Timeshrinking
From SpeechWiki
(Difference between revisions)
| Line 2: | Line 2: | ||
{| class="wikitable" | {| class="wikitable" | ||
| - | ! train <math>\tau</math> !! test <math>\tau</math> !! dev 2k utt WER !! frames dropped | + | ! train <math>\tau</math> !! test <math>\tau</math> !! dev 2k utt WER !! frames dropped !! comments |
|- | |- | ||
| 1 (old baseline)|| 1 || [{{SERVER}}/speech/akantor/fisher/exp/triphone/test2kUttOnConvGaus.noUnits/config73/LATEST.log 51.6%] || 0% | | 1 (old baseline)|| 1 || [{{SERVER}}/speech/akantor/fisher/exp/triphone/test2kUttOnConvGaus.noUnits/config73/LATEST.log 51.6%] || 0% | ||
| Line 12: | Line 12: | ||
| 1 || .9 || [{{SERVER}}/speech/akantor/fisher/exp/timeshrink/testOnBaseline/unit.tri.timeshrink.point9/LATEST.log 53.9%]|| ~5% | | 1 || .9 || [{{SERVER}}/speech/akantor/fisher/exp/timeshrink/testOnBaseline/unit.tri.timeshrink.point9/LATEST.log 53.9%]|| ~5% | ||
|- | |- | ||
| - | | .9 || 1 || TE || ~5% | + | | .9 || .9 || [{{SERVER}}/speech/akantor/fisher/exp/timeshrink/test/unit.tri.timeshrink.point9.noTsStr/LATEST.log 57.2] || ~5% || using the non-timeshrinking str file for test |
| + | |- | ||
| + | | .9 || 1 || [{{SERVER}}/speech/akantor/fisher/exp/timeshrink/.../LATEST.log TE] || ~5% | ||
|} | |} | ||
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. | 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== | ==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. | * 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. | ||
[[Category:Fisher Experiments]] | [[Category:Fisher Experiments]] | ||
Revision as of 20:27, 17 September 2009
Fisher experiments
| train <math>\tau</math> | test <math>\tau</math> | dev 2k utt WER | frames dropped | comments |
|---|---|---|---|---|
| 1 (old baseline) | 1 | 51.6% | 0% | |
| .6 | .6 | 69.3% | ~35% | |
| .9 | .9 | 56.3% | ~5% | |
| 1 | .9 | 53.9% | ~5% | |
| .9 | .9 | 57.2 | ~5% | using the non-timeshrinking str file for test |
| .9 | 1 | TE | ~5% |
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.
