Publications

Detailed Information

HymnTron: A DNN Based Automatic Music Composer

DC Field Value Language
dc.contributor.advisor강명주-
dc.contributor.author박진만-
dc.date.accessioned2019-05-07T04:31:45Z-
dc.date.available2019-05-07T04:31:45Z-
dc.date.issued2019-02-
dc.identifier.other000000153912-
dc.identifier.urihttps://hdl.handle.net/10371/151587-
dc.description학위논문 (석사)-- 서울대학교 대학원 : 자연과학대학 수리과학부, 2019. 2. 강명주.-
dc.description.abstract최근의 개발로 인해, 딥 뉴럴 네트워크 (Deep Neural Networks)를 사용하기위한 많은 어플리케이션이 존재한다. 그 중 하나는 음악 작곡이다. 이 논문에서 우리는 인간 작곡과 구별 할 수없는 음악을 작곡하려고 시도한다. 시드 노트로 작곡을 시작한다. 다음으로 LSTM을 사용하여 주어진 시드 노트와 코드 진행에서 수많은 멜로디를 작곡한다. 마지막으로, 우리는 빔 검색 알고리즘을 통해 원하는 수의 컴포지션 만 유지하기 위해 다양한 스코어링 알고리즘 함수를 사용한다. 인간 조사는 품질 테스트를 위해 수행된다.-
dc.description.abstractWith recent developments, there are many different applications for the use of Deep Neural Networks. One of which is music composition. In this thesis, we attempt to compose music that is indistinguishable from human composition. We initialize the composition with a seed note. Next, we utilize the long-short term memory network to compose numerous melodies from the given seed note and the chord progression. Finally, we use a variety of scoring functions to keep only a desired number of compositions through the beam search algorithm. Human survey is conducted for quality testing.-
dc.description.tableofcontentsContents
Abstract i
1 Introduction 1
2 Data Processing 3
2.1 Transformation . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Quantization . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.3 Renement . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.4 Encoding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3 Model Descriptions 7
3.1 Melody Sequential Model . . . . . . . . . . . . . . . . . . . . . 7
3.1.1 Model Architecture . . . . . . . . . . . . . . . . . . . . 7
3.1.2 Loss . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
4 Generation 10
4.1 Beam Search . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
4.2 Temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
4.3 Scoring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
4.4 Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
5 Experiments 15
5.1 Experiment Premises . . . . . . . . . . . . . . . . . . . . . . . 15
5.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
6 Conclusion 17
The bibliography 19
Abstract (in Korean) 20
-
dc.language.isoeng-
dc.publisher서울대학교 대학원-
dc.subject.ddc510-
dc.titleHymnTron: A DNN Based Automatic Music Composer-
dc.typeThesis-
dc.typeDissertation-
dc.description.degreeMaster-
dc.contributor.affiliation자연과학대학 수리과학부-
dc.date.awarded2019-02-
dc.identifier.uciI804:11032-000000153912-
dc.identifier.holdings000000000026▲000000000039▲000000153912▲-
Appears in Collections:
Files in This Item:

Altmetrics

Item View & Download Count

  • mendeley

Items in S-Space are protected by copyright, with all rights reserved, unless otherwise indicated.

Share