Description

This is an STT plugin for OpenVoiceOS. It uses the speech recognition API of the Google Chrome browser.

The plugin ships with a bundled API key. Google stopped issuing new keys for this API (announcement), but the bundled key still works, most likely because it is also used inside Chromium itself.

A list of supported languages is in this Stack Overflow comment.

This is the same speech recognition service as the "google" module in mycroft-core. That module is supported in mycroft-core but cannot be used there, because mycroft-core cannot obtain new API keys. A pull request to fix this was proposed and blocked.

Using this plugin removes the need for a backend proxy, so latency is lower and you get more configuration options. Accuracy stays the same, since this is the same speech recognition service used by Mycroft Selene.

Install

pip install ovos-stt-plugin-chromium

Configuration

By default, the plugin uses the global language set in your OVOS configuration.

  "stt": {
    "module": "ovos-stt-plugin-chromium"
  }

Advanced configuration

  "stt": {
    "module": "ovos-stt-plugin-chromium",
    "ovos-stt-plugin-chromium": {
        "lang": "en-US",
        "pfilter": false,
        "debug": false
    }
  }

• lang — overrides the global language. Use a code from the supported languages list.
• pfilter — profanity filter. When true, the plugin censors profanity, showing only the first character and replacing the rest with asterisks, for example f***.
• debug — when true, logs the confidence score and alternative transcriptions.

Docker

You can run this plugin behind ovos-stt-http-server to expose it as an HTTP STT server, without needing device pairing.

docker run -p 8080:8080 ghcr.io/openvoiceos/google-stt-proxy:master

Public instances of ovos-stt-http-server that run this plugin:

• https://stt.smartgic.io/chromium, hosted by @goldyfruit