Undergrad Research Project - Singing Voice Separation From Music Recordings And Personalization Through Remix

Spring 2015

Yangyang Xia
Richard Stern
Project description

Separating singing voices from music recordings is an appealing topic because it enables one to personalize the sounding of music through remix with altered vocal or instrumentation. In this research project, I will focus on the implementation and optimization of an algorithm which separates singing voice from recordings, and eventually develop an application that is capable of personalizing a given piece of music. I will first replicate an algorithm in the literature (by Paris Smaragdis) that produces state-of-the-art separation of singing voice. Standard digital signal processing techniques will be coupled with statistical models and machine learning techniques in the development process. A correct implementation will then be optimized in terms of the degree of separation and computation complexity. Different combinations of statistical models and DSP and ML algorithms will be experimented for optimal performance. Finally, an application will be developed based on the algorithm, which gives users control of a range of parameters that enables them to remix music to fit their own tastes. The anticipated separation should be comparable to the state-of-the-art results, and the remixed music should show an effective amount of personalization through subjective listening.

Return to project list