Recent issues

Heh, this month I discovered few critical issues in CMUSphinx.
  1. Pocketsphinx doesn't properly decode short silences in FSG/JSGF mode
  2. Sphinx4 doesn't really work with OOV loop in grammar
  3. Pocketsphinx n-best lists are useless because of too many repeated entries
  4. Pocketsphinx accuracy is way lower than sphinx3 one
  5. Supposedly-working sphinxbase LM stuff doesn't work with 32-bit DMP, thus no MMIE training for very large vocabulary
  6. MMIE itself doesn't improve accuracy (tested on Voxforge and Fisher)
  7. It's impossible to extract mixture_weights from recent sendumps in pocketsphinx models, python scripts in SphinxTrain are outdated
  8. PTM model adaptation doesn't work
  9. TextAligner demo from sphinx4 requires way more work to align properly

That's getting crazy, I wonder if I'll be able to find the time to fix all that.

11 comments:

Post a Comment