Wyoming Faster Whisper

Wyoming Faster Whisper converts speech to text on your own hardware for Home Assistant and other Wyoming-compatible clients. It's for home voice systems that need local transcription and better recognition of household names.

• Speech recognition backends: Use faster-whisper, Qwen3-ASR, Parakeet TDT, SenseVoice or GigaAM, with automatic backend selection by language and faster-whisper as the fallback.
• Home Assistant names: An optional connection supplies exposed device names, aliases, areas and floors to help faster-whisper and Qwen3-ASR recognize words specific to your home; Distil-Whisper models don't support this name biasing.
• Pause detection: Optional Silero voice detection lets the server finish a command after a pause and return its transcript without waiting for the client to end the recording.
• Language matching: The server reports the languages its configured backend supports and treats Cantonese separately from Mandarin when choosing a transcription model.

The project is open source under the MIT license. It runs as a Home Assistant add-on, Docker container or local Python service. CPU processing is supported. The GPU Docker image supports NVIDIA GPUs on amd64 hardware and requires a separate build; Parakeet through the sherpa backend still runs on the CPU in that image.

Home Assistant OS doesn't provide GPU passthrough, so GPU use requires a separate Docker or local deployment. The optional Home Assistant connection reads names through its API using an access token and doesn't call services.