Presentation | 2011-12-19 Study on extraction of vocal part in music signal by using non-negative matrix factorization Yuta YASUI, Hideki BANNO, Fumitada ITAKURA, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | This paper describes extraction methods of vocal signal in music signal which is a mixture of vocal signal and accompaniment signal by using non-negative matrix factorization. Non-negative matrix factorization (NMF) can factorize an input spectrogram into a finite number of basis vectors and its temporal activity information, because it represents similar spectral patterns appeared on the input spectrogram with a single basis vector. However, NMF is not suitable for extraction of vocal signal because factorization of vocal signal including temporal spectral fluctuation appeared in vibrato of singing voice into a finite number of basis vectors is quite difficult. To solve this problem, we propose a preprocessing method that removes the spectral fluctuation by using a linear frequency axis warping of the spectrum so that a fundamental frequency of vocal signal included in the input music signal aligns to a reference frequency. Then, NMF is applied to this preprocessed signal. We have performed evaluation by SNR of extracted vocal signal and extracted accompaniment signal, in comparison with the conventional method. As a result, it was found that the generated signals by the proposed method had lower quality and SNR. However, the proposed method obtained slight better results for some music signals. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | Non-negative Matrix Factorization / Sound Source Separation / Music Signal / Vibrato |
Paper # | NLC2011-43,SP2011-88 |
Date of Issue |
Conference Information | |
Committee | NLC |
---|---|
Conference Date | 2011/12/12(1days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | |
Chair | |
Vice Chair | |
Secretary | |
Assistant |
Paper Information | |
Registration To | Natural Language Understanding and Models of Communication (NLC) |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | Study on extraction of vocal part in music signal by using non-negative matrix factorization |
Sub Title (in English) | |
Keyword(1) | Non-negative Matrix Factorization |
Keyword(2) | Sound Source Separation |
Keyword(3) | Music Signal |
Keyword(4) | Vibrato |
1st Author's Name | Yuta YASUI |
1st Author's Affiliation | Graduate School of Science and Technology, Meijo University() |
2nd Author's Name | Hideki BANNO |
2nd Author's Affiliation | Meijo University |
3rd Author's Name | Fumitada ITAKURA |
3rd Author's Affiliation | Meijo University |
Date | 2011-12-19 |
Paper # | NLC2011-43,SP2011-88 |
Volume (vol) | vol.111 |
Number (no) | 364 |
Page | pp.pp.- |
#Pages | 6 |
Date of Issue |