An unofficial PyTorch implementation of VALL-E (last updated: 2025.05.30), utilizing the EnCodec encoder/decoder.

A demo is available on HuggingFace here.

Besides a working PyTorch environment, the only hard requirement is espeak-ng for phonemizing text:

• Linux users can consult their package managers on installing espeak/espeak-ng.
• Windows users are required to install espeak-ng.
  • additionally, you may be required to set the PHONEMIZER_ESPEAK_LIBRARY environment variable to specify the path to libespeak-ng.dll.
• In the future, an internal homebrew to replace this would be fantastic.

Simply run pip install git+https://git.ecker.tech/mrq/vall-e or pip install git+https://github.com/e-c-k-e-r/vall-e.

This repo is tested under Python versions 3.10.9, 3.11.3, and 3.12.3.

An "HF"-ified version of the model is available as ecker/vall-e@hf, but it does require some additional efforts (see the __main__ of ./vall_e/models/base.py for details).

Additionally, vall_e.cpp is available. Consult its README for more details.

Pre-trained weights can be acquired from

• here or automatically when either inferencing or running the web UI.
• ./scripts/setup.sh, a script to setup a proper environment and download the weights. This will also automatically create a venv.
• when inferencing, either through the web UI or CLI, if no model is passed, the default model will download automatically instead, and should automatically update.

The provided documentation under ./docs/ should provide thorough coverage over most, if not all, of this project.

Markdown files should correspond directly to their respective file or folder under ./vall_e/.