convez.ai
MCP ServerIntegração & APIsIntermediário
falcoschaefer99-eng

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 GitHub
InicialPluginScore · de 0 a 100

Servidor 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-tts

1 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

Estrelas
1
Forks
1
Acompanhando
0
Saúde
42%
Último commit
há 6 meses
Versão
v2.0.0
Issues abertas
0
Contribuidores
3

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.

EngineWhat it doesPlatform
Kokoro-82M54 preset voices, ~1s generationAll platforms
IndexTTS-1.5Natural voice cloningApple Silicon
Chatterbox OGVoice cloning, cross-platform fallbackWindows / 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?

PlatformFile
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

VariableDefaultDescription
KOKORO_VOICEam_onyxDefault voice ID
KOKORO_SPEED1.0Speed multiplier (0.5 - 2.0)

Voices

54 preset voices across 9 languages:

LanguageFemaleMale
American Englishaf_alloy, af_aoede, af_bella, af_heart, af_jessica, af_kore, af_nicole, af_nova, af_river, af_sarah, af_skyam_adam, am_echo, am_eric, am_fenrir, am_liam, am_michael, am_onyx, am_puck, am_santa
British Englishbf_alice, bf_emma, bf_isabella, bf_lilybm_daniel, bm_fable, bm_george, bm_lewis
Spanishef_doraem_alex, em_santa
Frenchff_siwis--
Hindihf_alpha, hf_betahm_omega, hm_psi
Italianif_saraim_nicola
Japanesejf_alpha, jf_gongitsune, jf_nezumi, jf_tebukurojm_kumo
Portuguesepf_dorapm_alex, pm_santa
Mandarinzf_xiaobei, zf_xiaoni, zf_xiaoxiao, zf_xiaoyizm_yunjian, zm_yunxi, zm_yunxia, zm_yunyang

Use muse_list_voices inside Claude to browse them interactively.

Tools

ToolWhat it does
muse_speakSpeak text — preset voice, named clone, or custom ref audio
muse_list_voicesBrowse all voices and clones, filter by language
muse_checkVerify engines, platform, and configuration

How It Works

Auto-detects the best engine for your platform:

PlatformPreset EngineCloning Engine
macOS Apple Siliconmlx_audioIndexTTS-1.5
Windowskokoro PyTorchChatterbox OG
Linuxkokoro PyTorchChatterbox OG
Intel Mackokoro PyTorchChatterbox 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) or kokoro + soundfile (any platform)
  • Optional: chatterbox-tts for 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.


Ver o original no GitHub

Bom para

Relacionados