Voicebox: The Open-Source AI Voice Studio — 43,562 GitHub Stars

Quick Answer

Voicebox is a local-first AI voice studio — a free and open-source alternative to ElevenLabs and WisprFlow in one app. With 43,562 GitHub stars and 5,315 forks, it's the fastest-growing open-source voice AI project. Voicebox lets you clone voices from a few seconds of audio, generate speech in 23 languages across 7 TTS engines, dictate into any text field with a global hotkey, and give any MCP-aware AI agent a voice of your choosing. Built with Tauri (Rust) for native performance, everything runs locally on your machine — models, voice data, and captures never leave your device. Use Voicebox if you need: A privacy-focused voice AI tool for voice cloning, text-to-speech generation, dictation, or giving AI coding agents (Claude Code, Cursor, Cline) a custom voice.

---

The Problem: Why Voicebox Exists

Voice AI tools have exploded in popularity, but they come with significant drawbacks:

• Privacy concerns: Your voice data and audio are processed on third-party servers
• Fragmented tools: ElevenLabs handles output (TTS), WisprFlow handles input (STT) — you need both
• Cost: Per-character or per-minute pricing adds up quickly for heavy users
• Vendor lock-in: Proprietary formats, no way to self-host or customize
• Platform limitations: Most tools are cloud-only, no offline capability

Voicebox solves these problems with an open-source, local-first approach. The project launched in late 2025 and exploded in popularity, reaching 43,562 stars in under a year — a testament to the demand for privacy-preserving voice AI tools.

---

Key Features

1. Complete Privacy — Everything Runs Locally

Unlike cloud-based alternatives, Voicebox processes everything on your machine. Models, voice data, and audio captures never leave your device. This matters for:

• Corporate content: Sensitive product demos, internal training materials
• Client work: NDA-protected voiceovers and presentations
• Personal content: Private voice clones and recordings
• Regulatory compliance: GDPR, HIPAA, and other data protection requirements

2. Seven TTS Engines — Switch Per Generation

Voicebox ships with seven different text-to-speech engines, each with unique strengths:

Engine	Languages	Strengths
Qwen3-TTS (0.6B / 1.7B)	10	High-quality multilingual cloning, delivery instructions ("speak slowly", "whisper")
Qwen CustomVoice	10	9 curated preset voices with natural-language delivery control
LuxTTS	English	Lightweight (~1GB VRAM), 48kHz output, 150x realtime on CPU
Chatterbox Multilingual	23	Broadest language coverage — Arabic, Turkish, Hindi, Swahili, and more
Chatterbox Turbo	English	Fast 350M model with paralinguistic emotion/sound tags
TADA (1B / 3B)	10	HumeAI speech-language model — 700s+ coherent audio
Kokoro	8	50 curated preset voices, tiny 82M model, fast CPU inference

You can switch engines per generation, choosing the best one for your use case. Need Turkish? Use Chatterbox Multilingual. Need emotional tags like [laugh] and [sigh]? Use Chatterbox Turbo.

3. Voice Cloning & Preset Voices

Clone any voice from a few seconds of audio:

• Zero-shot cloning: Upload a reference sample, get a voice profile in seconds
• Multi-sample support: Higher quality cloning with multiple audio samples
• 50+ preset voices: Curated voices via Kokoro and Qwen CustomVoice (no cloning required)
• Import/export profiles: Share or back up voice profiles
• Per-profile defaults: Set default effects chains for each voice

4. 23 Languages

From English to Arabic, Japanese, Hindi, Swahili, and more. Voicebox supports:

Arabic, Danish, Dutch, English, Finnish, French, German, Greek, Hebrew, Hindi, Italian, Japanese, Korean, Malay, Mandarin, Norwegian, Polish, Portuguese, Russian, Spanish, Swahili, Swedish, Turkish

This makes it ideal for multilingual content creation, localization, and accessibility.

5. Post-Processing Effects

Apply professional audio effects after generation:

• Pitch Shift: Up or down by up to 12 semitones
• Reverb: Configurable room size, damping, wet/dry mix
• Delay: Echo with adjustable time, feedback, and mix
• Chorus / Flanger: Modulated delay for metallic or lush textures
• Compressor: Dynamic range compression
• Gain: Volume adjustment (-40 to +40 dB)
• High-Pass Filter: Remove low frequencies
• Low-Pass Filter: Remove high frequencies

Ships with 4 built-in presets (Robotic, Radio, Echo Chamber, Deep Voice) and supports custom presets.

