Find topic
WS06 topics
- WS06 home
- Members
- Team papers
- Team presentations
- Feature transcriptions
- Pre-workshop meetings
- Additional papers
- Feature sets
- Data
- Links
Members' area
- Papers/final report
- Member journals
- Meeting notes
- Ongoing work
- Results
- Structures
- Tools & code
- Compute/space
- Project ideas
- Final presentation
Tools
Help!
Same as MonoFeat1State but with the following differences
fine grid
ckbeam=3000
- trained on missisipi state forced alignments (runs around 5 times faster)
- No beam for training (runs fast enough not to need a beam)
- wider beam for testing (ckbeam 3000)
| WER | comp | lm_sc | lm_pen |
| 73.3 | 256 | 4 | -4 |
| 73.5 | 512 | 9 | 0 |
| 76.3 | 128 | 9 | 0 |
| acc | mix | lm_sc | lm_pen |
| 26.9 | 256mixFine | 6 | -2 |
| 26.7 | 256mixFine | 4 | -4 |
| 26.7 | 256mixFine | 7 | -1 |
| 26.6 | 256mixFine | 4 | -2 |
| 26.3 | 256mixFine | 6 | -3 |
| 26.3 | 256mixFine | 5 | -3 |
| 26.3 | 256mixFine | 5 | -2 |
| 26.3 | 256mixFine | 4 | -3 |
| 26.0 | 256mixFine | 5 | -4 |
| 26.0 | 256mixFine | 8 | -1 |
| 25.8 | 256mixFine | 6 | -4 |
| 25.6 | 256mixFine | 3 | -5 |
| 25.5 | 256mixFine | 5 | -5 |
| 25.5 | 256mixFine | 4 | -5 |
| 25.1 | 256mixFine | 3 | -4 |
| 24.6 | 256mixFine | 3 | -3 |
| 24.5 | 256mixFine | 4 | -6 |
| 24.4 | 256mixFine | 6 | -5 |
| 24.3 | 256mixFine | 5 | -6 |
| 24.2 | 256mixFine | 3 | -2 |
| 23.6 | 256mixFine | 6 | -6 |
| 22.3 | 256mixFine | 2 | -6 |
| 21.6 | 256mixFine | 2 | -5 |
| 20.8 | 256mixFine | 2 | -4 |
| 18.6 | 256mixFine | 2 | -3 |
| -8.0 | 256mixFine | 3 | -6 |
| 32.2 | 256mixFine | 2 | -2 |
- tuning complete lms/lmp is 6/-2, 256 gausians
- tight training complete
- final test WER 74.8