- HTKBooks,
- 苏统华, 哈尔滨工业大学人工智能研究室, 2006年10月30日,
- Howard Hung-Ju Chou, Intelligence Information Retrieval Lab., NCKU, Taiwan(R.O.C.).
- HTK 3.4
- Cygwin NT-5.1 1.5.25
- Step 1 - the Task Grammer 辨識模型要用的文法資料(gram->wdnet)
- Step 2 - the Dictionary 辨識模型要用的字典資料 (利用字典將wlist裡的單字翻譯為phones字串, one -> w ah n)
- Step 3 - Recording the Data 錄製辨識用的語音檔 (產生劇本並依照劇本錄製*.wav)
- Step 4 - Creating the Transcription Files 建立翻譯檔 (根據字典翻譯實驗資料
trainprompts
testprompts
First of all, we have to convert the two prompts files to be words files. That means
====================
S0002 DIAL ZERO ZERO EIGHT SIX OH ONE ZERO NINE THREE FIVE EIGHT FIVE THREE THREE NINE ZERO
====================
will be convert to
====================
"S0002.lab"
DIAL
ZERO
ZERO
EIGHT
SIX
OH
ONE
ZERO
NINE
THREE
FIVE
EIGHT
FIVE
THREE
THREE
NINE
ZERO
.
====================
Prompts level converts to word leavel.
How to read MLF(Master Label File), you can refer Section 6.3 in HTKbook.
How to read MLF(Master Label File), you can refer Section 6.3 in HTKbook.
Command is below,
------------------------------------------------------------------------------
$ perl scripts/prompts2mlf labels/trainwords.mlf labels/trainprompts
------------------------------------------------------------------------------
For testwords.mlf,
------------------------------------------------------------------------------
$ perl scripts/prompts2mlf labels/testwords.mlf labels/testprompts
------------------------------------------------------------------------------
Then use HLEd (Label Editor) to convert word level to phone level.
--------------------------------------------------
$ HLEd -l '*' -d dict/dict1 -i labels/phones0.mlf mkphones0.led labels/trainwords.mlf
--------------------------------------------------
mkphones0.led is edit script filr to tell HLEd how to generate labels/phones0.mlf from labels/trainwords.mlf.
==============
EX
IS sil sil
DE sp
==============
- EX, means replace all the words in labels/trainwords.mlf to phones level.
- IS sil sil, means insert sil to the start and the end of utterance.
- DE sp, means that deleting all short-pause.
HLEd will refer the dict/dict1 to transcript each word to its phones.
Then labels/trainwords.mlf will be replaced to labels/phones0.mlf.
Like,
===============
"*/S0001.lab"
DIAL
EIGHT
FIVE
.
===============
replaced to
===============
"*/S0001.lab"
sil
d
ay
ax
l
ey
t
f
ay
v
sil
.
===============
sil is inserted to the start and the end. DIAL is replaced to d and ay.
No comments:
Post a Comment