OVOS MaryTTS Plugin

This is a text-to-speech (TTS) plugin for OpenVoiceOS that connects to a MaryTTS server. The plugin sends text to the MaryTTS server and returns the audio it gets back. You must install and run MaryTTS separately, on the same machine or on a different one.

Install

$ pip install ovos-tts-plugin-marytts

Configuration

"tts": {
    "module": "ovos-tts-plugin-marytts",
    "ovos-tts-plugin-marytts": {
      "url": "http://0.0.0.0:59125",
      "voice": "cmu-slt-hsmm"
    }
}

Usage

Use the plugin directly, without OVOS, like this:

from ovos_tts_plugin_marytts import MaryTTS

engine = MaryTTS(config={"url": "http://0.0.0.0:59125"})
engine.get_tts("hello world", "test.wav")

Docker

This repository also builds a container that serves the plugin behind ovos-tts-server, an ElevenLabs-compatible HTTP API, on port 9666. The image is published to ghcr.io/openvoiceos/ovos-tts-plugin-marytts.

This plugin is a thin client. It forwards text to a separate MaryTTS Java server and returns the WAV file it gets back. The plugin container cannot synthesize speech on its own. It fails to serve requests unless you configure a reachable MaryTTS server URL, either with the url config key or with the MARYTTS_URL build argument (default http://marytts:59125).

Compose (recommended)

The docker-compose.yml file in this repository wires both pieces together: a MaryTTS server sidecar (synesthesiam/marytts:5.2, with 19 HSMM voices for 8 languages, on port 59125) and this plugin container, pointed at that sidecar.

$ docker compose up

The ElevenLabs-compatible API is then available at http://localhost:9666, backed by the MaryTTS server at http://localhost:59125.

To use a different voice, or to point at your own MaryTTS server, rebuild the plugin image with the MARYTTS_URL and MARYTTS_VOICE build arguments:

$ docker build --build-arg MARYTTS_URL=http://my-marytts:59125 \
               --build-arg MARYTTS_VOICE=dfki-spike-hsmm -t marytts-tts .

The MaryTTS backend

The synesthesiam/marytts image bundles HSMM voices for several languages and runs on amd64, arm/v7, and arm64.

$ docker run -it -p 59125:59125 synesthesiam/marytts:5.2

This may use a lot of RAM on a Raspberry Pi. Use the --voice flag to load only the voices you need, and list the available voices with --voices:

$ docker run -it -p 59125:59125 synesthesiam/marytts:5.2 --voice cmu-slt-hsmm --voice cmu-rms-hsmm
$ docker run -it synesthesiam/marytts:5.2 --voices

Related projects

• OpenVoiceOS/ovos-tts-server — the ElevenLabs-compatible HTTP API this plugin's container serves behind.
• MaryTTS — the backend TTS engine this plugin connects to.

Compatible projects

A few TTS server projects offer MaryTTS-compatible APIs and work with this plugin.

• Larynx
• OpenTTS

License

This project is licensed under the Apache License 2.0. See LICENSE for the full text.