Description

This plugin adds Vosk speech-to-text support to OpenVoiceOS. It transcribes audio with a local Kaldi model, either after recording finishes or in real time as you speak.

Install

pip install ovos-stt-plugin-vosk

Get official models from alphacephei.

Configuration

Give the plugin a Kaldi model. Point model at a local folder or a direct download URL.

  "stt": {
    "module": "ovos-stt-plugin-vosk",
    "ovos-stt-plugin-vosk": {
        "model": "/path/to/unzipped/model/folder"
    }
  }

Advanced configuration

  "stt": {
    "module": "ovos-stt-plugin-vosk-streaming",
    "ovos-stt-plugin-vosk-streaming": {
        "model": "http://alphacephei.com/vosk/models/vosk-model-small-en-us-0.15.zip",
        "verbose": true
    },
    "ovos-stt-plugin-vosk": {
        "model": "http://alphacephei.com/vosk/models/vosk-model-small-en-us-0.15.zip"
    }
  }

ovos-stt-plugin-vosk runs regular STT. Transcription happens after recording finishes.

ovos-stt-plugin-vosk-streaming runs streaming STT. Transcription happens in real time.

verbose prints partial transcriptions.

model sets the full path or direct download URL for the model.

lang is optional. If you do not set model, the plugin downloads a default small model for this language, if one exists.

Docker

Use this plugin together with ovos-stt-http-server.

docker run -p 8080:8080 ghcr.io/openvoiceos/vosk-stt-http-server:master