6. Expressive Speech — Paralinguistic Tags

With Chatterbox Turbo selected, add emotional and paralinguistic tags inline:

[laugh] [chuckle] [gasp] [cough] [sigh] [groan] [sniff] [shush] [clear throat]

Type / in the text input to open the tag inserter. This enables natural-sounding speech with emotional nuance — perfect for audiobooks, podcasts, and character voices.

7. Unlimited Length — Auto-Chunking with Crossfade

Generate speech for long-form content:

• Configurable auto-chunking: 100–5,000 characters per chunk
• Crossfade slider: 0–200ms for smooth transitions between chunks
• Max text length: 50,000 characters
• Smart splitting: Respects abbreviations, CJK punctuation, and [tags]

Perfect for scripts, articles, book chapters, and long-form narration.

8. Stories Editor — Multi-Track Timeline

Create conversations, podcasts, and narratives with multiple voices:

• Multi-track composition: Drag-and-drop clips onto a timeline
• Inline audio trimming: Split and trim clips directly in the editor
• Auto-playback: Synchronized playhead for preview
• Version pinning: Pin specific generation versions to each clip

9. Voice Input — Global Dictation Hotkey

The other half of the voice I/O loop. Hold a hotkey anywhere on your system, speak, release — on macOS the transcript pastes straight into the focused text field.

• Configurable chord bindings: Hold-to-speak and tap-to-toggle modes
• Target-aware paste (macOS): Accessibility-verified injection into the focused text field
• First-run permissions UX: In-app guides for macOS Accessibility and Input Monitoring grants
• In-app mic button: Dictate directly into any Voicebox text field
• LLM refinement: Optional cleanup of ums, stutters, and false starts before paste
• On-screen pill: Floating overlay showing recording, transcribing, refining, and speaking states

Runs OpenAI Whisper for transcription (Base / Small / Medium / Large / Turbo). Same model that backs dictation, the Captures tab, and the /transcribe API.

10. Agent Voice Output — MCP Integration

Every agent gets a voice. One tool call and any MCP-aware agent can speak to you in a voice you've cloned:

// In any MCP-aware agent (Claude Code, Cursor, Cline):
await voicebox.speak({
  text: "Deploy complete.",
  profile: "Morgan",
});

Also exposed as POST /speak for anything that doesn't speak MCP — ACP, A2A, shell scripts, custom harnesses.

Key features:

• Bidirectional pill: Recording, transcribing, refining, and speaking are all states of the same OS-level overlay
• Per-agent voice binding: Pin Claude Code to Morgan and Cursor to Scarlett in Settings → MCP
• Always visible: No silent background TTS; every agent-initiated speak surfaces the pill
• HTTP + stdio transports: Install as a URL in Claude Code / Cursor / Windsurf / VS Code MCP, or point stdio-only clients at the bundled voicebox-mcp binary

11. Voice Personalities — Local LLM Integration

Attach a free-form personality to any voice profile — who this voice is, how they speak, what they care about. Two actions appear when a personality is set:

• Compose: A shuffle button that drops a fresh in-character line into the textarea
• Speak in character: A toggle that routes your input text through the personality LLM to be rewritten in-character

Powered by a bundled Qwen3 LLM running entirely locally. Agents can invoke the same modes over MCP.

12. API-First Architecture

Voicebox exposes a REST API plus a built-in MCP server for integrating voice I/O into your own apps and agents:

# Start the API server
voicebox serve

# REST API available at http://127.0.0.1:3333
curl -X POST http://127.0.0.1:3333/v1/generate \
  -H 'Content-Type: application/json' \
  -d '{
    "text": "Hello world",
    "profile": "Morgan",
    "engine": "qwen3-tts"
  }'

Full OpenAPI 3.1 contract available at /openapi.json.

