jamiepine/voicebox

What it solves

Voicebox is a local-first AI voice studio that provides a unified interface for both voice input (speech-to-text) and voice output (text-to-speech). It eliminates the need for cloud-based services like ElevenLabs or WisprFlow by allowing users to clone voices, generate expressive speech, and dictate text into any application on their own machine, ensuring complete privacy.

How it works

The application runs a suite of local AI models for different tasks:

• TTS Engines: It bundles seven different text-to-speech engines (including Qwen3-TTS, Kokoro, and HumeAI TADA) to handle multilingual cloning and expressive speech.
• STT Engine: It uses OpenAI Whisper for high-accuracy transcription of dictation and recordings.
• Local LLM: A bundled Qwen3 LLM is used for refining dictation (removing stutters) and powering "Voice Personalities" that rewrite text to match a specific character's style.
• Integration: It provides a REST API and a Model Context Protocol (MCP) server, allowing AI agents (like Claude Code or Cursor) to speak using cloned voices.
• Performance: Built with Tauri (Rust) and FastAPI (Python), it supports hardware acceleration across macOS (MLX/Metal), Windows (CUDA/DirectML), and Linux (ROCm/CUDA).

Who it’s for

• Content Creators: Those producing podcasts, narratives, or game dialogue using the multi-voice timeline editor.
• AI Agent Developers: Developers who want to give their agents a local, cloned voice via MCP.
• Accessibility Users: People who need speech assistance or a way to communicate in a cloned version of their own voice.
• General Users: Anyone wanting a private, local alternative to cloud TTS/STT services.

Highlights

• Zero-shot Voice Cloning: Clone any voice from a few seconds of audio.
• Global Dictation: A system-wide hotkey that pastes transcribed text directly into any focused text field (macOS).
• Expressive Speech: Support for paralinguistic tags like [laugh] or [sigh] via Chatterbox Turbo.
• Agent Voice Output: Integration with MCP-aware agents to allow them to speak in specific cloned voices.
• Multi-track Stories Editor: A timeline for composing complex conversations and narratives.
• Post-Processing: Built-in audio effects like pitch shift, reverb, and compression.