What platforms does Voicebox support?

Voicebox is available on MacOS, Windows and Linux.

What are the limitations of Voicebox?

Output quality and speed depend heavily on the user's GPU; AMD, Intel Arc and CPU-only machines will see slower or weaker results than CUDA setups. Seven engines mean seven downloads, separate strengths and uneven language or emotion support between them, so picking one takes experimentation. Version 0.5.0 signals early software; expect rough edges in a desktop app that is still adding core surfaces like MCP and personalities.

Voicebox

Name: Voicebox
Author: Voicebox

Open-source desktop voice cloning with seven TTS engines

Intermediate API

Visit Voicebox Back to AI tools

Own Voicebox? Verify ownership to get a dofollow link

FeaturedHand-picked by aiwizard

AI Angels400+ customizable AI girlfriends for NSFW chat, voice calls, and roleplay

Candy AIExperience the future of companionship with AI friends who chat, role-play, and connect with you on a human level.

ourdream.aiA vast AI roleplay platform with millions of characters for chat, images, video and voice

Screenshot of Voicebox, Open-source desktop voice cloning with seven TTS engines

What is Voicebox?

Voicebox is a free, open-source desktop app for macOS, Windows and Linux that clones voices from a few seconds of audio and generates speech across seven TTS engines.

Voicebox is an open-source desktop application for voice cloning, speech generation, dictation and agent voice output, built to run entirely on the user's own machine. It positions itself as a free and local alternative to cloud voice platforms such as ElevenLabs and dictation tools such as WisprFlow. The app ships for macOS, Windows and Linux, and the homepage reports version 0.5.0 with more than 3.1 million downloads. Cloning starts from a short sample. A clip can be dragged in as WAV, MP3, FLAC or WebM, recorded from a microphone for up to 30 seconds, or captured directly from system audio, which allows a voice to be pulled from a YouTube video, podcast or any app that is playing sound. From there, seven text to speech engines are available, including Qwen3-TTS, Chatterbox and Chatterbox Turbo from Resemble AI, LuxTTS, Qwen CustomVoice, TADA from Hume AI and Kokoro. They differ in language coverage, speed and hardware appetite: Kokoro is an 82M parameter model at CPU realtime, LuxTTS claims more than 150x realtime on CPU, and Qwen3-TTS is the only engine with natural-language delivery instructions. Whisper handles transcription across five sizes and 99 languages, with optional cleanup by a local Qwen3 model that removes filler words and fixes punctuation without rephrasing. Beyond single lines, Voicebox includes a timeline based Stories editor for multi-voice narratives, an effects pipeline with pitch shift, reverb, delay and compression that can be saved as presets per voice profile, and chunked generation up to 50,000 characters with crossfaded seams. A local REST API on port 17493 exposes generation, profiles, history and health, and an MCP endpoint lets agents such as Claude Code, Cursor and Cline speak through cloned voices. Personalities add a further layer: a profile can be given a character description and then rewrite text in that voice or improvise new lines. It suits podcasters, game developers, audiobook producers, accessibility engineers and developers wiring voice into agents, especially anyone who cannot send audio to a third party. The tradeoff is that quality depends on local hardware, and the experience is closer to a studio tool than a one-click web service.

How do you use Voicebox?

1Download and install the desktop app
Grab the build for macOS, Windows or Linux from the download page and install it. Voicebox runs as a local application rather than in a browser.
2Add a voice sample
Open the voice creation flow and choose Import Voice. Drop in an audio file, record up to 30 seconds from the microphone, or capture system audio from a video or podcast, then trim the clip and save the profile.
3Pick a TTS engine
Download one or more engines from the model catalog. Qwen3-TTS offers delivery instructions, Kokoro is light on hardware, and LuxTTS is fast on CPU. Each engine becomes available to every profile.
4Generate and refine audio
Type or paste text, choose a profile and engine, and generate. Longer text is split at sentence boundaries and crossfaded. Add pitch shift, reverb or compression from the effects pipeline if the voice needs shaping.
5Set up dictation and agent speech
Hold the shortcut (Command plus Option on macOS, Ctrl plus Alt on Windows) anywhere to dictate into the focused field. To let agents talk back, add the local MCP server URL to your agent config and call voicebox.speak with a profile name.
6Review Linux install notes if needed
Linux users should read the dedicated install page before setting up, since GPU drivers and packaging differ between distributions.

Pros and cons

Pros

Voice cloning, generation and transcription all run locally, so voice samples and audio never leave the machineAI
Seven TTS engines cover different needs, from CPU-friendly Kokoro and LuxTTS to long-form TADA and instruction-driven Qwen3-TTSAI
Accepts uploads, microphone recordings up to 30 seconds, and system audio capture, which makes sourcing a sample easyAI
Ships with a documented local REST API and an MCP endpoint, so scripts, games and agents can call it without keys or per-character feesAI
Multi-voice Stories editor, effects pipeline and 50,000 character chunked generation go well past basic text to speechAI

Cons

Output quality and speed depend heavily on the user's GPU; AMD, Intel Arc and CPU-only machines will see slower or weaker results than CUDA setupsAI
Seven engines mean seven downloads, separate strengths and uneven language or emotion support between them, so picking one takes experimentationAI
Version 0.5.0 signals early software; expect rough edges in a desktop app that is still adding core surfaces like MCP and personalitiesAI
The feature set (audio routing, effects chains, REST calls, MCP config) assumes some technical comfort and is not aimed at casual usersAI
No phone apps or browser version, so work is tied to a desktop install and cannot follow the user across devicesAI

How much does Voicebox cost?

Integrations

A local REST API on port 17493 covers generation, cancellation, profiles, model status, history and health. An MCP endpoint at /mcp connects agents such as Claude Code, Cursor and Cline, and a plain POST /speak route serves ACP, A2A, shell scripts and custom harnesses.

Features

Voice cloning from roughly 3 seconds of audio, seven local TTS engines (Qwen3-TTS, Chatterbox, Chatterbox Turbo, LuxTTS, Qwen CustomVoice, TADA and Kokoro), Whisper transcription in five sizes across 99 languages, optional local LLM transcript cleanup, a timeline Stories editor for multi-voice work, an effects pipeline with presets, generation up to 50,000 characters, a local REST API and an MCP endpoint for agent speech.

Frequently asked questions about Voicebox

Does Voicebox have an API?
Yes, Voicebox offers an API.
How do you use Voicebox?
The walkthrough on this page covers 6 steps: 1. Download and install the desktop app 2. Add a voice sample 3. Pick a TTS engine 4. Generate and refine audio 5. Set up dictation and agent speech 6. Review Linux install notes if needed.
What platforms does Voicebox support?
Voicebox is available on MacOS, Windows and Linux.
What does Voicebox integrate with?
A local REST API on port 17493 covers generation, cancellation, profiles, model status, history and health. An MCP endpoint at /mcp connects agents such as Claude Code, Cursor and Cline, and a plain POST /speak route serves ACP, A2A, shell scripts and custom harnesses.
What are the limitations of Voicebox?
Output quality and speed depend heavily on the user's GPU; AMD, Intel Arc and CPU-only machines will see slower or weaker results than CUDA setups. Seven engines mean seven downloads, separate strengths and uneven language or emotion support between them, so picking one takes experimentation. Version 0.5.0 signals early software; expect rough edges in a desktop app that is still adding core surfaces like MCP and personalities.