Skip to content

Audio-SDS

Apply score distillation to audio diffusion models

Visit official site

What Audio-SDS is

Audio-SDS is described as a framework applying Score Distillation Sampling to audio diffusion models, using large pretrained models for several audio tasks.

What you can do with it

  • Separate mixed audio into independent sources
  • Synthesise impact sounds from physical models
  • Tune FM synthesis parameters for a timbre
  • Render audio in real time from a prompt

Application scenarios

  • Split vocals and background music apart
  • Make impact sounds for game and film effects
  • Adjust a synthesiser in music production

Getting started

  • Open the official site for docs and examples
  • Prepare mixed audio and define source prompts
  • Feed the audio into the model with parameters
  • Run the model and wait for separated audio

Pros & cons

✓ What we like

  • Real-time rendering creates audio from a prompt
  • Instruments and ambient sounds are both covered

! What to watch out for

  • Tasks run through iterative optimisation
  • Separation depends on a defined source prompt

FAQ

What is the core idea?

The material states Score Distillation Sampling is applied to audio diffusion models.

Is a dataset needed?

The material states no dedicated training dataset is needed and pretrained models are used directly.

Last reviewed: 2026-11-15

More AI audio tools tools

View all →

How we review