HTK yes/no识别

yes/no识别

预先建立yn文件夹

第一步：创建训练文件

DOS下打开yn文件夹，命令：HSLab yes.sig 创建10个yes音，10个no音,10个sil（静音）。保存在yn/sig下。

 

第二步：声学分析

抽取yes和no的mfcc特征参数。保存在yn/mfcc下。

命令：HCopy –C extract.conf  –S iofile.txt

其中  extract.conf（在yn文件夹下）为抽取参数配置文件，

内容为：

#

# Example of an acoustical analysis configuration file

#

SOURCEFORMAT = HTK # Gives the format of the speech files

TARGETKIND = MFCC_D_A_0 # Identifier of the coefficients to use

# Unit = 0.1 micro-second :

WINDOWSIZE = 250000.0 # = 25 ms = length of a time frame

TARGETRATE = 100000.0 # = 10 ms = frame periodicity

NUMCEPS = 12 # Number of MFCC coeffs (here from c1 to c12)

USEHAMMING = T # Use of Hamming function for windowing frames

PREEMCOEF = 0.97 # Pre-emphasis coefficient

NUMCHANS = 26 # Number of filterbank channels

CEPLIFTER = 22 # Length of cepstral liftering

# The End

 

iofile.txt（在yn文件夹下）说明抽取源文件路径和目标文件的保存路径，

内容如下：

sig/yes0.sig mfcc/yes0.mfcc

sig/yes1.sig mfcc/yes1.mfcc

sig/yes2.sig mfcc/yes2.mfcc

sig/yes3.sig mfcc/yes3.mfcc

sig/yes4.sig mfcc/yes4.mfcc

sig/yes5.sig mfcc/yes5.mfcc

sig/yes6.sig mfcc/yes6.mfcc

sig/yes7.sig mfcc/yes7.mfcc

sig/yes8.sig mfcc/yes8.mfcc

sig/yes9.sig mfcc/yes9.mfcc

sig/no0.sig mfcc/no0.mfcc

sig/no1.sig mfcc/no1.mfcc

sig/no2.sig mfcc/no2.mfcc

sig/no3.sig mfcc/no3.mfcc

sig/no4.sig mfcc/no4.mfcc

sig/no5.sig mfcc/no5.mfcc

sig/no6.sig mfcc/no6.mfcc

sig/no7.sig mfcc/no7.mfcc

sig/no8.sig mfcc/no8.mfcc

sig/no9.sig mfcc/no9.mfcc

sig/sil0.sig mfcc/sil0.mfcc

sig/sil1.sig mfcc/sil1.mfcc

sig/sil2.sig mfcc/sil2.mfcc

sig/sil3.sig mfcc/sil3.mfcc

sig/sil4.sig mfcc/sil4.mfcc

sig/sil5.sig mfcc/sil5.mfcc

sig/sil6.sig mfcc/sil6.mfcc

sig/sil7.sig mfcc/sil7.mfcc

sig/sil8.sig mfcc/sil8.mfcc

sig/sil9.sig mfcc/sil9.mfcc

 

第三步：HMM原型定义

建立文件hmm_yes.hmm、hmm_no.hmm、hmm_sil.hmm保存在yn/model/proto下。

hmm_yes.hmm内容如下：

~o <VecSize> 39 <MFCC_D_A_0>

~h "yes"

<BeginHMM>

<NumStates> 6

<State> 2

<Mean> 39

0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0

0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0

0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0

<Variance> 39

1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0

1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0

1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0

<State> 3

<Mean> 39

0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0

0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0

0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0

<Variance> 39

1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0

1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0

1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0

<State> 4

<Mean> 39

0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0

0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0

0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0

<Variance> 39

1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0

1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0

1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0

<State> 5

<Mean> 39

0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0

0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0

0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0

<Variance> 39

1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0

1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0

1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0 1.0

<TransP> 6

0.0 0.5 0.5 0.0 0.0 0.0

0.0 0.4 0.3 0.3 0.0 0.0

0.0 0.0 0.4 0.3 0.3 0.0

0.0 0.0 0.0 0.4 0.3 0.3

0.0 0.0 0.0 0.0 0.5 0.5

0.0 0.0 0.0 0.0 0.0 0.0

<EndHMM>

 

hmm_no.hmm、hmm_sil.hmm有着相同的内容

 

第四步：HMM训练

1.初始化：(建立文件夹yn/train/init)

 命令: HInit -S trainlist_yes -M train -H model/proto/hmm_yes.hmm yes

trainlist_yes（在yn文件夹下）内容为：

mfcc/yes0.mfcc

mfcc/yes1.mfcc

mfcc/yes2.mfcc

mfcc/yes3.mfcc

mfcc/yes4.mfcc

mfcc/yes5.mfcc

mfcc/yes6.mfcc

mfcc/yes7.mfcc

mfcc/yes8.mfcc

mfcc/yes9.mfcc

no、sil类似地初始化。

 2．训练

 命令：HRest -H train/init/hmm_yes.hmm -S trainlist_yes -M train yes

 no、sil类似地训练。

 

第五步：任务定义

1．          建立语法规则和字典

建立语法规则文件gram.txt（在yn文件夹下），内容为：

/*

* Task grammar

*/

$WORD = YES | NO;

( { SIL } [ $WORD ] { SIL } )

 

建立字典文件dict.txt（在yn文件夹下），内容为：

YES [yes] yes

NO [no] no

SIL [sil] sil

 

2．          建立任务网络

命令：HParse gram.txt net.slf

      HSGen -s net.slf dict.txt

 

第六步：识别

1．  建立文件hmmlist.txt（在yn文件夹下）文件，内容为：

yes

no

sil

注意每一列最后留一个空格。

 

2．          创建待识别信号并抽取mfcc参数（文件为test.mfcc）保存在mfcc文件夹下。

 

3．          命令：HVite -H train/hmm_yes.hmm -H train/hmm_no.hmm -H train/hmm_sil -i reco.mlf -w net.slf dict.txt hmmlist.txt mfcc/test.mfcc

识别结果自动存放在文件reco.mlf。