Wyoming Whisper API client

Wyoming protocol server for the Whisper API speech to text system.

My motivation was to establish a one working STT system for whole household, additionally to Home Assistant I use it for Blurt gnome shell extension.

You need a running Whisper compatible API service.

Free Whisper compatible API servers

Nvidia Parakeet

Do not let the name fool you, this model works very efficiently on a CPU and it's the model I am currently running for my needs.

I am using a docker version from this repository, but it's up to you how do you like to make it running.

https://github.com/chand1012/parakeet-server

Whisper.cpp

Another example is Whisper.cpp instance which requires a GPU for speedy inference.

https://github.com/ggerganov/whisper.cpp/tree/master/examples/server

I run it on nvidia GPU with fantastic results with detailed inference on large model in usually about a second:

whisper.cpp/server -m whisper.cpp/models/ggml-large-v3-q5_0.bin --host 0.0.0.0 --port 8910 --print-realtime --print-progress

You need to study whisper.cpp to get more information about running its STT service.

Installation

Local

Clone the repository and set up Python virtual environment:

git clone https://github.com/ser/wyoming-whisper-api-client
cd wyoming-whisper-api-client
script/setup

Run a server anyone can connect to:

./script/run --uri tcp://0.0.0.0:7891 --debug --api http://192.168.41.49:8910/inference

To save the audio of every request as a timestamped .wav file, add --log-dir:

./script/run --uri tcp://0.0.0.0:7891 --debug --api http://192.168.41.49:8910/inference --log-dir ./recordings

Docker

(Optional) The image is already published in GHCR. Build locally with:

git clone https://github.com/ser/wyoming-whisper-api-client
cd wyoming-whisper-api-client
docker build -t ghcr.io/ser/wyoming-whisper-api-client:latest .

Run the container:

docker run -p 7891:7891 -it --rm --name wyoming-whisper-api-client ghcr.io/ser/wyoming-whisper-api-client:latest \
    --debug \
    --uri tcp://0.0.0.0:7891 \
    --api http://192.168.41.49:8910/inference

To save the audio of every request as a .wav file, mount a host directory and point --log-dir at it (otherwise the files stay inside the container and are lost when it is removed):

docker run -p 7891:7891 -it --rm --name wyoming-whisper-api-client \
    -v "$(pwd)/recordings:/recordings" \
    ghcr.io/ser/wyoming-whisper-api-client:latest \
    --debug \
    --uri tcp://0.0.0.0:7891 \
    --api http://192.168.41.49:8910/inference \
    --log-dir /recordings

Attach for logging:

docker logs -f wyoming-whisper-api-client

Stop the container:

docker stop wyoming-whisper-api-client

Acknowledgements

1. It's a rewrite of Michael Hansen's wyoming-faster-whisper.
2. Tests are not functioning as there is no public Whisper API service to test it out.

Support

Matrix: #serdev:sergevictor.eu

Contribution

I do not accept any code or documentation contributions which have not been previously agreed upon. This means the order of steps is as follows:

1. Open an issue or discussion topic first.
2. Have a conversation about the matter at hand.
3. Reach a consensus about both form and content of the contribution.
4. Wait for a yay/nay signal from me.

Please do not submit any pull requests before I ask for them, as they will be closed. This outlined, proven practice will save time for all involved.

Alternatives

1. https://github.com/roryeckel/wyoming_openai - A proxy which handles STT and TTS in one.