# 專題研究 (3) Viterbi Decoding Triphone Acoustic Model

## 專題研究 (3) Viterbi Decoding Triphone Acoustic Model

- - - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - - -
##### Presentation Transcript

1. 專題研究 (3)Viterbi DecodingTriphone Acoustic Model Prof. Lin-Shan Lee, TA. Yun-Chiao Li

2. Viterbi Decoding 03.04.mono0a.viterbi.sh 04.04.tri1.viterbi.sh

3. Viterbi Decoding • Instead of using WFST, we use Viterbi now • Converted Kaldi Acoustic model to HTK by Vulcan • (02.02.convert.htk.feat.sh) Convert the acoustic model from Kaldi to HTK

4. Viterbi Decoding Using the dev set to find the best acoustic weight (acwt)

5. Triphone Acoustic Model 04.01~04.04

6. Triphone Acoustic Model • In monophone acoustic model, • ㄅ、ㄆ、ㄇ they use their own model • In triphone acoustic model, • ㄅ-ㄆ-ㄇ is a model • There will be too many model and lack of training data

7. Decision Tree • Use decision tree to tie similar models together

8. 04.01.tri1.train.sh (1/3) • It is very similar to 03.01

9. 04.01.tri1.train.sh (2/3)

10. 04.01.tri1.train.sh (3/3)

11. Homework bash 04.01.tri1.train.sh bash 04.02.tri1.mkgraph.sh bash 04.03.tri1.fst.sh bash 04.04.tri1.viterbi.sh

12. Some Helpful References • “使用加權有限狀態轉換器的基於混合詞與次詞 以文字及語音指令偵測口語詞彙” – 第三章 • https://www.dropbox.com/s/dsaqh6xa9dp3dzw/wfst_thesis.pdf • Check HDecode, HLRescore in HTK Book