VoiceCraft
Token infilling neural codec model for zero-shot speech editing and TTS.
Zero-shot speech editing and text-to-speech for in-the-wild audio data.
VoiceCraft is a token infilling neural codec language model designed for zero-shot speech editing and text-to-speech tasks. It suits audio engineers and game developers who need to clone or edit unseen voices using only a few seconds of reference audio.
What it is
The model performs speech editing and zero-shot text-to-speech on in-the-wild audio such as internet videos, podcasts, and audiobooks. It utilizes a token infilling approach and provides enhanced 330M and 830M model weights via Hugging Face.
Who it's for
It suits developers and audio professionals working on procedural dialogue, voice replacement, and audio post-production. It does not suit teams looking for a simple plug-and-play UI without handling Python dependencies or Docker containers.
Key features
Token Infilling Architecture
Employs a neural codec language model for precise speech editing.
Zero-Shot TTS
Clones or edits unseen voices using only a few seconds of reference audio.
Flexible Inference Options
Run via Gradio in Colab, Docker, local environment setup, or command line.
In-The-Wild Adaptation
Trained to handle messy real-world audio sources like internet videos and podcasts.
Up and running in 3 steps
Our honest verdict
What's great
- Requires only a few seconds of reference audio to clone or edit a voice.
- Performs well on messy in-the-wild data like podcasts and internet videos.
- Multiple running options including Docker, command line scripts, and Colab notebooks.
- Provides pre-trained 330M and 830M TTS enhanced model weights.
Where it falls short
- Requires setting up Python environments or Docker images to run locally.
- Model training and fine-tuning demand significant technical skill and hardware resources.
- License is not stated in the repository metadata, creating ambiguity for commercial use.
Pricing & license
The software is free to use, though the exact license is not stated in the repository. The only cost is running it yourself on your own hardware or cloud infrastructure.
Open remote AI-training roles
870 open remote AI-training roles, updated November 14, 2026; most pay $20–$170 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
What audio inputs does VoiceCraft support for voice cloning?
It requires only a few seconds of reference audio from sources like audiobooks, podcasts, or internet videos.
How can I run inference locally?
You can run it using Docker or by setting up the Python environment locally and using the provided Jupyter notebooks.
