 2.1.2 14 2.2 17 2.2.1 17 2.2.2 20 2.3 21 2.3.1 21 2.3.2 22 24 3 25 3.1 25 3.1.1 25 3.1.2 28 3.2 30 3.3 34 3.3.1 34 3.3.2 37 41
iv 4 42 4.1 NN 42 4.1.1 42 4.1.2 43 4.1.3 45 4.2 46 4.2.1 47 4.2.2 48 4.2.3 2 49 4.2.4 52 4.3 k-nn 55 4.3.1 55 4.3.2 56 4.3.3 57 4.3.4 k-nn 58 60 5 61 5.1 62 5.2 63 5.3 65 5.3.1 65 5.3.2 Widrow Hoff 66 5.3.3 68 5.4 69 5.4.1 69 5.4.2 70 71 6 1 72 6.1 72 6.2 73 6.2.1 73 6.2.2 74 6.3 80 6.3.1 80 6.3.2 81 6.3.3 82 85
v 7 2 86 7.1 86 7.2 88 7.2.1 89 7.2.2 89 7.2.3 90 7.2.4 95 7.3 95 7.3.1 96 7.3.2 97 7.3.3 99 102 8 103 8.1 103 8.1.1 103 8.1.2 104 8.1.3 105 8.1.4 p(x) 106 8.1.5 P (ω i ) 106 8.1.6 p(x ω i ) 107 8.2 107 8.2.1 108 8.2.2 109 8.2.3 110 8.3 114 8.3.1 114 8.3.2 114 8.3.3 115 116 9 117 9.1 117 9.1.1 117 9.1.2 118 9.2 120 9.2.1 121 9.2.2 121 9.2.3 124 128
vi 2 129 10 131 10.1 131 10.2 133 10.3 137 10.3.1 HMM 138 10.3.2 139 10.3.3 140 10.4 142 10.4.1 142 10.4.2 143 10.4.3 Baum Welch 144 10.5 146 10.5.1 146 10.5.2 147 10.5.3 149 149 11 HTK 151 11.1 HTK 151 11.2 153 11.3 155 11.4 156 11.5 158 11.6 HMM 160 11.7 161 11.8 164 165 12 166 12.1 166 12.2 167 12.3 169 12.3.1 169 12.3.2 170 12.4 Julius 171 12.5 176 179
vii 13 180 13.1 180 13.2 N- 181 13.2.1 N- 182 13.2.2 183 13.2.3 183 13.3 184 13.3.1 184 13.3.2 185 13.3.3 Good Turing 185 13.4 186 13.4.1 187 13.4.2 187 13.5 189 13.6 SRILM 190 193 14 194 14.1 194 14.1.1 194 14.1.2 195 14.2 196 14.2.1 196 14.2.2 197 14.3 WFST 198 14.4 199 14.5 201 14.6 203 14.6.1 203 14.6.2 204 14.6.3 204 206 15 207 15.1 207 15.2 208 15.3 211 15.3.1 MMDAgent 211 15.3.2 MMDAgent 212 15.3.3 213 15.3.4 214 216
viii 217 A 230 A.1 230 A.2 231 A.3 233 A.4 234 A.4.1 234 A.4.2 235 A.5 236 B Scilab 238 B.1 238 B.2 239 B.2.1 240 B.2.2 241 B.3 241 B.3.1 241 B.3.2 241 B.3.3 241 B.4 242 B.5 243 B.6 243 B.7 243 C Weka 244 C.1 DL4J 244 C.2 DL4J 245 C.3 DL4J 246 D 248 250 251 252
3 1 1.1 pattern recognition (pattern) recognition re cognition 1970 1.1(a) 1.1(b)
4 1 1.1 1.1(c) 1.2 1.2 1.2 SegNet http://mi.eng.cam.ac.uk/projects/ segnet/
1.3 5 1.3 2 1.3 1.3 2
6 1 1.4 1 x =(x 1,x 2,...,x d ) T (1.1) d d 2 d 1 2 T x
1.5 7 1.4 x 1 1.4 3 1.5 1 1.5.1 x (1.1) d c ω 1,ω 2,...,ω c
8 1 1.5 d p 1, p 2,...,p c 1.5 x x p 1, p 2,...,p c p i ω i 1.6 nearest neighbor NN NN 4 1.6 1.5.2
42 4 4.1 4.1 1 NN NN NN 4.1 NN NN NN 4.1.1 c ω 1,ω 2,...,ω c d ω i p i
4.1 NN 43 p i =(p i1,p i2,...,p id ) T (i =1,...,c) (4.1) x x =(x 1,x 2,...,x d ) T (4.2) x p i D(x, p i ) D(x, p i )= (x 1 p i1 ) 2 +(x 2 p i2 ) 2 + +(x d p id ) 2 (4.3) 8 min minimum min D(x, p i ) (i =1,...,c) (4.4) arg min D(x, p i ) (i =1,...,c) (4.5) i arg min argument minimum argument f(x) x C arg min i arg min min arg min 4.1.2 (4.5) 2
