- Home
- Text-to-Speech Software
- VoiceCraft
On the recordRead 10 Oct 202622 facts
VoiceCraft
On the record
- API access
- Nogithub.com · read 20 Sep 2026
- Code license
- The codebase is licensed under CC BY-NC-SA 4.0, with some identified components under separate licenses.github.com · read 9 Oct 2026
- Commercial use
- Nogithub.com · read 20 Sep 2026
- Compute requirement
- The Docker instructions assume Docker with NVIDIA Container Toolkit and pass GPUs into the container.github.com · read 10 Oct 2026
- Consent guidance
- The project says people must not use the technology to generate or edit someone's speech without that person's consent.github.com · read 9 Oct 2026
- Dataset access
- The training instructions say downloading GigaSpeech requires signing an agreement and using an authentication token.github.com · read 10 Oct 2026
- Editing and TTS
- The Gradio interface offers zero-shot TTS, speech editing, and long-text TTS modes.github.com · read 10 Oct 2026
- Hardware requirements
- The Docker instructions assume Docker and NVIDIA container support, and the examples check GPU visibility with nvidia-smi.github.com · read 9 Oct 2026
- Integrations
- The repository provides scripts for integrating VoiceCraft into other projects and documents demos on Hugging Face Spaces and Replicate.github.com · read 10 Oct 2026
- Intended users
- The repository documents inference use as well as model training and fine-tuning for developers preparing their own datasets.github.com · read 10 Oct 2026
- License
- The repository states that the code is under CC BY-NC-SA 4.0 and the model weights are under the Coqui Public Model License 1.0.0.github.com · read 10 Oct 2026
- Model length limit
- For the giga330M TTS model described in its news section, the repository says maximum prompt plus generation length should be 16 seconds.github.com · read 10 Oct 2026
- Model license
- The model weights use Coqui Public Model License 1.0.0, which permits only non-commercial use of the model and its outputs.github.com · read 9 Oct 2026
- Model limits
- The README says the enhanced TTS models were trained with utterances no longer than 16 seconds and recommends keeping prompt plus generation length at or below 16 seconds for those models.github.com · read 9 Oct 2026
- Purpose
- VoiceCraft is a token-infilling neural codec language model for zero-shot speech editing and text-to-speech on in-the-wild data such as audiobooks, internet videos, and podcasts.github.com · read 10 Oct 2026
- Training
- Training requires utterances and transcripts, encoded audio codes, phoneme sequences, and a manifest; the project includes preparation guidance.github.com · read 10 Oct 2026
- Transcript control
- Its smart transcript feature lets users write only the text they want to generate.github.com · read 10 Oct 2026
- Voice cloning
- To clone or edit an unseen voice, VoiceCraft needs only a few seconds of reference audio.github.com · read 9 Oct 2026
