Verified project record

jamiepine/voicebox

A local-first voice I/O stack that combines dictation, text-to-speech, voice cloning, and agent voice output. It is built with Tauri (Rust) and routes audio processing through local models such as Whisper and Kokoro.

Project overview

Bridges voice input and output locally, including a bundled local LLM for applying personas to voice profiles and an MCP server that enables coding agents to produce speech.

Project type: MCP · Model Runtime · Audio & Speech
Use cases: Meeting Notes · Audio & Speech · Automation
Deployment: Refer to project documentation
License: MIT

Best for

Developers and AI engineers using MCP-aware coding agents who want to add local voice output.
Creators requiring local voice cloning and multi-track audio editing without sending data to cloud-dependent studios.
Users who need global dictation alongside speech generation managed within a single native application.

Key capabilities

Supports zero-shot cloning from a reference sample and includes over 50 curated preset voices via Kokoro and Qwen CustomVoice.
Generates speech across 7 TTS engines and 23 languages, which can be switched per-generation.
Provides voice input through a global hotkey with push-to-talk and toggle modes. Includes auto-paste functionality that is verified for accessibility on macOS.
Exposes a tool call (voicebox.speak) that allows any MCP-aware agent to speak to the user in a cloned voice.
Allows a free-form persona to be attached to any voice profile. Users can then Compose, Rewrite, or Respond using a bundled local LLM.
Runs OpenAI Whisper for transcription, which backs the dictation feature, the Captures tab, and the /transcribe API.
Features a multi-track timeline with drag-and-drop functionality for conversations, podcasts, and narratives.
Includes 8 audio effects, such as pitch shift, reverb, delay, chorus, and compression, powered by pedalboard.

Limitations and risks

The auto-paste functionality for dictation is currently available only on macOS.
Pre-built binaries are not yet available for Linux.
Running multiple local models simultaneously may cause GPU contention or require significant memory.
Telemetry practices are currently not documented.

Getting started

Setup is documented as easy because pre-built binaries are available for macOS and Windows. The first success path is to download the installer for your OS, install it, and open the application.

Alternatives and comparisons

A local-first macOS application providing voice cloning, text-to-speech, document reading, audiobook creation, and an agentic MCP server optimized for Apple Silicon via MLX.
A tokenizer-free text-to-speech model supporting 30 languages, voice design from text descriptions, and voice cloning.
An industrial-grade speech recognition toolkit handling offline, streaming, and edge deployment with ASR, speaker diarization, and emotion recognition.

Project comparisons

Evidence and sources

GitHub project description: The open-source AI voice studio. Clone, dictate, create.
README: Native performance — built with Tauri (Rust), not Electron
README: Voicebox is a **local-first AI voice studio** — a free and open-source alternative to **ElevenLabs** and **WisprFlow** in one app. Clone voices from a few seconds of audio, genera…
Release: v0.5.0
README: Windows / Linux auto-paste | Dictation paste parity — `SendInput` on Windows, `uinput` / AT-SPI on Linux

Source updated at: 2026-07-16T18:10:45Z

Open original GitHub project