title	labels	dataset
LatentDiffSep	Source Separation Diffusion Models	WSJ0-2mix WSJ0-3mix Libri2mix Libri3mix

Latent Diffusion Model for Source Separation

Master Thesis Project at University of Cambridge.

**Contributor: ** Eduard Burlacu \

Abstract:

About this project

What's implemented: Source code used for producing the results in _____ paper.

Datasets: Libri2Mix, WSJ0-2mix *

Hardware Setup: These experiments were run on ___

Contributor: Eduard Burlacu

Environment Setup

To construct the Python environment follow these steps:

#Setup source separation env
conda env create -f env/environment.yaml

Experiments

Training the OobleckVAE:

We use the StabilityAI's stable-audio-tools to train an OobleckVAE specially-designed for source separation, being able to encode and reconstruct multi-speaker audio samples.

Useful for these Tasks: Blind Source Separation, Speech Enhancement, Target Speaker Extraction

Models:

Datasets: The settings are as follows:

Dataset	#speakers	target method	SI-SDR