vapid mica
I got it all working with cuda after some massaging of the dockerfile.
It returns all exclation marks for any audio.
It takes a long time because it retries several times, and this occurs with the same values every time no matter how long or what audio is sent.
I simply added it via the web UI - is this a sample rate setup problem or something?
The exact same thing happens with small and medium models
halcyon drum
Could try turning on debug in your ha configuration.yaml and listen to the recording to see if something is wrong there, spend be instructions in the ha docs. Does it work with other stt providers?
Also seems like something is a bit off, log shows the whisper model has no initial prompt, i think usually the model has some basic instruction prompt to tell it how to transcribe
novel orbit
I got it all working with cuda after some massaging of the dockerfile.
It return...
can you post your faster-whisper config? e.g. its compose file (assuming your running it on docker)
vapid mica
podman stop wyoming-faster-whisper
podman rm wyoming-faster-whisper
podman run -d \
--name wyoming-faster-whisper \
-p 10300:10300 \
--volume /tank/services/wyoming-faster-whisper/data:/data \
--restart=unless-stopped \
--device nvidia.com/gpu=all \
wyoming-whisper-cuda \
--model medium \
--language en --debug
matt@queen:~$ cd wyoming-faster-whisper-cuda/
matt@queen:~/wyoming-faster-whisper-cuda$ ls
Dockerfile README.md run.sh
matt@queen:~/wyoming-faster-whisper-cuda$ cat run.sh
#!/bin/bash
export LD_LIBRARY_PATH=/usr/local/lib/python3.11/dist-packages/nvidia/cudnn/lib/:/usr/local/lib/python3.11/dist-packages/nvidia/cublas/lib/
#export LD_LIBRARY_PATH=`python3 -c 'import os; import nvidia.cublas.lib; import nvidia.cudnn.lib; print(os.path.dirname(nvidia.cublas.lib.__file__) + ":" + os.path.dirname(nvidia.cudnn.lib.__file__))'`
echo $LD_LIBRARY_PATH
python3 -m wyoming_faster_whisper --uri tcp://0.0.0.0:10300 --data-dir /data --download-dir /data --device cuda "$@"
it works with ha cloud
novel orbit
```matt@queen:~$ cat /opt/admin/update_wyoming_whisper.sh
podman stop wyoming-f...
can you try medium-int8 as the model?
novel orbit
... haha, works perfectly and quickly
and we have a winner 🙂