Voicebox: A Free App for Voice Cloning and Narration Without a Subscription

Voicing a video through ElevenLabs or a similar service costs money every month, and the minute limit runs out faster than you'd like. A separate dictation app is usually paid too. An open-source tool has shown up that covers both jobs at once and runs free on your own computer.

It's called Voicebox. Here's an honest breakdown based on the official repository: what's inside, what's genuinely free, and what's worth checking yourself before building a workflow on top of it.

What Voicebox is

Voicebox is an open-source app for voice cloning, text-to-speech, and dictation. The repository github.com/jamiepine/voicebox had collected over 51,000 stars at the time of checking. The code is open under the MIT license, meaning you're free to use it and see how it works inside.

It runs locally: your voice and audio recordings stay on your computer and don't go to someone else's server for processing. That's a separate advantage if you're narrating someone else's text or personal stories and don't want to hand those recordings to a third party.

What's inside, technically

Voice cloning is built on the Qwen3-TTS model: a short sample recording is enough to get a copy of a voice. Besides that model, the app has six more text-to-speech engines, including Chatterbox and Kokoro, supporting 23 languages in total.

Whisper, from OpenAI, handles speech-to-text; it's the same engine many paid transcription services use. So dictation and transcription are included in the app too; there's nothing extra to buy.

One more detail for anyone working with AI agents: Voicebox has a built-in MCP server. Through it, you can connect the app directly to Claude Code, Cursor, or Windsurf, and have the agent's responses read aloud in the voice you choose, with no manual file exports between apps.

What it runs on

The app is built natively (on Tauri, not Electron, if that matters to you) and is available for macOS, Windows, and Linux, including builds for AMD ROCm and Intel Arc. Mac and Windows have ready-made installers; Linux needs to be built from source, though there's a Docker option.

Synthesis and cloning are computed by the app itself on your hardware, with no cloud subscription. That means speed and quality depend on your computer: on a weaker machine, the process will be noticeably slower than a paid cloud service.

Who this is actually useful for

Anyone who regularly voices videos, podcasts, or Stories and keeps hitting the limits of paid services. Anyone who wants to dictate text by voice instead of typing and isn't ready to pay for a separate app just for that. And separately, anyone already working with Claude Code or similar agents who wants a voice response with no extra steps.

What to check before building a process on this

The app is open-source and very new. I keep an eye on tools like this precisely because voice quality in every video the factory produces is critical, but I personally haven't yet run Voicebox on my own working tasks. Before switching to it fully, it's worth: checking the cloning quality on your own voice and language, not on the demo examples in the repository; making sure the speed on your computer is acceptable; and, just in case, keeping a paid backup option ready for a while.

Separately, about cloning someone else's voice: only do it with that person's consent. The app technically lets you clone a voice from a short recording; what to do with that capability is up to the person, not the tool.

Bottom line

Voicebox covers two paid jobs, speech synthesis and dictation, with one free, open-source, local app. For content work, that's real savings, if you're willing to spend the time once on installation and checking quality against your own voice and computer.

I write about other open-source tools I test for my content system in the channel «Растём с ИИ». If you need help building a system for your own niche, you can start with a free consultation.