Voicebox

Voicebox combines speech creation and dictation for people producing audio or working with AI agents. It's a free, open-source alternative to ElevenLabs and WisprFlow that processes voice data on your own machine.

• Voice cloning and speech generation: Create voices from short recordings or use presets with Qwen3-TTS, Qwen CustomVoice, LuxTTS, Chatterbox Multilingual, Chatterbox Turbo, HumeAI TADA and Kokoro; language support depends on the engine.
• Dictation and transcription: Whisper transcribes recordings and hotkey dictation, with automatic insertion into the focused text field on macOS and optional local LLM cleanup of fillers and false starts.
• Audio production: Arrange multiple voices in the Stories timeline, trim and split clips, generate long scripts, and apply effects such as pitch shift, reverb and compression while preserving the original audio.
• Agent voices and personalities: Give Claude Code, Cursor or Cline a cloned voice through MCP, and use a bundled Qwen3 local LLM to compose or rewrite speech according to a voice profile's persona; a REST API connects custom apps.
• Recording library: Keep audio alongside its transcript, edit or re-transcribe captures, and reuse recordings as voice samples.

Voicebox uses the MIT license and runs on macOS, Windows, Linux or Docker. Models, voice data and captures stay on your machine. It supports CPU inference and acceleration through Apple Silicon, NVIDIA CUDA, AMD ROCm and Intel Arc. Linux requires a source build because prebuilt desktop binaries aren't available.

Speech controls vary by engine: Chatterbox Turbo interprets laughter and sigh tags, while Qwen CustomVoice accepts written delivery instructions. Automatic dictation paste into other apps is available on macOS.