Skip to content

LLaSA_training

Train LLaMA-based speech synthesis models

Visit official site

What LLaSA_training is

The project is described as a LLaMA-based speech synthesis training effort that improves efficiency by optimising compute in training and inference.

What you can do with it

  • Train LLaMA-based speech synthesis models
  • Work with open datasets such as LibriHeavy and Emilia
  • Use distributed training through Slurm

Getting started

  • Clone the repository to a local machine
  • Download the open datasets or prepare your own
  • Choose a config file and run the training script

Pros & cons

✓ What we like

  • Several training config files are provided
  • Total open data reaches 160,000 hours

! What to watch out for

  • Training uses both open and internal datasets
  • Config files and a training script are part of setup

FAQ

What datasets are named?

LibriHeavy and Emilia are listed among the open datasets.

Which models are on Hugging Face?

The material names Llasa-3B, Llasa-1B and Llasa-8B.

Last reviewed: 2026-11-15

More AI audio tools tools

View all →

How we review