matrix-stt-bot

Matrix Speech-To-Text Bot

Transcribes audio messages using OpenAI Whisper.

This bot is based on Simple-Matrix-Bot-Lib and whisper.cpp. It downloads audio messages from your homeserver, transcribes them locally and responds with the result as a text message.

Usage

The bot is available as an image on DockerHub. You can deploy it using docker-compose:

version: "3.7"

services:
  matrix-stt-bot:
    image: ftcaplan/matrix-stt-bot
    restart: on-failure
    volumes:
      - ./data/:/data/
    environment:
      - "HOMESERVER=https://matrix.example.com"
      - "USERNAME=@stt-bot:example.com"
      - "PASSWORD=<password>"
      - "ASR_MODEL=tiny"
      - "XDG_CACHE_HOME=/data/cache"

Configuration

The bot will download the model file on first run to reduce image size. Available models are tiny.en, tiny, base.en, base, small.en, small, medium.en, medium, and large. The default is ASR_MODEL=tiny.

You can authenticate using tokens instead of a password by setting LOGIN_TOKEN=<login-token> or ACCESS_TOKEN=<access-token> instead of PASSWORD=<password>.