---

Technical Architecture

Tauri (Rust) — Native Performance

Voicebox is built with Tauri, not Electron. This provides:

• Smaller binary size: ~50MB vs Electron's ~150MB+
• Lower memory usage: Rust backend, no Chromium overhead
• GPU acceleration: Direct access to Metal (macOS), CUDA (Windows), ROCm (Linux)
• Security: Sandboxed webview, minimal attack surface
• Cross-platform: Same codebase compiles to macOS, Windows, Linux

Platform-Specific Optimization

macOS (Apple Silicon):

• MLX acceleration for Whisper and TTS models
• Metal GPU backend for model inference
• Accessibility-verified paste for global dictation

Windows:

• CUDA acceleration for NVIDIA GPUs
• DirectML support for AMD and Intel GPUs

Linux:

• ROCm support for AMD GPUs
• CPU fallback for systems without discrete GPUs

Docker:

• Containerized deployment for server environments
• docker compose up for quick setup

Generation Pipeline

Voicebox uses a sophisticated generation pipeline:

1. Text preprocessing: Sentence boundary detection, abbreviation handling, tag parsing
2. Chunking: Split text into chunks based on configurable limit
3. Inference: Generate audio for each chunk using selected engine
4. Crossfade: Smooth transitions between chunks
5. Post-processing: Apply effects chain (pitch, reverb, delay, etc.)
6. Version tracking: Store original, effects versions, and takes with provenance

Generation is non-blocking — submit and immediately start typing the next one. Serial execution queue prevents GPU contention. Real-time SSE status streaming. Failed generations can be retried. Stale generations from crashes auto-recover on startup.

---

Voicebox vs ElevenLabs vs WisprFlow

Feature	Voicebox	ElevenLabs	WisprFlow
Price	Free (open-source)	Freemium (per-character)	Freemium (per-minute)
Privacy	Local processing	Cloud-based	Cloud-based
Voice cloning	✅ (unlimited)	✅ (limited by tier)	❌
Text-to-speech	✅ (7 engines)	✅ (1 engine)	❌
Dictation (STT)	✅ (Whisper)	❌	✅
Languages	23	29	1 (English)
MCP integration	✅ (built-in)	❌	❌
Agent voice output	✅	❌	❌
Offline support	✅	❌	❌

When to Choose Voicebox

✅ Choose Voicebox if:

• Privacy is critical (corporate, client, or personal content)
• You need both TTS and STT in one tool
• You want to clone voices without subscription fees
• You're building AI agent workflows with MCP
• You need offline support
• You want to self-host or customize

When to Choose ElevenLabs

✅ Choose ElevenLabs if:

• You need the absolute highest quality TTS (their models are state-of-the-art)
• You don't want to manage local models and GPU requirements
• You need 29+ languages (Voicebox has 23)
• You're okay with cloud processing and per-character pricing

---

Installation & Setup Guide

Option 1: Download Pre-Built Binaries (Easiest)

macOS (Apple Silicon): Download DMG from voicebox.sh

macOS (Intel): Download Intel DMG from voicebox.sh

Windows: Download MSI from voicebox.sh

Docker:

git clone https://github.com/jamiepine/voicebox.git
cd voicebox
docker compose up

Linux: Pre-built binaries not yet available. See build-from-source instructions at voicebox.sh/linux-install.

Option 2: Install MCP Server for AI Agents

Claude Code:

# In Claude Code, add Voicebox MCP server
claude mcp add voicebox -- npx voicebox-mcp

Cursor: Add to your MCP config:

{
  "mcpServers": {
    "voicebox": {
      "command": "npx",
      "args": ["voicebox-mcp"]
    }
  }
}

Other MCP clients: Point to the bundled voicebox-mcp binary or use the HTTP transport at http://127.0.0.1:3333/mcp.

---

Use Cases

1. Content Creation

Podcast production:

• Clone your voice for consistent narration
• Use multi-track editor for interviews and conversations
• Apply effects for professional sound quality

Video narration:

• Generate voiceovers in multiple languages
• Use expressive tags for emotional nuance
• Auto-chunking for long scripts

Audiobooks:

• Clone author's voice for narration
• Unlimited length generation for chapters
• Stories editor for character voices

2. Accessibility

Screen reader integration:

• Clone a familiar voice (family member, caregiver)
• Use as custom voice for screen readers via API

Dictation for disabilities:

• Global hotkey for hands-free text input
• Whisper-based STT with LLM refinement

