Voicebox Complete Guide: Open-Source AI Voice Workstation for Local Voice Cloning, Speech Generation & Global Dictation

TL;DR: Voicebox is a fully local, open-source AI voice workstation — it brings together ElevenLabs-level voice cloning and WisprFlow-level global dictation in a single app, and it's free.

What Is It?

Voicebox is an open-source AI voice workstation developed by jamiepine. It has already earned 45,000+ stars on GitHub, making it one of the hottest local voice tools out there.

The core problem it solves is simple: voice input and output shouldn't depend on the cloud.

• Voice output (TTS): Choose from 7 TTS engines, support for 23 languages, zero-shot voice cloning with just seconds of audio
• Voice input (STT): Global hotkey dictation powered by Whisper — local speech-to-text
• AI integration: Built-in MCP server lets AI coding assistants like Claude Code and Cursor "speak out loud"

Unlike ElevenLabs (TTS only) or WisprFlow (dictation only), Voicebox handles both ends and adds a local LLM as the bridge — text polishing, personality settings, tone adjustment, all running on your own machine.

Feature Overview

Installation

macOS (Recommended)

Download the DMG installer:

# Apple Silicon (M1/M2/M3/M4)
# Visit https://voicebox.sh/download/mac-arm

# Intel Mac
# Visit https://voicebox.sh/download/mac-intel

After downloading, double-click the DMG and drag it to Applications. On first launch, macOS may warn about an "unverified developer" — go to System Settings → Privacy & Security and click "Open Anyway."

Windows

# Download the MSI installer
# Visit https://voicebox.sh/download/windows
# Double-click to install — CUDA is auto-configured if you have an NVIDIA GPU

Docker (Cross-Platform)

# Clone the repo
git clone https://github.com/jamiepine/voicebox.git
cd voicebox

# Start
docker compose up -d

# Access the Web UI
open http://localhost:3000

Linux

No pre-built packages yet — you'll need to build from source:

# Install dependencies
sudo apt install -y build-essential pkg-config libssl-dev \
  libasound2-dev libjack-dev portaudio19-dev

# Install Rust + Tauri dependencies
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
source $HOME/.cargo/env

# Clone and build
git clone https://github.com/jamiepine/voicebox.git
cd voicebox
cargo tauri build

# Run
./src-tauri/target/release/voicebox

For a detailed Linux build guide, see: voicebox.sh/linux-install

Getting Started: Generate Your First Speech in 5 Minutes

Step 1: Choose a TTS Engine

After launching Voicebox, pick a TTS engine from the top selector. Beginners should start with Kokoro — it's only 82M parameters, runs on CPU, and is the fastest option.

Step 2: Enter Text and Generate Speech

Type your content into the text box and click generate:

Hello, welcome to Voicebox. This is a demo of local AI voice synthesis.

On first run, Voicebox will automatically download the model files for your chosen engine (typically 100MB–2GB). Just be patient — subsequent uses won't need any downloads.

Step 3: Preview and Adjust

Once generation is complete, you can: - Click play to preview - Adjust speed and pitch - Add audio effects (reverb, pitch shift, etc.) - Re-generate with a different random seed for a different delivery

Advanced Features

Voice Cloning: Copy Any Voice in Seconds

This is one of Voicebox's most powerful features.

1. Prepare a clean voice recording of 5–30 seconds (WAV/MP3/FLAC all work)
2. Go to Profiles → Create Profile
3. Upload the audio file, or use the built-in recorder
4. Name the voice and select a language tag
5. Save it, then pick this profile when generating speech

# Supported cloning engines:
# - Qwen3-TTS: High-quality multilingual cloning
# - Chatterbox series: Zero-shot cloning
# - Kokoro: Mostly preset voices, limited cloning capability

# Tips:
# - The cleaner your reference audio, the better the clone
# - Avoid background music and noise
# - Multiple reference clips can improve quality

Global Dictation: Voice Input in Any App

Voicebox's global dictation lets you speak into any text input field on your system:

macOS Setup: 1. Open Voicebox → Settings → Dictation 2. Enable Global Hotkey (default: Option + Space) 3. Grant Accessibility and Input Monitoring permissions 4. Choose a Whisper model size (Turbo recommended — fast and accurate)

