README.md

Description

This plugin adds Faster Whisper speech recognition to OpenVoiceOS. Faster Whisper is a fast inference engine for OpenAI's Whisper automatic speech recognition models.

The plugin also provides a language detection transformer for ovos-dinkum-listener.

Install

pip install ovos-stt-plugin-fasterwhisper

Models

The plugin supports these Whisper model names: tiny.en, tiny, base.en, base, small.en, small, medium.en, medium, large-v1, large-v2, large-v3, large, distil-large-v2, distil-medium.en, distil-small.en, distil-large-v3.

You can also pass a full path to a local model, or a Hugging Face repo ID, for example projecte-aina/faster-whisper-large-v3-ca-3catparla.

To use a model that is not already in Faster Whisper format, convert it, or pick a compatible model on Hugging Face.

Configuration

This example configures the large-v3 model to run on GPU:

  "stt": {
    "module": "ovos-stt-plugin-fasterwhisper",
    "ovos-stt-plugin-fasterwhisper": {
        "model": "large-v3",
        "use_cuda": true,
        "compute_type": "float16",
        "beam_size": 5,
        "cpu_threads": 4
    }
  }

This example uses Faster Whisper for language detection. It works only with ovos-dinkum-listener.

  "listener": {
    "audio_transformers": {
        "ovos-audio-transformer-plugin-fasterwhisper": {
            "model": "small"
        }
    }
  }

Related projects

• OpenVoiceOS/ovos-plugin-manager — the plugin manager that loads STT and audio transformer plugins.
• OpenVoiceOS/ovos-dinkum-listener — the listener service that runs audio transformer plugins such as the language detector in this repository.

License

Apache-2.0