Nonnegative Matrix Factorization for Music - Tuning the NMF Algorithm with Regularization 公开

Abstract

The mathematics behind music is a work of art in itself. Mathematicians have been utilizing mathematical tools to analyze music for decades. One such tool is Nonnegative Matrix Factorization (NMF) which has been used to decompose an audio signal into fundamental components in a source separation application. The NMF algorithm in a musical interpretation takes a spectral object known as a spectrogram represented by a matrix and separates the spectrogram into a two nonnegative sparse matrix product where one matrix takes temporal information of the sources, and the other matrix gives the frequency information of the sources. While the basic NMF algorithm excels at handling small in complexity problems with little noise, it fails to successfully separate the sources for problems with many sources or bad-quality audio data. One solution to this limitation is the implementation of regularization into the NMF algorithm. Regularization aims to induce qualities into our matrix factorization such as promoting sparsity or smoothing temporal readings that will improve the source separation accuracy. In this paper, we hope to introduce the simple NMF algorithm, display the source separation application of NMF, and demonstrate the effects of a regularized NMF algorithm.

Table of Contents

1 Introduction........................................................................1

2 Related Works......................................................................4

3 From Recording to Spectrogram............................................7

4 Deriving the NMF Algorithm................................................11

4.1 NMF Convergence.............................................................15

5 Illustrative Examples...........................................................18

5.1 C Scale Example...............................................................18

5.2 Drum Sound Separation....................................................21

6 Regularized NMF.................................................................25

7 Experiments.......................................................................29

7.1 NMFD: Higher Complexity Audio Sample...........................29

7.2 NMFD: Symphony with Real Noise.....................................32

7.3 Regularized NMF: C Scale with Induced Noise.....................34

8 Conclusions and Future Directions.......................................38

Appendix A Example Derivation of Regularized NMF................41

Bibliography..........................................................................44

About this Honors Thesis

Rights statement

• Permission granted by the author to include this thesis or dissertation in this repository. All rights reserved by the author. Please contact the author for information regarding the reproduction and use of this thesis or dissertation.

School	Emory College
Department	Mathematics
Degree	B.S.
Submission	Honors Thesis
Language	English
Research Field	Mathematics Applied Mathematics
关键词	source separation nonnegative matrix factorization math music matrix decomposition optimization regularization audio data
Committee Chair / Thesis Advisor	Elizabeth Newman, Emory University
Committee Members	Simon Blakey, Emory University Lars Ruthotto, Emory University

最新修改

Primary PDF

Thumbnail	Title	Date Uploaded	Actions
	Nonnegative Matrix Factorization for Music - Tuning the NMF Algorithm with Regularization ()	2022-04-07 22:07:23 -0400	Press to Select an action Download

Supplemental Files

Thumbnail	Title	Date Uploaded	Actions