Skip to content
Get packs

Guide · Open source

How to use Voicebox: install, clone a voice and generate speech

Updated · written with AI from the project's own documentation and checked against our catalogue data

Voicebox is a free, open-source voice studio that runs on your own computer. It can clone a voice from a short audio sample, read text aloud in that voice, and turn your speech into text for other apps. This guide covers what you need, how to install it, and how to make your first generation.

Voiceboxjamiepine/voicebox · ★ 59.2k · MIT

Quick start

  1. 1Check your system. Confirm you have an Apple Silicon or Intel Mac, or Windows. Linux users must build from source.
  2. 2Install the app. Download the DMG or MSI from the Download table and install it, or run docker compose up with Docker.
  3. 3Pick an engine. Choose the engine that fits your language and hardware, such as Kokoro for fast CPU inference or Qwen3-TTS for cloning.
  4. 4Create a voice profile. Add a reference sample of a few seconds, or use a preset or a text description.
  5. 5Generate speech. Enter your text or script and generate audio. Long scripts are chunked automatically.
  6. 6Set up dictation or agents. Configure the global hotkey for voice input, or connect the MCP server to an agent such as Claude Code or Cursor.

Check whether your computer can run Voicebox

The README lists downloads for Apple Silicon Macs, Intel Macs and Windows, plus a Docker option. Linux users currently need to build from source, because pre-built Linux binaries are not yet available.

The README does not publish full system requirements. It does note that several engines depend on GPU memory or capable CPUs, so check your hardware before you commit to the larger models.

  • Apple Silicon or Intel Mac, or Windows, for the app download
  • Linux: build from source using the instructions at voicebox.sh/linux-install
  • Enough GPU memory or CPU capability for the engine you plan to use

Install Voicebox on macOS or Windows

Download the file that matches your system from the Download table: a DMG for Apple Silicon, a DMG for Intel, or an MSI for Windows. Open it and install the app as you would any other desktop program.

If you prefer Docker, the README lists a single command in its download table: docker compose up. Run it from the project folder that contains the compose file.

  • Apple Silicon Mac: Download DMG from voicebox.sh/download/mac-arm
  • Intel Mac: Download DMG from voicebox.sh/download/mac-intel
  • Windows: Download MSI from voicebox.sh/download/windows
docker compose up

Clone a voice from a short audio sample

Voice cloning in Voicebox is zero-shot: you give a reference sample and the engine reproduces that voice. The README says cloning works from a few seconds of audio. Engines such as Qwen3-TTS and Chatterbox Multilingual support cloning from a reference sample.

If you don't have a sample, some engines work without one. Qwen CustomVoice offers 9 curated preset voices, Kokoro offers 50 curated preset voices, and Qwen VoiceDesign builds a voice from a written description.

  • Use a clean recording of the voice you want to copy
  • Choose an engine that supports cloning, such as Qwen3-TTS or Chatterbox Multilingual
  • Use a preset voice (Kokoro or Qwen CustomVoice) or a text description (Qwen VoiceDesign) if you have no sample

Generate speech and long scripts

Pick a voice profile and an engine, then enter your text to generate speech. You can switch engines per generation. Voicebox supports 23 languages, though the number available depends on the engine you choose.

For long material such as articles or chapters, Voicebox auto-chunks the text and joins the parts with crossfade. The Stories editor provides a multi-track timeline for conversations, podcasts and narratives.

  • Choose a voice profile and engine for each generation
  • Use Chatterbox Turbo for paralinguistic tags such as [laugh], [sigh] and [gasp]
  • Use post-processing effects such as pitch shift, reverb, delay, chorus, compression and filters

Dictate into any app with voice input

Voicebox includes a global dictation hotkey with push-to-talk and toggle modes. It uses Whisper-based speech-to-text. On macOS, the README describes accessibility-verified auto-paste. There is also an in-app microphone for every text field.

Dictation runs on your machine, so your captures stay local.

  • Push-to-talk mode: hold the hotkey while you speak
  • Toggle mode: press once to start and again to stop
  • Auto-paste is accessibility-verified on macOS

Use Voicebox with the REST API or an MCP agent

Developers can add local voice input or output to their own apps through the REST API. Voicebox also includes a built-in MCP server. With MCP-aware agents such as Claude Code, Cursor or Cline, the voicebox_speak tool call lets the agent speak in a voice you have cloned.

Voice personalities let you attach a free-form persona to a voice profile, then use Compose, Rewrite or Respond through a bundled local LLM. Agents can call the same modes over MCP.

  • REST API for integrating voice I/O into your own apps
  • MCP server for agents, using the voicebox_speak tool
  • Voice personalities for Compose, Rewrite and Respond modes

What people use it for

  • Narrating articles, chapters or audiobook-style scripts locally
  • Producing podcast or conversation audio with the Stories timeline
  • Dictating text into any app with a global hotkey
  • Letting an MCP-aware coding agent speak results aloud in a chosen voice
  • Adding local voice input or output to your own app through the REST API

Frequently asked questions

Is Voicebox free?

The README describes Voicebox as a free and open-source app released under the MIT license. It says the app has no splash screens, banners or interruptions.

How do I install Voicebox on Linux?

The README says pre-built Linux binaries are not yet available. Linux users need to follow the build-from-source instructions at voicebox.sh/linux-install.

Does Voicebox send my audio to the cloud?

The README says models, voice data and captures stay on your machine and never leave it. Voicebox runs locally.

How much audio do I need to clone a voice?

The README says Voicebox clones voices from a few seconds of audio. Some engines, such as Kokoro and Qwen CustomVoice, use preset voices instead and need no sample.

What are the alternatives to Voicebox?

The project's comparison points to ElevenLabs and WisprFlow, which are cloud services, plus Coqui TTS and Piper as open-source options. Voicebox combines text-to-speech and dictation in one local app.

Why is my generation slow or failing?

The README warns that several engines depend on GPU memory or capable CPUs and does not give full system requirements. It points to the Troubleshooting Guide in docs/content/docs/overview/troubleshooting.mdx for install, generation, model-download and GPU issues.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Voicebox suits people who want local text-to-speech, voice cloning, dictation or agent voice output without sending audio to cloud services, and who have capable hardware. It is a poor fit if you need a simple Linux install today, have limited hardware, or only want one narrow feature rather than a multi-engine studio.

Install options & downloads