| Presentation | 2020-01-17 A study sound source separation method for recorded multiple sounds Satoru Jomae, Kenko Ota, Hideaki Yoshino, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | It is useful to separate multiple sounds into single sounds in order to improve the accuracy of fundamental frequency analysis and chord analysis of multiple sounds included in recorded music. In previous research, we applied the basis enhancement based on Deep Neural Network classification to the learning process of gamma process non-negative matrix factorization in order to improve separation accuracy. However, there was a slight improvement in accuracy depending on the sound source used in the evaluation, but no significant improvement was confirmed. The reason why the accuracy was not improved was clarified by analyzing changes in the number of bases and errors during separation. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | Sound source separation / non-negative matrix factorization / DNN / machine learning |
| Paper # | CQ2019-131 |
| Date of Issue | 2020-01-09 (CQ) |
| Conference Information | |
| Committee | CQ |
|---|---|
| Conference Date | 2020/1/16(2days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | NHK Science & Technology Research Laboratories |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | AR/VR, Broadcasting Service, Video/Voice Services Quality, High Realistic, User Behavior/Psychology, User Experience, Media Quality, Network Quality and QoS Control, Networks and Communications at Disaster, User Behavior, Machine Learning, etc. |
| Chair | Hideyuki Shimonishi(NEC) |
| Vice Chair | Jun Okamoto(NTT) / Takefumi Hiraguri(Nippon Inst. of Tech.) |
| Secretary | Jun Okamoto(Nippon Inst. of Tech.) / Takefumi Hiraguri(Doshisha Univ.) |
| Assistant | Chikara Sasaki(KDDI Research) / Yoshiaki Nishikawa(NEC) / Takuto Kimura(NTT) |
| Paper Information | |
| Registration To | Technical Committee on Communication Quality |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | A study sound source separation method for recorded multiple sounds |
| Sub Title (in English) | |
| Keyword(1) | Sound source separation |
| Keyword(2) | non-negative matrix factorization |
| Keyword(3) | DNN |
| Keyword(4) | machine learning |
| 1st Author's Name | Satoru Jomae |
| 1st Author's Affiliation | Nippon Institute of Technology(NIT) |
| 2nd Author's Name | Kenko Ota |
| 2nd Author's Affiliation | Nippon Institute of Technology(NIT) |
| 3rd Author's Name | Hideaki Yoshino |
| 3rd Author's Affiliation | Nippon Institute of Technology(NIT) |
| Date | 2020-01-17 |
| Paper # | CQ2019-131 |
| Volume (vol) | vol.119 |
| Number (no) | CQ-367 |
| Page | pp.pp.137-140(CQ), |
| #Pages | 4 |
| Date of Issue | 2020-01-09 (CQ) |