You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Vosk is v1.6.0, and Piper is v1.5.0. Assigned in Assist them, used Russian language for TTS and STT. But when I enable debug_recording_dir: /share/assist_pipeline and explore my records in that folder, so it cut off the end of my speech too soon and some long commands like 'set brightness to 70' are cut off on 'set brightness...> in my wav files in reality. Vosk ignores even 'Finished speaking detection' =>Relaxed. And whisper is good enough on that, but it's slower with the same level of positive/negative STT detections as on the Vosk. Vosk is really fast and precise, but only on the short commands. So this cut-off looks like a bug.
The text was updated successfully, but these errors were encountered:
Vosk is v1.6.0, and Piper is v1.5.0. Assigned in Assist them, used Russian language for TTS and STT. But when I enable debug_recording_dir: /share/assist_pipeline and explore my records in that folder, so it cut off the end of my speech too soon and some long commands like 'set brightness to 70' are cut off on 'set brightness...> in my wav files in reality. Vosk ignores even 'Finished speaking detection' =>Relaxed. And whisper is good enough on that, but it's slower with the same level of positive/negative STT detections as on the Vosk. Vosk is really fast and precise, but only on the short commands. So this cut-off looks like a bug.
The text was updated successfully, but these errors were encountered: