Audio Examples 🔊 | Saurjya Sarkar

This page presents audio examples for my various projects: Choral Separation, Ensemble Separation, and Lead Singer Separation. The audio files presented are from models I’ve trained during my PhD and PostDoc research.

Lead Singer Separation

These examples are related to my work in collaboration with AudioStrip during my postdoc. This example highlights how lead vocal separation works, where you can listen that the lead vocalist swaps in the prediction for a single phoneme, as one of the backing vocalists temporarily becomes louder than the lead singer.

Mixture:

Separated Lead Vocal:

Reference Lead Vocal:

Separated Backing Vocals:

Reference Backing Vocals:

Choral Separation

These examples are related to the work from the Interspeech 2021 paper "Vocal Harmony Separation using Time-domain Neural Networks". The examples are at a sampling rate 22.05 kHz due to the dataset used.

Mixture:

Separated Vocal 1:

Reference Vocal 1:

Separated Vocal 2:

Reference Vocal 2:

Ensemble Separation

These examples are related to the work from the IEEE WASPAA 2023 paper "Leveraging Synthetic Data for Improving Chamber Ensemble Separation". The test examples are real recordings from the URMP dataset at a sampling rate 44.1 kHz.

Mixture:

Separated Violin 1:

Reference Violin 1:

Separated Violin 2:

Reference Violin 2: