Singing Voice Separation System Design for Monaural Polyphonic Music Signals

26 May 2017 Lin Kin Wah Edward ISTD Singing Voice Separation

PhD Student
Lin Kin Wah Edward, Information Systems Technology and Design
 
Supervisor
Simon Lui, Assistant Professor, Information Systems Technology and Design


Separating singing voice from the music accompaniment in monaural (i.e. mono) polyphonic music recordings is challenging. This is because, like solving an equation with 2 unknown variables (e.g. z = x + y), the singing voice and music accompaniment are not easily distinguishable by machines.
 
This research project proposes a system for separating singing voice by (1) tracking the singing characteristics in very fine sound fragments and (2) using machine learning to classify these sound fragments into the singing voice and music accompaniment.
 
The success of the system will stimulate the development of other applications such as:

  1. musical information retrieval systems (e.g. melody extraction);

  2. musical tools (e.g. automatic music score transcription);

and many others.

Read more here.