Voicebox: The Open-Source Voice Studio That Replaces ElevenLabs and WisprFlow on Your Own Machine

Voicebox packages voice cloning, speech generation, and dictation into one local desktop app, so the paid features of ElevenLabs and WisprFlow run for free on your own hardware. The video walks through the GitHub repo jamiepine/voicebox and its landing page at voicebox.sh, showing a project that trends near the top of GitHub with tens of thousands of stars. On screen the app clones and generates speech across 23 languages, dictates into any app through a global hotkey, and gives AI agents a chosen voice. It positions itself as a local-first alternative that keeps voice data off the cloud.

What Voicebox Actually Does

The repo README shown in the video describes Voicebox as a "local-first AI voice studio," a free and open-source alternative to ElevenLabs and WisprFlow in one app. The framing is specific: the two cloud incumbents sit on opposite halves of the voice loop. ElevenLabs handles output, generating speech. WisprFlow handles input, turning your speech into text. Voicebox does both sides and bridges them with a bundled local LLM for refinement and per-profile personas.

The feature list on screen is concrete. 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. Give any MCP-aware AI agent a voice you own. The README names the engines: Qwen3-TTS, Qwen CustomVoice, LuxTTS, Chatterbox, and HumeAI among them.

The Cost Argument

The pitch in the video is a direct cost comparison. Cloning and generating speech at this level normally requires a paid subscription to ElevenLabs or WisprFlow. Voicebox runs the same class of work locally for free. The landing page frame states it plainly: "A free and local alternative to ElevenLabs and WisprFlow, running entirely on your machine."

Running locally also changes the privacy picture. The README bullet shown on screen reads "Complete privacy — models, voice data, and captures never leave your machine." For anyone cloning their own voice or dictating sensitive text, that is the practical difference from a cloud service.

Dictation Through a Global Hotkey

One frame draws the comparison to WisprFlow directly. Voicebox has its own hotkey for speech to text, so you can dictate verbally into any app you choose. You hold a customizable key chord anywhere on the machine, speak, release, and the transcribed text drops into whatever field you were in. This is the input half of the voice loop, and it is the piece that competes with WisprFlow rather than ElevenLabs.

The Voice Studio Interface

The desktop app frames show a library of voice profiles built for demonstration: Jarvis, Samuel L. Jackson, Bob Ross, Sam Altman, Morgan Freeman, Linus Tech Tips, Fireship, Scarlett Johansson, and Dario Amodei. Each profile carries a short persona description, a language tag, and a model such as Qwen 1.7B or Chatterbox. A generation panel takes typed text, and a history feed on the right lists recent renders with timestamps and durations. The interface makes the "per-profile personas" claim tangible: each voice is a saved profile you can generate against on demand.

Repo Signals Worth Noting

The GitHub page shown in the video reports roughly 45,000 stars, 5.5k forks, and an MIT license, with a "#1 Repository Of The Day" GitHub Trending badge on the README. The commit history is active, with changes to the app, backend, docs, and Tauri directories dated within the last day. The topic tags include ai, cuda, whisper, mlx, voice-ai, voice-clone, and qwen3-tts, which point at the underlying stack. One note on the numbers: the narrator says 33,000 stars, but the repo page on screen shows about 45,000, so the count had already grown past the figure in the script.

Key Takeaways

Voicebox is an open-source, MIT-licensed desktop app that runs voice cloning, speech generation, and dictation locally.
It targets both ElevenLabs (speech output) and WisprFlow (speech input) in a single tool.
It generates speech in 23 languages across 7 TTS engines, including Qwen3-TTS, Chatterbox, and HumeAI.
A global hotkey lets you dictate into any application, and MCP support lets AI agents speak in a chosen voice.
Everything runs on your machine, so models, voice data, and captures stay local.
The repo jamiepine/voicebox had trended to #1 on GitHub with around 45,000 stars by the time of the video.