- 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.).
Environment:
- HTK 3.4
- Cygwin NT-5.1 1.5.25
Section 1 is Data Preparation - 資料準備
Now, we can use the tool HSGen to generate the experimental sentences constructing by the lattice file.- 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)
For example,
$ HSGen -l -n 200 wdnet dict > testpromts
The HSGen will generate 200 entries of training sentences and save them in testpromts file by wdnet.
-------------------------------------------
$HSGen -l -n 140 wdnet /dict/dict1 > labels/trainprompts
-------------------------------------------
-------------------------------------------
$HSGen -l -n 15 wdnet /dict/dict1 > labels/testprompts
-------------------------------------------
So we can generate two files, trainprompts and testprompts, 140 entries and 15 entries respectively.
- trainprompts
S0001 DIAL EIGHT FIVE
S0002 DIAL ZERO ZERO EIGHT SIX OH ONE ZERO NINE THREE FIVE EIGHT FIVE THREE THREE NINE ZERO
S0003 DIAL ZERO SIX ZERO EIGHT THREE ZERO EIGHT SEVEN SEVEN THREE FIVE ONE TWO TWO FOUR NINE SIX
S0004 CALL DAVE WOOD
S0005 CALL STEVE YOUNG
S0006 DIAL NINE OH THREE NINE ONE ZERO ONE NINE NINE OH THREE THREE FOUR TWO FOUR SEVEN FOUR ZERO ZERO SIX ONE ZERO ONE FOUR ZERO NINE TWO
S0007 DIAL NINE FOUR TWO NINE OH
S0008 DIAL SEVEN EIGHT SIX ONE EIGHT ZERO NINE ZERO SIX SIX SIX ONE ONE THREE
S0009 PHONE DAVE WOOD
S0010 CALL TYLER
....
=======================
- testprompts
T0001 PHONE LAW
T0002 PHONE JULIAN TYLER
T0003 CALL JULIAN TYLER
T0004 CALL WOOD
T0005 PHONE LAW
T0006 PHONE STEVE YOUNG
T0007 PHONE STEVE YOUNG
T0008 DIAL FIVE FIVE TWO SIX SEVEN SIX EIGHT
T0009 PHONE PHIL LEE
T0010 DIAL TWO EIGHT FOUR SIX THREE ZERO OH OH EIGHT NINE FOUR NINE
T0011 DIAL NINE SIX OH NINE SIX THREE SIX EIGHT THREE OH
T0012 CALL DAVE WOOD
T0013 PHONE YOUNG
T0014 DIAL ONE OH OH ZERO EIGHT TWO OH
T0015 PHONE WOOD
=======================
Then use HSLab to record the wav files.
No comments:
Post a Comment