Wyoming Whisper ASR Bridge

This python module mimcs Wyoming STT protocol of Faster Whisper but rather than loading an individual WhisperModel, it consumes Whisper ASR Box endpoint to transcribe audio.

Whisper ASR Box supports cuda by default, thus reducing the resources loaded on GPU or target hardware.

Naturally, the beam size is defaulted to 5 as implemented in Faster Whisper ASR Engine

Additionally can also configure voice activity detection (VAD) parameter

Local Install & Run

pip install -e /dir/wyoming_whisper_bridge

exec python3 \ 
    -m wyoming_whisper_bridge \ 
    --uri 'tcp://0.0.0.0:10300' \ 
    --language en \           # Language to Transcribe
    --language-codes en,uk \  # Language code(s) set for Wyoming Whisper Model (HA)
    --model-name Systran/faster-distil-whisper-small.en \  # WhisperModel loaded on ASR Webservice
    --asr-endpoint http://whisper-asr-service.tld/asr  \   # ASR Webservice Endpoint
    --vad true \    # Enable the voice activity detection (VAD) to filter out parts of the audio without speech (default: false)
    --asr-endpoint-timeout 5  # (default: 5)

Docker Image

docker run -it \ 
    -p 10300:10300 \ 
    ghcr.io/gyaneesh/wyoming-whisper-bridge:latest \ 
    --model-name Systran/faster-distil-whisper-small.en \ 
    --language en \ 
    --asr-endpoint http://whisper-asr-service.tld/asr

Acknowledgements

• ahmetoner
• rhasspy team