Musical Source Separation: A Review of Techniques and the Proposal of Cepstromarg FilteringIn this thesis, we give a review of modern blind, semi-blind, and informed source separation techniques. Specifically, the details of independent component analysis are derived, and several non-negative matrix factorization algorithms and adaptations are introduced. We focus on source separation applied to monaural musical signals in an attempt to develop a real-time musical decomposition system with applications in audio visualization, vocal recognition, karaoke generation, and more. Harmonic- percussive source separation algorithms are investigated, and a new filtering technique is proposed by defining a new transform space, the cepstromarg domain, which allows for instantaneous harmonic-percussive separation.Charous, AaroncreatorSandstede, Björnthesis advisorRose, ChristopherreaderBrown University. Applied Mathematicssponsor2019textborn digitalEnglishSenior thesis (ScB)--Brown University, 2019Applied MathematicsthesesInformation storage and retrieval systems--MusicMusic information retrievalBlind source separationIndependent component analysisNon-negative matricesAuditory scene analysisSignal processing--Digital techniques--Mathematics10.26300/pcpz-6m06In CopyrightCollection is open for research.