MUSE TTS — Síntese de voz local com 54 vozes
falcoschaefer99-eng/muse-tts
54 vozes, 3 engines, clonagem de voz, 100% local. Nada sai da máquina.
Ver no GitHubServidor MCP que integra três motores de TTS (Kokoro, IndexTTS, Chatterbox) para síntese de voz totalmente local, com 54 vozes pré-configuradas e clonagem de voz a partir de áudio de referência. Funciona em macOS, Windows e Linux sem enviar dados para servidores externos.
Instalação
npx convez-ai falcoschaefer99-eng/muse-tts1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Dar voz a agentes de IA com personalidades distintas
- Clonar vozes customizadas para aplicações de áudio
- Gerar narração e audiobooks offline em múltiplos idiomas
- Integrar síntese de voz em workflows privados sem dependência de APIs externas
Veredito

Solução madura e bem-arquitetada para TTS local com suporte multi-plataforma. Especialmente valiosa para quem precisa de privacidade garantida e quer evitar custos recorrentes de APIs de voz na nuvem.
README
A tradução para o português deste README ainda está na fila.
Install
Quickest path — install from PyPI:
pip install muse-tts # preset voices only (Kokoro)
pip install muse-tts[cloning] # + voice cloning (Chatterbox, cross-platform)
pip install muse-tts[macos] # + Apple Silicon optimization (mlx_audio + IndexTTS cloning)
Note for Apple Silicon users:
[macos]includes both Kokoro optimization AND IndexTTS voice cloning. You don't need[cloning]in addition.
Then start the server:
muse-tts
What Is This?
Three TTS engines, one MCP server. Ask Claude to speak and it does — through your speakers, in any of 54 voices, with cloning from any reference audio. Nothing leaves your machine.
Looking for a persistent player embedded in Claude's chat? See MUSE TTS Embed.
| Engine | What it does | Platform |
|---|---|---|
| Kokoro-82M | 54 preset voices, ~1s generation | All platforms |
| IndexTTS-1.5 | Natural voice cloning | Apple Silicon |
| Chatterbox OG | Voice cloning, cross-platform fallback | Windows / Linux |
Add to Claude
Claude Desktop
{
"mcpServers": {
"muse-tts-live": {
"command": "muse-tts"
}
}
}
Claude Code
claude mcp add muse-tts-live muse-tts
Then ask Claude to speak. It now has muse_speak, muse_list_voices, and muse_check.
Install from source (developers)
If you're cloning the repo to hack on it or need pinned hashes:
Which file for your platform?
| Platform | File |
|---|---|
| Apple Silicon Mac (M1/M2/M3/M4) | requirements-macos.txt |
| Intel Mac, Windows, Linux (preset voices only) | requirements-standard.txt |
| Intel Mac, Windows, Linux (+ voice cloning) | requirements-cloning.txt |
macOS (Apple Silicon — fastest):
pip install -r requirements-macos.txt --require-hashes
Windows / Linux / Intel Mac — preset voices only:
pip install -r requirements-standard.txt --require-hashes
Windows / Linux / Intel Mac — add voice cloning:
pip install -r requirements-cloning.txt --require-hashes
On Linux, you also need
espeak-ng:sudo apt install espeak-ng
Then run the server directly:
python server.py
Or add the cloned repo path to Claude Desktop:
{
"mcpServers": {
"muse-tts-live": {
"command": "python3",
"args": ["/path/to/muse-tts/server.py"]
}
}
}
Voice Cloning
Clone any voice from a reference clip:
"Speak this using the reference audio at ~/Downloads/my_voice.wav"
Adding Permanent Clones
Bring your own reference audio: drop a .wav into voices/, detected on restart, available as clone="filename" (without the .wav extension). Reference samples are not bundled with the package — bring your own.
Reference Audio Tips
- Format: WAV, 24kHz, mono
- Length: 10-30 seconds of clean speech
- Quality: Clear audio, minimal background noise
Convert your audio:
ffmpeg -i input.mp3 -ar 24000 -ac 1 -t 15 reference.wav
Configuration
| Variable | Default | Description |
|---|---|---|
KOKORO_VOICE | am_onyx | Default voice ID |
KOKORO_SPEED | 1.0 | Speed multiplier (0.5 - 2.0) |
Voices
54 preset voices across 9 languages:
| Language | Female | Male |
|---|---|---|
| American English | af_alloy, af_aoede, af_bella, af_heart, af_jessica, af_kore, af_nicole, af_nova, af_river, af_sarah, af_sky | am_adam, am_echo, am_eric, am_fenrir, am_liam, am_michael, am_onyx, am_puck, am_santa |
| British English | bf_alice, bf_emma, bf_isabella, bf_lily | bm_daniel, bm_fable, bm_george, bm_lewis |
| Spanish | ef_dora | em_alex, em_santa |
| French | ff_siwis | -- |
| Hindi | hf_alpha, hf_beta | hm_omega, hm_psi |
| Italian | if_sara | im_nicola |
| Japanese | jf_alpha, jf_gongitsune, jf_nezumi, jf_tebukuro | jm_kumo |
| Portuguese | pf_dora | pm_alex, pm_santa |
| Mandarin | zf_xiaobei, zf_xiaoni, zf_xiaoxiao, zf_xiaoyi | zm_yunjian, zm_yunxi, zm_yunxia, zm_yunyang |
Use muse_list_voices inside Claude to browse them interactively.
Tools
| Tool | What it does |
|---|---|
muse_speak | Speak text — preset voice, named clone, or custom ref audio |
muse_list_voices | Browse all voices and clones, filter by language |
muse_check | Verify engines, platform, and configuration |
How It Works
Auto-detects the best engine for your platform:
| Platform | Preset Engine | Cloning Engine |
|---|---|---|
| macOS Apple Silicon | mlx_audio | IndexTTS-1.5 |
| Windows | kokoro PyTorch | Chatterbox OG |
| Linux | kokoro PyTorch | Chatterbox OG |
| Intel Mac | kokoro PyTorch | Chatterbox OG |
Audio playback is handled natively (afplay on Mac, SoundPlayer on Windows, aplay/paplay on Linux).
Requirements
- Python 3.10+
- One of:
mlx_audio(Mac M-series) orkokoro+soundfile(any platform) - Optional:
chatterbox-ttsfor voice cloning on non-Apple platforms - ~200MB disk (Kokoro model) + ~1.5GB (IndexTTS-1.5) or ~2.5GB (Chatterbox)
License
Licensed under the Apache License, Version 2.0.
Copyright 2026 The Funkatorium (Falco & Rook Schäfer). Protected under German Copyright Law (Urheberrechtsgesetz). Jurisdiction: Amtsgericht Berlin.