VoiceCraft
Local speech editing and text-to-speech software clones voices from a few seconds of reference audio and offers a Gradio interface and Docker support.
VoiceCraft generates speech from text and edits existing recordings using a few seconds of a speaker's voice as reference. It's for researchers and developers who want to work with voice generation on their own hardware without training a separate model for each speaker.
- Voice cloning and text-to-speech: Generate speech in a reference speaker's voice, including voices the model hasn't encountered before.
- Speech editing: Change speech within an existing recording using reference audio from the speaker.
- Long text generation: A dedicated Gradio mode handles longer passages and lets you regenerate individual parts.
- Browser interface: The local Gradio interface provides generation and editing modes, with a smart transcript feature that accepts only the text you want to generate.
- Custom models and integration: Standalone scripts support use in other projects, and training and fine-tuning code lets researchers work with their own audio datasets.
VoiceCraft runs locally with or without Docker, and the Docker setup supports Linux and Windows. Google Colab provides a cloud option for speech editing and text-to-speech; those sessions run outside your machine. The project prohibits generating or editing someone's speech without their consent.