[$B]

$B#R#W#C#P(B $B
$B8&5f>R2p!'%^%$%/%m%[%s%"%l!<$rMQ$$$?(BHMM$B$K4p$E$/2;8;<1JL$H2;@

ATR$B2;@<8@8lDL?.8&5f=j(B $BBh0l8&5f<<(B $BCfB<(B $BE/!"@>1:(B $B7I?.(B
http://www.slt.atr.co.jp/dept1/index.html

$BK\%Z!<%8$N:G?7>pJs$O(B
http://www.slt.atr.co.jp/~tnishi/DB/micarray/research.htm
$B$r;2>H$N$3$H!#(B
$B#1!%35MW(B

RWCP$B$N(BDB$B$rv$_9~$s$G!"2;@<(I%$BHs2;@<$N<1JL$r;n$_$?!#$5$i$K!"%$%s%Q%k%91~Ez$H2;@<$N%I%i%$%=!<%9$rMQ$$$F2;@
$B#2!%2;8;<1JL%"%k%4%j%:%`(B

$B?^#1$N:8?^$N$h$&$K%^%$%/%m%[%s%"%l!<$KBP$7$F@5LLJ}8~$+$i2;@l9g$r9M$($k!#$3$3$G9b2;u67$K$*$$$F2;8;J}8~$r?dDj$9$k$3$H$,$G$-$F$b!"OC


$B?^#1(B $B2;8;<1JL%"%k%4%j%:%`(B

$B#3!%I>2A

$B?^#2$Kr7o$r<($9!#@5LLJ}8~$KL\(B $BE*2;8;!"B&LL$K;(2;8;$,B8:_$9$k>u67$rA[Dj$9$k!#$3$Nr7o2<$K$*$$$F%7(B $B%s%0%k%^%$%/%m%[%s$*$h$S%^%$%/%m%[%s%"%l!<$rMQ$$$F@5LLJ}8~$K%S!<%`%U%)!<(B $B%_%s%0$7$?$H$-$N!"3F(BSNR$B$K$*$1$k2;8;<1JL@-G=$rI>2A$7$?!#$^$?!"2;@<$KBP(B $B$7$F$O(BIPA(Imformation-technology Promotion Agency) 1998$BG/EY$N2;6A%b%G(B $B%k$rMQ$$$F2;@


$B?^#2!'
$BI=#1!'(B $Br7o(B
$B%^%$%/%m%[%s%"%l!<(B $BAG;R?t(B14, $BAG;R4V3V(B2.83 cm
Beamformer $BCY1dOB%"%l(B-
$B%5%s%W%j%s%0<~GH?t(B 16KHz
$B%U%l!<%`D9(B 32 msec.$B!J%O%_%s%0Ak!K(B
$BFCD'%Y%/%H%k(B MFCC, $B&$(BMFCC, $B&$%Q%o!<(B
$B<1JLMQ(BHMM Gaussian mixture$B7?(B
$B2;6A%b%G%k?t(B $B2;@<(B 1$B%b%G%k(B
$BHs2;@<(B 1$B%b%G%k(B
$B2;@<(BDB ATR$B2;@<%G!<%?%Y!<%9(BSetA
$B2;@<%b%G%k3X=,(B $BCK=w3F(B8$BL>(B 400$B8l(B
$BHs2;@<(BDB RWCP-DB
$BHs2;@<%b%G%k3X=,(B $B4D6-2;(B92$B
$B%F%9%H(B($B%*!<%W%s(B) $B2;@<(B: MHT$B2;1$%P%i%s%9(B216$B8l(B
$BHs2;@<(B: $B4D6-2;(B92$B
$B2;6AEAC#FC@-(B RWCP-DB
$B;D6A;~4V(B 0.0, 0.3, 1.3 sec
SNR -5, 0, 5, 10, 15, 20 ,25,30,clean dB
$B#3!%#1!%@-G=I>2A(B

$BK\2A$9$k!#$^$?2;@2A$r9T$&!#(B

$B#3!%#2!%

$B?^#3(B(a)$B$K%7%s%0%k%^%$%/%m%[%s!"(B(b)$B$K%^%$%/%m%[%s%"%l!<$rMQ$$$?>l9g(B $B$Ne$7$F$$$k!#$3$l(B $B$h$j%^%$%/%m%[%s%"%l!<$NM-8z@-$,3NG'$G$-$k!#$5$i$K!"2;8;<1JLN($O(BSNR$B$,(B 0dB$B$N7k2L$H(B20dB$B$N7k2L$O$[$H$s$IJQ$o$i$J$$$3$H$+$i!"(BSNR$B$,Dc$$4D6-2<$G$b(B $B9b@:EY$J2;8;<1JL$,2DG=$G$"$k$3$H$,$o$+$k!#$7$+$7!$(BSNR$B$,(B0dB$B$G$"$k>l9g$N(B $B2;@l9g$h$j$bBg$-$/Nt2=$7$F$$$k!#:#8e!"$h$j9b2;2A$7$?>l9g!"9b;D6A2<(B($B;D6A2DJQ<<(B [T60=1.3 sec.])$B$K$*$$$FDc(BSNR$B$N4D6-$G$b9b<1JL@-G=$G$"$k$3$H$+$i!"2;8;$N(B $B0LCV$,M=$a$o$+$C$F$$$k$J$i$P!"$=$N2;8;$,OC



(a) A single microphone

(b) A microphone array
$B?^(B3 $B2;8;<1JLN($H2;@

$B#4!%$^$H$a(B

$BK\8&5f$G$OOCl9g$K$D$$$F8!F$$9$kM=Dj$"$k!#(B

[$B]
RWCP Sound Scene Database in Real Acoustical Environments
Copyright (c) 1998-2001 Satoshi Nakamura, and Takanobu Nishiura, ATR Spoken Language Translation Research Laboratories.