44 4 2 x x ω 1 p 1 ω 2 p 2 p 1 p 2 x 2 2 2 4.2 4.2 2 4.2 NN x NN 4.2 4.1 ω 1 p 1 =(2, 8) ω 2 p 2 =(8, 4) x =(1, 6) NN x p 1, p 2 D D(x, p 1 )= (1 2) 2 +(6 8) 2 = 5 D(x, p 2)= (1 8) 2 +(6 4) 2 = 53 D(x, p 1 ) <D(x, p 2 ) x ω 1
4.1 NN 45 (5,6) p 1 p 2 3x 1 2x 2 =3 4.3 4.3 4.1.3 NN ω 1 4.4 p 1 ω 2 p 2 4.4 4.5 4.5
46 4 4.4 4.5 ω 1 ω 2 ω 2 ω 1 2 4 4.2 4.6 4.6
10.2 133 (10.2) w 10.1 11 12 13 14 1 10.2 (10.2) p(x 1,...,x t w 1,...,w n ) w x 8 p(x w) p(x 1,...,x t w) /a/ A 2 1 3 3 1 1 [10] [11] 2 /a/ A 3
134 10 /a/ A (/a/ /s/ /u/) AAASSSUUU 1 0 10.2 1 10.2 3 /a/ /s/ /u/ 3 9 2 A 10.2 AAASSSUUU 1 0 1 2 11
10.2 135 P (AAASSSUUU ) 1 P (AAASSUUU ) P (AAASSSUUE ) 0 2 1 2 /a/ 0.8 A 0.1 I 0.1 O 10.3 10.3 10.3 AAASSSUUU P (AAASSSUUU ) =0.8 0.8 0.8 0.7 0.7 0.7 0.7 0.7 0.7 =6.0236288 10 2 AAISSSUEU P (AAISSSUEU ) =0.8 0.8 0.1 0.7 0.7 0.7 0.7 0.1 0.7 =1.075648 10 3
136 10 3 3 2 1 /a/ 4 10.4 S 1 3 S 2 3 1 0.75 0.25 10.4 AAISSTSSUE P (AAISSTSSUE ) = P (S B S 1 ) P (A S 1 ) P (S 1 S 1 ) P (A S 1 ) P (E S 3 ) P (S 3 S E ) =1.0 0.8 0.75 0.8 0.1 0.1 =1.9611853 10 7 4
10.3 137 10.4 S 1 A, I, O 0 0 10.5 10.5 AAISSTSSUE A 10.5 S 1 2 A S 1 S 1 S 1 S 2 S 2 hidden Markov model; HMM HMM 10.3 HMM
138 10 10.3.1 HMM HMM 10.1 10.6 HMM AAB 10.6 10.1 HMM AAB HMM A S 1 B S 2 3 2 S B S 1 S 1 S 2 S E S B S 1 S 2 S 2 S E P (S B S 1 ) P (A S 1 ) P (S 1 S 1 ) P (A S 1 ) P (S 1 S 2 ) P (B S 2 ) P (S 2 S E ) =1.0 0.6 0.8 0.6 0.2 0.5 0.5 =0.0144 P (S B S 1 ) P (A S 1 ) P (S 1 S 2 ) P (A S 2 ) P (S 2 S 2 ) P (B S 2 ) P (S 2 S E ) =1.0 0.6 0.2 0.5 0.5 0.5 0.5 =0.0075 0.0144 + 0.0075 = 0.0219 HMM S B S E
252 A* 197 Baum Welch 144, 160 BNF 169 EM 116 forward-backward 146 forward 140 Good Turing 185 HMM 137 HMM 152 Julius 171 Kneser Ney 188 k-nn 58 LSTM 102 MAP 104 MFCC 27 MMDAgent 211 Modified Kneser Ney 188 NN 8 N- 182 ReLU 98 SIFT 28 SRGS 176 SRILM 190 VoiceXML 177 WFST 199 Widrow Hoff 66 Witten Bell 188 110 24 97 134, 171 110 48, 57 50 59 49 132 25 52 82 95, 125 47 52 9 135 69, 90 137 107 137 185 87 81 82 21 170 231 121 62 26 233 93, 158 55 56 3 122 9 122 122 107
253 126 132 118 91 96 89 180 115 65 8 134 63 194 110, 182 196 185 74 74 69 87 47 7 7 9, 44 72 92 104 104 136 97 107 86 168 17 38 135 88 235 100 138 21 134 144 59 18 22 19 9 108 170 82 183 136 44 187 208 235 88 110 82 23 99 24, 99 195 168 203 203 133, 194 114 76 207 88 149 102 180 17, 95 183 181 166, 180 132 6 6 7 170 149 209
254 139 98 88 87 86 110 125 21 49 52 52 3 32 3 6 188 188 68 124 110 137 168 89 80 141 119 195 198 196 196 31 232 14 14 15 87 88 23 26 56 57 230 230 231 99 7 117 231 166 166 169 231 23 123 115 105 59 186 170 5 74 197 69 172 23 20 149 110 195 75 75 75, 234 101, 189 14 14 136
1993 1997 1998 1999 2007 2 2007 10 10 1 1 2016 11 22 1 11 2017 4 11 2 1 2017 1 4 11 102 0071 03 3265 8341 FAX 03 3264 8709 http://www.morikita.co.jp/ Printed in Japan ISBN978 4 627 84712 5