Practical overview
When it fits
Use it to explore diffusion-based music synthesis, try pretrained model variants, or build a training workflow around audio datasets and mel spectrograms.
Who it is for
- People exploring AI-generated music loops using pretrained audio diffusion models.
- Developers and researchers who want to train audio diffusion models on audio-file datasets.
- Creators experimenting with audio variations, interpolation, remixing or masked sample stitching.
Documented uses
- Generate music loops from pretrained diffusion models.
- Train a diffusion model on mel spectrograms generated from a directory of audio files.
- Create audio variations or remix-like outputs based on an existing sample.
- Interpolate between audio samples in latent noise space or mask sections to stitch samples.
- Train conditional models using an encoding such as text or audio.
Requirements & limits
The repository is public and identifies its license as GPL-3.0. The README links to a Google Colab notebook and a Hugging Face Space for trying pretrained models.
- The workflow converts audio to mel spectrograms and back; the README states that higher spectrogram resolution preserves more audio information.
- Training with 64×64 mel spectrograms is described as possible on a single commercial-grade GPU, with an RTX 2080 Ti given as an example.
- The project provides pretrained standard, latent, DDIM and conditional latent model variants.
- Updated pretrained models that save mel configuration are not compatible with earlier repository versions; the README points to migration guidance for previously trained models.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Install Audio Diffusion from the GitHub repository
Install the project package from its repository, which includes training scripts.
Before you start
- Git
- Python package installation with pip
Steps
Clone https://github.com/teticio/audio-diffusion.git.
Change into the audio-diffusion directory.
Run pip install .
Expected outcome
The Audio Diffusion package is installed from the repository, including its training scripts.
Guide limitations
- These installation steps do not describe running the application or training a model.
- Training has separate GPU and mel-spectrogram considerations described in the README.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 6 November 2026
- Original website title
- teticio/audio-diffusion · Official homepage title · retained in its source language