search

alumae / ruby-pocketsphinx-server

Ruby-based web service for speech recognition, using the PocketSphinx gstreamer module
Other
91 stars 19 forks source link

unpublished pocketsphinx patch dependencies #5

Closed tkharris closed 12 years ago

tkharris commented 12 years ago

It looks like, and I think it's mentioned in the README, that there are dependencies on unpublished pocketsphinx modifications. What are the modifications, and why are they not public? Can a patch be made available?

alumae commented 12 years ago

The dependencies are not yet public because I'm struggling to make them play nice with the current examples (particularly, I'm using 16 kHz sample rate while the current SVN version of pocketsphinx GST plugin is fixed to 8 kHz). I hope to work on the patch in April.

tkharris commented 12 years ago

Understood. Thanks alumae.

I hacked versions of pocketsphinx and ruby-pocketsphinx-server this weekend to get them to talk to each other, but it was at 8kHz. Getting it to work at 16kHz looked like it would have been easy, but, as you mentioned, not without breaking the 8kHz. It seems to be hard-coded in about 3 different places. I look forward to your 16kHz version that plays nicely with 8kHz.

I threw a couple features onto ruby-pocketsphinx-server - an ability to strip audio data out of video files, and the ability to fetch data from an url (instead of posting the data itself, a client can post an url as a parameter, and the audio is fetched by the server. Maybe when we get a common base that works I can contribute those features.

alumae commented 12 years ago

The patch is here: https://sourceforge.net/tracker/?func=detail&aid=3521952&group_id=1904&atid=301904

wiseman commented 11 years ago

Have you tried to get the patch accepted by the pocketsphinx maintainers? The easier access to cmn priors especially seems useful.

alumae commented 11 years ago

I submitted the patch to cmusphinx long time ago (http://sourceforge.net/p/cmusphinx/patches/90/), it's being ignored. I'll ask the cmusphinx developers.