Log:Kaldi 863
来自cslt Wiki
initial
- basic info
total data range: disc01 ~ disc 06
about 5% speech data for dev set, 65% data for train set
- dev set
| type | best WER% |
|---|---|
| monophone | 21.55 |
| triphone | 12.38 |
| LDA+MLLT | 10.91 |
| +MMI | 11.13 |
| +boosting | 10.62 |
| +fMMI | 18.33 |
expanded train set
about 5% speech data for dev set, 85% data for train set
- dev set
| type | best WER% |
|---|---|
| monophone | 18.57 |
| triphone | 10.44 |
| LDA+MLLT | 8.45 |
| +MMI | 7.14 |
| +boosting | 7.02 |
| +fMMI | 8.51 |
expanded again
data range: disc01 ~ disc 11
about 5% speech data for dev set, 85% data for train set
- dev set
| type | best WER% |
|---|---|
| monophone | 10.19 |
| triphone | 3.97 |
| LDA+MLLT | 3.26 |
| +MMI | 2.93 |
| +boosting | 2.75 |
| +fMMI | 4.92 |
with tone
- dev set
| type | best WER% |
|---|---|
| monophone | 8.39 |
| triphone | 3.36 |
| LDA+MLLT | 3.11 |
| +MMI | 2.48 |
| +boosting | 2.38 |
| +fMMI | 4.37 |
