Ticket #105 (closed defect: fixed)
Check latest Acoustic Model build 798 - 8kHz:16bit model not recognizing as well as it should
Reported by: | kmaclean | Owned by: | kmaclean |
---|---|---|---|
Priority: | major | Milestone: | Acoustic Model 0.1.2 |
Component: | Acoustic Model | Version: | 0.1-alpha |
Keywords: | Cc: |
Description
Problem might be related to Audacity's noise removal algorithm.
- the 16kHz:16bit (build 798) Acoustic Model seems to work OK with audio that has been modified with Audacity's noise removal algorithm.
- the 8kHz:16 bit (build 798) Acoustic Model has terrible recognition ... need further testing to isolate the problem.
Need to compare and test performance against the current release build 726.
Change History
Note: See
TracTickets for help on using
tickets.
Ran Acoustic Model creation script only on audio that had *no* noise reduction.
8kHz:16bit Acoustic Model performed much better on manual voice tests.
Need some more experimentation with noise reduction audio to understand impacts of noise reduction on acoustic model creation.