
yovoice
Open-source local studio for speech, voice cloning, songs and MIDI scores on Mac, Windows and Linux.

About yovoice
yovoice: A Local Studio for Voice and Music
yovoice is a free, open-source desktop studio for speech, voice cloning, songs and multi-track scores that runs on your own computer. It brings speech and music models such as IndexTTS, Qwen3-TTS, VoxCPM2 and ACE-Step into one app on macOS, Windows and Linux. Inference runs locally on the native C++ engine audio.cpp, so no Python setup is needed.
Key Features
Story Dubbing: Import text, EPUB or subtitle files, split them into lines, assign a voice to each character and edit the generated clips on a timeline.
Voice Cloning and Design: Clone a voice from a clip you upload or record, or describe a voice in words to design a new one.
Emotion and Delivery: Shape delivery with emotion tags, a reference performance or text prompts, and add tags such as laughter. Controls vary by model.
Speech Recognition: Turn audio or video into timed subtitles with Qwen3 ASR or Parakeet TDT, exported as SRT or WebVTT.
Music and Sound Effects: Generate songs from a style prompt and lyrics with ACE-Step 1.5 or YuE2, or instrumentals and sound effects with Stable Audio 3, plus 6-stem separation.
Multi-Track Score: Write parts bar by bar, import MIDI or transcribe audio to MIDI, and render with General MIDI sound fonts without a GPU.
Audio Checks and Loudness: Flags clipping, clicks and long silences, and exports at audiobook (-18 LUFS), podcast (-16 LUFS) or video (-14 LUFS) loudness.
Local and Offline: Uses Metal on Apple Silicon, CPU, CUDA or Vulkan on Windows, and CPU or Vulkan on Linux. Once models are downloaded it works offline, and files stay on disk.
CLI, HTTP API and MCP: A CLI and Agent Skill script speech and score rendering, and yovoice serve exposes an authenticated HTTP API and MCP service.
Use Cases
Audiobooks and Stories: Narrate long texts with several characters.
Voiceovers and Subtitles: Voice a video in a cloned or designed voice, or subtitle it.
Music Creation: Draft songs, instrumentals and scores on one machine.
Agent Workflows: Let agents such as Claude Code or Cursor generate speech through MCP or the CLI.
Pricing and Licensing
yovoice is free and open source under Apache-2.0. Model weights carry their own licenses, and some, such as OmniVoice, YuE2 and Voxtral 4B TTS, are non-commercial only.
Getting Started
Website: https://yovoice.leemysw.com/en/
Source code: https://github.com/leemysw/yovoice
yovoice gives creators and developers one local app for voices, songs and scores.