3. AI Agent Workflows

Coding assistants:

• Give Claude Code a custom voice for task completions
• Per-agent voice binding (different voices for different agents)
• Always-visible pill for agent speech

Customer support bots:

• Clone brand voice for consistent customer interactions
• Integrate with support ticket systems via API

4. Localization

Multilingual content:

• Generate speech in 23 languages from one text source
• Clone voice across languages (zero-shot multilingual cloning)

Language learning:

• Generate pronunciation examples in target language
• Dictation practice with instant feedback

---

Performance Benchmarks

Inference Speed

LuxTTS (English, lightweight):

• 150x realtime on CPU (1 second of audio generated in 6.7ms)
• ~1GB VRAM usage

Chatterbox Turbo (English, 350M model):

• 50x realtime on GPU
• Supports paralinguistic tags

Qwen3-TTS (1.7B model):

• 20x realtime on GPU
• High-quality multilingual cloning

Kokoro (82M model):

• Fast CPU inference
• 50 preset voices

Quality Metrics

Voicebox uses subjective quality metrics (MOS - Mean Opinion Score) for evaluation:

• Voice cloning similarity: 4.2/5 (zero-shot, 10-second reference)
• Naturalness: 4.5/5 (Qwen3-TTS)
• Intelligibility: 4.8/5 (all engines)

These are comparable to or exceed cloud-based alternatives like ElevenLabs.

---

FAQ

1. Is Voicebox really free?

Yes, Voicebox is 100% free and open-source. You can use it for personal or commercial projects without restrictions. There are no premium tiers or paywalled features. The only costs are your own GPU/compute resources.

2. What GPU do I need to run Voicebox?

It depends on the TTS engine:

• LuxTTS: ~1GB VRAM (runs on CPU at 150x realtime)
• Kokoro: 82M model, fast CPU inference
• Qwen3-TTS (0.6B): ~2GB VRAM
• Qwen3-TTS (1.7B): ~4GB VRAM
• Chatterbox Turbo (350M): ~2GB VRAM

For best performance, a discrete GPU with 4GB+ VRAM is recommended. However, lightweight engines like LuxTTS and Kokoro work well on CPU.

3. Can I use Voicebox for commercial projects?

Yes, Voicebox is open-source and can be used for commercial projects. However, be aware that some TTS engines (like Qwen models) may have their own licensing terms. Check each engine's license before commercial use.

4. How accurate is voice cloning?

Voice cloning quality depends on reference audio quality and length. Voicebox achieves 4.2/5 MOS (Mean Opinion Score) for voice similarity, which is comparable to cloud-based alternatives.

5. Does Voicebox work offline?

Yes, once models are downloaded, Voicebox works completely offline. Models, voice data, and audio captures never leave your device.

6. Can I use Voicebox with AI coding agents?

Yes! Voicebox has built-in MCP (Model Context Protocol) server support. Any MCP-aware agent (Claude Code, Cursor, Cline, Windsurf) can use Voicebox to speak task completions, questions, and notifications in a cloned voice.

7. How does Voicebox compare to ElevenLabs?

Voicebox advantages: Free, local processing (privacy), 7 TTS engines, dictation included, MCP integration, offline support.

ElevenLabs advantages: State-of-the-art TTS quality, 29+ languages, no GPU requirements, managed service.

Choose Voicebox if privacy, cost, and agent integration matter. Choose ElevenLabs if you need the absolute highest TTS quality.

---

Learn More

Interested in building tools like Voicebox? Check out these courses on CoddyKit:

• AI with Python: Master AI development with Python — the language powering Voicebox's ML models and Whisper STT.
• AI Agents: Learn to build AI agents with MCP integration — the same protocol Voicebox uses to give coding assistants a voice.
• TypeScript: Learn TypeScript for building type-safe applications like Voicebox's Tauri frontend.
• Rust: Deep dive into Rust — the language powering Voicebox's high-performance Tauri backend.

---

Resources

• GitHub: jamiepine/voicebox
• Website: voicebox.sh
• Documentation: docs.voicebox.sh
• Troubleshooting: Troubleshooting Guide

---

Star count verified on July 20, 2026. Voicebox is the fastest-growing open-source voice AI project on GitHub.