yovoice

Project Overview

yovoice is an open-source, local-first audio and music creation studio for macOS and Windows that turns text into expressive speech, cloned voices, complete vocal songs and multi-track scores.

Powered by the native C++ audio.cpp engine, all inference runs on your own GPU or CPU with zero cloud API calls and zero per-character fees and no Python environment to configure; a built-in MCP server lets AI agents such as DeepSeek and Claude call its speech and scoring abilities.

Item	Value
Platforms	macOS 14+ (Apple Silicon), Windows 10/11 x64
Stack	Go + C++ (audio.cpp) + web frontend
Local models	IndexTTS 2.x, VoxCPM2, Qwen3-TTS, Kokoro, ACE-Step 1.5 and more
License	Apache-2.0

Text to speech, voices cloned, ideas composed into songs, all on your machine at zero API cost.

Key Features

• Speech with performance control: generate voice from a reference timbre, emotion adjustments or a delivery described in words
• Voice design and cloning: VoxCPM2 offers text-guided design, controllable cloning and transcript-assisted cloning at 48 kHz
• Song generation: ACE-Step 1.5 creates full songs or instrumentals from a style and section-marked lyrics, keeping every take
• Multi-track scores: write parts by bar and beat, import MIDI or let AI score it, render with GM sounds and export MIDI and stems
• Multi-take workspace: version comparison, waveform trimming and timeline scrolling with clean stem export
• Local model management: resumable downloads for IndexTTS, VoxCPM2, OmniVoice, Qwen3-TTS, Breeze TTS, Kokoro and more, with GGUF import
• Hardware acceleration: Metal on Apple Silicon, CUDA and experimental Vulkan on Windows
• MCP and agents: built-in MCP server for Cursor, Claude Code and others to write lyrics, scores and voiceovers

Target Audience

• Short-video creators who need batch voiceovers
• Music lovers who want to generate complete songs
• Authors producing audiobooks and podcasts
• Privacy-minded users who require offline generation
• Developers wiring AI agents into voice workflows

Quick Start

1. Download the installer for your platform from Releases
2. After launching, download the local models you need in model management
3. Type text to generate speech, or import reference audio for cloning
4. Generate songs from lyrics and a style, keeping versions to compare
5. Write or import MIDI in the score page, render and export stems

Core Advantages

• Fully local: manuscripts, reference voices and creations never leave the computer, no per-character fees
• Native performance: the C++ engine avoids Python environment conflicts, ready right after install
• Rich model choice: from the 189 MB Kokoro to full-song ACE-Step, pick by hardware
• Agent ecosystem: the MCP server lets AI coding assistants call speech and scoring directly

System Requirements

Item	Requirement
OS	macOS 14+ (Apple Silicon) or Windows 10/11 (x64)
Voice cloning	8 GB+ RAM or 4-6 GB VRAM recommended
Full songs	12 GB+ VRAM or 16 GB+ unified memory recommended
Optional acceleration	NVIDIA CUDA (Windows), Metal (macOS)

Notes

• Models download on demand: local models range from a hundred-plus MB to several GB, expect a wait on first use
• Full songs are demanding: ACE-Step 1.5 BF16 needs about 10 GB VRAM, start with lightweight models on modest hardware
• AI services optional: DeepSeek, OpenAI and similar services are only for writing lyrics and scores, keys stay on this machine