search

synesthesiam / rhasspy

Rhasspy voice assistant for offline home automation
https://rhasspy.readthedocs.io
MIT License
941 stars 101 forks source link

Mixed Language Model Causing STT to fail #241

Closed athanasiusuk closed 4 years ago

athanasiusuk commented 4 years ago

Replicated this on two installs. Whenever the mix ratio is above 0 on either pocketsphinx or kaldi then STT and therefore intent recognition fails. Retraining/restarting does not solve the issue. I notice that whilst the log shows no errors it feels like training is not successful as I don’t get the training duration note. Recordings happening fine. Is this a known issue?

synesthesiam commented 4 years ago

Please try upgrading to 2.5 and see if the issue persists. Mixed modeling hasn't been thoroughly tested on 2.4 in a while.