Voicebox

Voicebox is an open-source AI voice studio for cloning voices, generating speech, dictating into applications, and giving MCP-aware agents a cloned voice. It runs entirely on the user’s machine and is presented as a local alternative to ElevenLabs and WisprFlow. The desktop app is available for macOS, Windows, and Linux.

Its core functions include voice cloning from as little as three seconds of audio, speech generation across seven TTS engines, and dictation anywhere through a system-wide shortcut. Voicebox can take a sample by uploading an audio file, recording from a microphone, or capturing system audio from another app or from sources such as a YouTube video or podcast. Supported upload formats listed on the site are WAV, MP3, FLAC, and WebM. It also offers a timeline-based Stories Editor for arranging multi-voice narratives, trimming clips, and mixing conversations between characters.

The app includes an audio effects pipeline with pitch shift, reverb, delay, compression, and preset handling, plus live preview and per-profile defaults. It supports local or remote GPU inference with Metal, CUDA, ROCm, Intel Arc, and DirectML, and offers one-click server setup with automatic discovery. Transcription is powered by Whisper, with Base, Small, Medium, Large, and Turbo model sizes shown, and a local language model can refine transcripts by cleaning ums, self-corrections, and punctuation without rephrasing. The site also says generated speech can reach up to 50,000 characters in one go.

Voicebox is available to download and to view on GitHub, and the page includes Donate and Pricing links. It describes itself as free and local, and the featured MCP support lets any MCP-aware agent speak through voicebox.speak in a cloned voice.