How to Use: - Hold-to-talk: Hold the hotkey while speaking, release to auto-transcribe and paste - Toggle mode: Tap the hotkey to start/stop recording

# Whisper model recommendations:
# - Turbo: ~8x faster than Large with minimal quality loss (recommended)
# - Base: Smallest and fastest, good for simple scenarios
# - Medium: Balanced choice
# - Large: Highest quality, needs a decent GPU

Audio Effects: 8 Post-Processing Options

After generating speech, you can apply professional-grade audio effects:

# 4 built-in presets:
# - Robotic
# - Radio
# - Echo Chamber
# - Deep Voice

# 8 adjustable effects:
# 1. Pitch Shift    — ±12 semitones
# 2. Reverb         — Room size, damping, wet/dry ratio
# 3. Delay          — Echo delay
# 4. Chorus/Flanger — Chorus and flanging effects
# 5. Compressor     — Dynamic range compression
# 6. Gain           — Volume adjustment (-40 to +40 dB)
# 7. High-Pass      — High-pass filter
# 8. Low-Pass       — Low-pass filter

Effects can be saved as presets or bound to specific voice profiles as defaults.

MCP Integration: Make Your AI Coding Assistant "Speak"

Voicebox has a built-in MCP server, supporting AI tools like Claude Code, Cursor, and Cline to directly access voice features.

Setup:

// Add this to your MCP config:
{
  "mcpServers": {
    "voicebox": {
      "command": "voicebox",
      "args": ["--mcp"]
    }
  }
}

Once configured, your AI assistant can call the voicebox.speak tool to read out code explanations, debug info, and more — in your cloned voice.

Advanced — Voice Personality:

Attach a "personality description" to a voice profile, and Voicebox will use its built-in local LLM to stylize the text:

Example personalities:
"You're an experienced Linux sysadmin. You speak concisely and directly,
with occasional technical humor. You like using analogies when explaining."

AI assistants can also trigger these modes via MCP: Compose, Rewrite, and Respond.

Stories Editor: Multi-Track Voice Narratives

Stories is a multi-track timeline editor, perfect for: - Multi-character dialogues - Podcast episodes - Audiobook chapters - Voice presentations

# Basic workflow:
# 1. Create a Story → Add multiple Tracks (each linked to a voice)
# 2. Generate or drag voice clips onto each track
# 3. Arrange clips on the timeline
# 4. Supports clip trimming and splitting
# 5. Auto-play with synced playhead
# 6. Export as a complete audio file

REST API for Developers

Voicebox provides a full REST API for integrating into your own apps:

# Generate speech
curl -X POST http://localhost:9876/api/generate \
  -H "Content-Type: application/json" \
  -d '{
    "text": "Hello world",
    "engine": "kokoro",
    "voice": "af_heart",
    "speed": 1.0
  }'

# Clone a voice
curl -X POST http://localhost:9876/api/profiles \
  -F "name=my-clone" \
  -F "audio=@reference.wav"

# Speech-to-text
curl -X POST http://localhost:9876/api/transcribe \
  -F "audio=@recording.wav" \
  -F "model=turbo"

# List all voice profiles
curl http://localhost:9876/api/profiles

The API supports SSE (Server-Sent Events) for real-time status updates — generation is non-blocking.

Hardware Requirements

Platform acceleration support: - macOS: MLX / Metal (Apple Silicon native optimization) - Windows: CUDA (NVIDIA), DirectML (AMD/Intel) - Linux: CUDA, ROCm (AMD), Intel Arc

Competitor Comparison

Summary

Voicebox is the most complete open-source local voice solution available today. It combines voice cloning, speech synthesis, global dictation, audio post-processing, and AI integration into a single native app — all free, all private, data never leaves your machine.

Who it's for: - Developers: MCP integration gives your AI coding assistant a voice - Content creators: Generate voiceovers locally, no paid APIs needed - Privacy-conscious users: All data processed locally - Podcast / audiobook makers: Multi-track Stories editor

Project: github.com/jamiepine/voicebox Website: voicebox.sh Docs: docs.voicebox.sh License: Open Source

If you found this guide helpful, feel free to share it with fellow developers. Got questions? Drop a comment below!