Exploring how multichannel recordings can be used to estimate source distance and separate sounds based on spatial proximity rather than source type.

Various methods from optimisation and learning based models were introduced for the task of interference or bleeding reduction in live multi-track recordings.

Developing source separation models tailored for Indian Art Music (Carnatic), addressing microphone bleed and complex harmonic–percussive interactions to extract instruments such as vocals, mridangam, violin, and ghatam.

A deep convolutional neural network-based architecture is trained to completely remove the music in a given music+speech audio for a other NLP task