What is Voicebox?
Voicebox is an open-source AI voice studio that combines voice cloning, speech generation, dictation, and agent speech in a desktop application. Voicebox serves people producing spoken content as well as developers adding voice features to software. The website presents Voicebox as a free, local alternative to ElevenLabs and WisprFlow, with downloads for macOS, Windows, and Linux.
Voicebox runs models on the user's hardware and supports offline use. Voicebox also offers connections to a remote machine for inference rather than requiring every workload to run on the desktop itself. Capy's agents build and maintain Voicebox, and the website links to GitHub alongside the application download.
What Voicebox can do
Voicebox brings voice sampling, speech generation, editing, and transcription into a single studio. Voicebox includes multiple speech engines, with capabilities that differ by model, such as natural-language delivery instructions, preset speakers, and expressive speech tags. Voicebox also supports voice-profile personalities for rewriting supplied text or composing fresh character lines.
- Voicebox clones voices from as little as 3 seconds of audio.
- Voicebox accepts uploaded audio, microphone recordings, and captured system audio.
- Voicebox arranges, trims, and mixes multi-voice content in a timeline editor.
- Voicebox applies pitch shift, reverb, delay, and compression with reusable presets.
- Voicebox transcribes audio with Whisper and extracts reference text from samples.
- Voicebox generates up to 50,000 characters with sentence-based splitting and crossfading.
How Voicebox works
Voicebox lets users obtain a voice sample by uploading a clip, recording through a microphone, or capturing audio playing on the system. Voicebox supports WAV, MP3, FLAC, and WebM uploads. Users can choose downloaded models for speech generation or transcription, while local inference supports Metal, CUDA, ROCm, Intel Arc, and DirectML.
Voicebox desktop dictation uses a held keyboard shortcut to capture speech and place the transcript in a focused text field or the clipboard. Optional local refinement removes filler words, self-corrections, and punctuation problems without rephrasing. For developers, Voicebox exposes REST endpoints and an MCP speech tool that agents such as Claude Code, Cursor, and Cline can call.
Who Voicebox is for
Voicebox addresses audio production, everyday text entry, and software development workflows. Voicebox's Stories Editor supports character conversations and multi-voice narratives, while the API supports applications, games, and automation scripts. Agent users can bind different MCP clients to different voice profiles, making the speaking agent identifiable by its voice.
- Voicebox helps creators produce narration, scripts, dubs, and character dialogue.
- Voicebox supports desktop users who want to dictate into other applications.
- Voicebox lets game developers generate NPC dialogue and localized character voices.
- Voicebox lets developers add narration, accessibility readouts, and voice replies to apps.
- Voicebox supports scripts that batch-generate audiobook chapters or podcast introductions.
Voicebox pricing and access
Voicebox is free and open source, and the website provides download and GitHub links. Voicebox describes local generation as subscription-free and states that the built-in API requires no API keys, imposes no rate limits, and charges no per-character fees. Downloaded models run locally, with the website describing model access as download once, use forever.
Voicebox strengths and limitations
Voicebox combines local audio processing with editing tools and programmatic access, making the same voice profiles useful across content production and agent workflows. Voicebox offers multiple model sizes so users can choose according to hardware and quality needs. However, the supported languages, expressive controls, and computing requirements depend on the selected model rather than being uniform across all engines.
- Voicebox microphone recordings have a maximum duration of 30 seconds.
- Voicebox local model choices depend on the user's hardware.
- Voicebox reserves some controls, such as delivery instructions, for specific engines.
Voicebox FAQ
What is Voicebox?
Voicebox is a free, open-source AI voice studio for voice cloning, speech generation, transcription, and desktop dictation. Voicebox also includes a multi-voice editor, audio effects, and developer endpoints for adding speech to applications and AI agents.
Is Voicebox free?
Voicebox is free and open source. Voicebox runs locally without subscriptions, and the built-in voice API requires no API keys, has no rate limits, and charges no per-character fees. Downloads and a GitHub link are available on the website.
Does Voicebox work offline on macOS, Windows, and Linux?
Voicebox supports macOS, Windows, and Linux and works offline with downloaded local models. Voicebox runs speech generation and transcription on the user's hardware, while also offering the option to connect to a remote machine for inference.
How does Voicebox clone a voice?
Voicebox clones a voice from as little as 3 seconds of audio supplied through a file upload, microphone recording, or system audio capture. Voicebox accepts WAV, MP3, FLAC, and WebM uploads and can extract reference text from samples using Whisper transcription.
Does Voicebox have an API or MCP integration?
Voicebox includes a local REST API and an MCP tool named voicebox.speak. Voicebox supports MCP-aware agents such as Claude Code, Cursor, and Cline, with per-agent voice profiles. A separate POST /speak endpoint supports scripts and other systems that do not use MCP.
Can Voicebox dictate into other apps?
Voicebox captures dictation through a held keyboard shortcut and sends the transcript to a focused text field or the clipboard. Voicebox uses local Whisper models and offers optional local transcript refinement to clean filler words, self-corrections, and punctuation without rephrasing.
From the community
No reviews yet
Be the first to review this tool. What you write helps the next person choose.




