VoiceStudio

VoiceStudio is a desktop application that puts an entire speech-AI pipeline on your own machine. Where hosted services bill per character and send your audio to their servers, VoiceStudio downloads the models and runs everything locally: no account, no API key, and no upload of the voices you clone.

One interface covers six jobs — cloning a voice from a short reference clip, designing a new voice from a description, dubbing video into another language, system-wide dictation behind a keyboard shortcut, transcription with speaker diarization, and long-form audiobook production. It claims coverage of 646 languages, and ships a batch queue plus vocal isolation for working through large jobs unattended.

Engines are swappable rather than fixed. Sixteen text-to-speech backends are bundled — OmniVoice is the default, alongside CosyVoice 3 and GPT-SoVITS — and eleven speech-recognition engines, defaulting to WhisperX with Faster-Whisper and MLX Whisper available. A built-in model catalogue handles downloading and switching between them.

VoiceStudio is also addressable by other software. The backend exposes an OpenAI-compatible REST API plus SSE and WebSocket transports, and mounts an MCP server at /mcp, so a coding agent such as Claude Code can generate speech, clone a voice or transcribe a file without leaving the terminal.

It is built with Tauri (Rust shell, React UI) and ships as a DMG, MSI and AppImage, with a Docker option. Requirements are macOS 13.3+ on Apple Silicon, Windows 10/11 x64, or Linux x86_64 with glibc 2.39+; 8 GB RAM minimum with 16 GB recommended, 10 GB of free disk, and an optional GPU (CUDA, Apple MPS/MLX, or ROCm). The application is licensed AGPL-3.0 — model weights keep their upstream terms, which is worth checking before commercial use. The project was previously named OmniVoice Studio.