KoljaB
commented
4 months ago This warning can be ignored, must be some other issue. The thing is, sometimes some libraries I use (mostly WebRTCVad) don't throw errors (or log any other information) if they don't detect anything like voice activity or wakewords (also when something is wrong with the inputs), so hard to detect. Can't do much to verify too with realtime chunks too.
The pipeline is: Microphone recording -> PvPorcupine Wakeword detection -> WebRTCVad -> SileroVAD -> Whisper All last 4 can fail. WebRTCVad (and SileroVAD) often don't log much, WebRTCVad also sensible towards incoming chunks. PvPorcupine is rarely used and logs bit better. Whisper would tell you if something was off.
Please try this as first lines of your script:
import logging
logging.basicConfig(level=logging.DEBUG)Then create AudioToTextRecorder object with recorder = AudioToTextRecorder(level=logging.DEBUG)
This sets highest loglevel for all used libs. Maybe you can then see anything in the logging why it fails (sometimes yes, sometimes not - if not you basically have to test every single lib for its own).
Denintendator
rajaiswal
rihp
M4a1x
Just spinner with text "speak now". Tried large-v3, small and tiny models, no difference. Mic working well, I just tested it by recording audio using python. It's probably trying, because CPU is heavily loaded, but there's no result.
MacOS Sonoma 14.4
There are no errors in console, just warning: [2024-04-24 23:53:30.031] [ctranslate2] [thread 8069444] [warning] The compute type inferred from the saved model is float16, but the target device or backend do not support efficient float16 computation. The model weights have been automatically converted to use the float32 compute type instead.