Timeshrinking
From SpeechWiki
(Difference between revisions)
| Line 18: | Line 18: | ||
| 1 || 1 || [{{FisherPath}}/exp/triphone/test2kUttOnConvGaus.noUnits/config73/LATEST.log 51.6%] || || old baseline | | 1 || 1 || [{{FisherPath}}/exp/triphone/test2kUttOnConvGaus.noUnits/config73/LATEST.log 51.6%] || || old baseline | ||
|- | |- | ||
| - | | 1 || 1 || [{{FisherPath}}/exp/.../LATEST.log TR] || [{{FisherPath}}/exp/timeshrink/test/unit.tri.timeshrink.1.onSingleGaussian/LATEST.log | + | | 1 || 1 || [{{FisherPath}}/exp/.../LATEST.log TR] || [{{FisherPath}}/exp/timeshrink/test/triphoneSingleGausian/unit.tri.timeshrink.1.onSingleGaussian/LATEST.log 78.2%] || baseline rerun exactly as timeshrinking to really make sure it's not getting an unfair advantage |
|- | |- | ||
| .6 || .6 || 69.3% | | .6 || .6 || 69.3% | ||
Revision as of 18:11, 25 September 2009
Fisher experiments
| <math>\tau</math> | frames dropped |
|---|---|
| 1 | 0% |
| .9 | ~5% |
| .6 | ~35% |
| train <math>\tau</math> | test <math>\tau</math> | dev 2000 utt WER | dev 2000 utt on triphone single-gaussian model WER | comments |
|---|---|---|---|---|
| 1 | 1 | 51.6% | old baseline | |
| 1 | 1 | TR | 78.2% | baseline rerun exactly as timeshrinking to really make sure it's not getting an unfair advantage |
| .6 | .6 | 69.3% | ||
| .9 | .9 | 56.3% | 80.4% | |
| 1 | .9 | 53.9% | ||
| .9 | .9 | 57.2 | 80.7% | using the non-timeshrinking str file for test |
| .9 | 1 | 54.6 | ||
| 1 | 1 | TR | PLP+MLP tandem |
I have to check for bugs. It could be that the threshold is too low or it could be something else too. We should probably rerun baseline too, just to make sure I didn't optimize it unfairly.
Things to try
- Test svitchboard with fisher-trained model to see if we still get good results
- Train and test on plp+mlp, like svitchboard timeshrinking was done.
- Do baseline train+test to see if something changed in going from baseline to timeshrink structure files.
Future Directions
- Can be viewed as a two-mode special case of best-first viterbi search. So make a real best-first lattice search. Mark mentioned some attempts in the 80'ies to do this.
