语音识别之HTK入门(三)训练文件准备
本节记录的是进行训练前的各种文件的准备,步骤稍微有些繁杂。
config文件夹下的操作
config文件夹下运行终端命令如下:
touch grammer
touch global.ded
touch mkphones_color.led
touch codetrain.scp
touch config_HCopy
touch proto
touch train.scp
touch config_color
grammer文件内容:
$color = RED | BLUE | YELLOW;
(SENT-START ($color) SENT-END)
global.ded文件内容:
AS sp
RS cmu
MP sil sil sp
mkphones_color.led文件内容
EX
IS sil sil
DE sp
codetrain.scp文件内容
data/train/speech/s01.wav data/train/mfc/s01.mfc
data/train/speech/s02.wav data/train/mfc/s02.mfc
data/train/speech/s03.wav data/train/mfc/s03.mfc
data/train/speech/s04.wav data/train/mfc/s04.mfc
data/train/speech/s05.wav data/train/mfc/s05.mfc
data/train/speech/s06.wav data/train/mfc/s06.mfc
data/train/speech/s07.wav data/train/mfc/s07.mfc
data/train/speech/s08.wav data/train/mfc/s08.mfc
data/train/speech/s09.wav data/train/mfc/s09.mfc
data/train/speech/s10.wav data/train/mfc/s10.mfc
config_HCopy文件内容
# Coding parameters
TARGETKIND = MFCC_0_D_A
TARGETRATE = 100000.0
SAVECOMPRESSED = T
SAVEWITHCRC = T
WINDOWSIZE = 250000.0
USEHAMMING = T
PREEMCOEF = 0.97
NUMCHANS = 26
CEPLIFTER = 22
NUMCEPS = 12
ENORMALISE = F
SOURCEFORMAT = WAV
proto文件内容
~o <VecSize> 39 <MFCC_0_D_A>
~h "proto"
<BeginHMM>
<NumStates> 5
<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
<TransP> 5
0.0 1.0 0.0 0.0 0.0
0.0 0.6 0.4 0.0 0.0
0.0 0.0 0.6 0.4 0.0
0.0 0.0 0.0 0.7 0.3
0.0 0.0 0.0 0.0 0.0
<EndHMM>
train.scp文件内容
data/train/mfc/s01.mfc
data/train/mfc/s02.mfc
data/train/mfc/s03.mfc
data/train/mfc/s04.mfc
data/train/mfc/s05.mfc
data/train/mfc/s06.mfc
data/train/mfc/s07.mfc
data/train/mfc/s08.mfc
data/train/mfc/s09.mfc
data/train/mfc/s10.mfc
config_color文件内容
# Coding parameters
TARGETKIND = MFCC_0_D_A
TARGETRATE = 100000.0
SAVECOMPRESSED = T
SAVEWITHCRC = T
WINDOWSIZE = 250000.0
USEHAMMING = T
PREEMCOEF = 0.97
NUMCHANS = 26
CEPLIFTER = 22
NUMCEPS = 12
ENORMALISE = F
lists文件夹下的操作
lists文件夹下运行终端命令如下:
touch wordlist
wordlist文件内容
BLUE
RED
SENT-END
SENT-START
YELLOW
dict文件夹操作
将字典文件的“beep”和“otherDict"置于dict下
labels文件夹操作
labels文件夹下运行终端命令如下:
touch trainprompts
trainprompts文件内容:
s01 SENT-START BLUE SENT-END
s02 SENT-START RED SENT-END
s03 SENT-START BLUE SENT-END
s04 SENT-START BLUE SENT-END
s05 SENT-START BLUE SENT-END
s06 SENT-START YELLOW SENT-END
s07 SENT-START RED SENT-END
s08 SENT-START BLUE SENT-END
s09 SENT-START BLUE SENT-END
s10 SENT-START RED SENT-END
scripts文件夹操作
脚本文件包括以下文件:
主目录树下的操作
注意注意,进行以下操作时需要将下载好的htk工具箱里的部分文件复制进主目录下,具体如图:
依次运行以下命令:
HParse ./config/grammer ./config/wordnet
config目录会生成wordnet;
HDMan -m -w ./lists/wordlist -n ./lists/monophones -g ./config/global.ded ./dict/dict_color ./dict/beep ./dict/otherDict
ists目录会生成monophones;
dict目录里会生成dict_color;
perl ./scripts/prompts2mlf ./labels/trainwords.mlf ./labels/trainprompts
labels目录会生成trainwords.mlf;
HLEd -l '*' -d ./dict/dict_color -i ./labels/phones_color.mlf ./config/mkphones_color.led ./labels/trainwords.mlf
labels目录会生成phones_color.mlf;
HCopy -T 1 -C ./config/config_HCopy -S ./config/codetrain.scp
data/train/mfc目录会生成s01.mfc;s02.mfc……s10.mfc;
好的,到此已经对训练文件准备完成,下一节写训练模型。
来源:CSDN
作者:小老弟哟
链接:https://blog.csdn.net/weixin_44747240/article/details/104157687