jamiepine/voicebox

What it solves

Voicebox is a local-first AI voice studio that provides a unified interface for voice cloning, speech generation, and system-wide dictation. It eliminates the need for separate cloud-based services for voice output (like ElevenLabs) and voice input (like WisprFlow), allowing users to run the entire voice I/O loop locally for complete privacy and no subscription fees.

How it works

The application combines several specialized AI models for different tasks:

• Text-to-Speech (TTS): It integrates seven different TTS engines (including Qwen3-TTS, Kokoro, and LuxTTS) to handle multilingual cloning and expressive speech.
• Speech-to-Text (STT): It uses OpenAI Whisper to provide high-quality transcription for dictation and audio captures.
• Local LLM: A bundled Qwen3 LLM handles text refinement (cleaning up stutters in dictation) and "Voice Personalities," which can rewrite text to match a specific character's style before it is spoken.
• Integration: Built with Tauri (Rust) and FastAPI (Python), it exposes a REST API and an MCP (Model Context Protocol) server, allowing AI agents like Claude Code or Cursor to speak using cloned voices.

Who it’s for

• Content Creators: People making podcasts, narratives, or game dialogue using the multi-voice timeline editor.
• Developers: Those building AI agents that need a local, private voice interface.
• Accessibility Users: Individuals who need speech assistance tools that run locally.
• Agentic AI Users: Users of MCP-aware agents who want their AI assistants to have a distinct, cloned voice.

Highlights

• Local-First Privacy: All models and voice data stay on the user's machine.
• ** uma 7 TTS Engines:** Support for 23 languages and zero-shot voice cloning from short audio samples.
• System-Wide Dictation: A global hotkey for push-to-talk dictation that pastes text directly into any focused app on macOS.
• MCP Server: Built-in support for the Model Context Protocol, enabling AI agents to trigger speech output.
• Stories Editor: A multi-track timeline for composing complex conversations and podcasts.
• Audio Effects: Integrated post-processing effects like pitch shift, reverb, and compression powered by Spotify's pedalboard.
• Broad Hardware Support: Optimized for Apple Silicon (MLX), NVIDIA (CUDA), AMD (ROCm), and Intel Arc GPUs.