Free local text to speech for LibreChat

1Start the speech server

Docker is the easiest choice: the image includes FFmpeg, and LibreChat plays MP3 audio.

docker run -d --name loudness-tts -p 8880:8880 -v loudness-tts:/data ghcr.io/loudness-ai/loudness-tts

Or with Python, after installing FFmpeg (see the installation guide):

python -m pip install loudness-tts
loudness-tts serve --host 0.0.0.0 --port 8880 --api-key choose-a-secret

2Add the speech settings to librechat.yaml

speech:
  tts:
    allowedAddresses:
      - "host.docker.internal:8880"
    openai:
      url: "http://host.docker.internal:8880/v1/audio/speech"
      apiKey: "${TTS_API_KEY}"
      model: "tts-1"
      voices: ["nova", "shimmer", "onyx", "echo", "alloy"]

The url is the full speech address, ending in /v1/audio/speech. LibreChat blocks addresses on your own network unless they are listed under allowedAddresses, written as host and port only. People can choose only the voices in the list.

3Set the API key

In LibreChat’s .env file, add the key that the settings refer to: the secret you passed to --api-key, or any text if the server has none.

TTS_API_KEY=choose-a-secret

4Restart LibreChat

docker compose down
docker compose up -d

LibreChat reads librechat.yaml when it starts. With Docker, the file must be mounted into the api container, as described in LibreChat’s speech documentation.

5Turn on text to speech

In LibreChat, open Settings → Speech, turn on Text to Speech, choose the OpenAI engine and a voice from your list. Then use the read-aloud button under a message.

6Optional: run the server inside LibreChat’s Docker Compose

Add loudness-tts to docker-compose.override.yml, next to LibreChat:

services:
  loudness-tts:
    image: ghcr.io/loudness-ai/loudness-tts
    volumes: ["loudness-tts:/data"]
volumes:
  loudness-tts:

Then use url: "http://loudness-tts:8880/v1/audio/speech" and allowedAddresses: ["loudness-tts:8880"].