This is still unstable! Expect breaking changes on version updates!

A Discord bot providing real-time Text-to-Speech (TTS) in voice channels using Pocket-TTS for fast, high-quality speech generation. Built with Python for cross-platform compatibility.

• Primarily needs .safetensors models generated from Pocket-TTS
• Automatic TTS for muted users

• Make sure docker is installed
• Add voices to the voices directory
• Configure .env (Discord bot token, etc.)
• docker compose up

Download WSL2, then follow the instructions above.

• Store custom trained voices in a voices directory (create this yourself).
• Add new voices by:
  1. Collecting audio samples (wav, mp3, flac, m4a, ogg, opus)
  2. Using Pocket-TTS and the export-voices.sh script to convert/truncate/export
  3. Placing .safetensors files in the voices directory
• .wav files may be used (no idea if this works) but are slower; .safetensors is strongly recommended.

Use the export-voices.sh script to convert supported audio formats to truncated 30s, mono, 44kHz .wav, then export to .safetensors using Pocket-TTS.

• Bash only (not Windows compatible)
• Requires ffmpeg and Python package uv
• Usage:

  ./export-voices.sh <input_file_or_directory> <output_directory>

  Example:

  ./export-voices.sh ./samples ./voices

The export_voices.py script should do the same thing, and is platform independent.

See Pocket-TTS documentation for full training details.

• Python 3.x required
• Install dependencies using uv:

  uv sync

• Install ffmpeg (see ffmpeg download page)
• Create a voices directory and add your models
• Place your Discord bot token and config in a .env file

• !join
  • Joins the voice channel the caller is in.
  • Listens for messages from muted users in VC in the channel this command was called in.
  • Recommend you use this command in a VC adjacent text channel (no idea if the text channels embedded in a VC work)
• !voice <voice_name>
  • Set your TTS voice
  • Example: !voice Joe
• !s <text>
  • Speak text directly (no username prefix)
  • Example: !s Hello world!
• !prefix <on|off>
  • Toggle the 'User says:' prefix (on/off).
• !multi
  • Used for playing dialog from different voices back to back. Example:

!multi
alba: Hello everyone! How are you doing?
marius: I'm doing good!

Depends on having alba.safetensors and marius.safetensors inside the voices directory.

• Automatic TTS: Muted user’s text messages are spoken aloud in voice channels.

There's a web ui built with FastAPI + HTMX. It's a really simple UI that will allow you to play text outside of Discord. Note that it doesn't have any protection, so be cautious when deploying it publicly.

• Pocket-TTS repository
• ffmpeg install guides
• Discord developer portal
• uv documentation

Contributions welcome! Please:

• Suggest new features
• Submit bug reports and pull requests
• Help extend functionality (such as adding a voice listing command)

Open an issue or PR on GitHub to get involved.