Change search
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf
Dissection of a Generative Network for Music Composition
KTH, School of Electrical Engineering and Computer Science (EECS).
2019 (English)Independent thesis Advanced level (degree of Master (Two Years)), 20 credits / 30 HE creditsStudent thesisAlternative title
Dissektion av ett generativt nätverk för musikkomposition (Swedish)
Abstract [en]

Controlling what a neural network generates has had great success when applied in the image domain. This thesis explores the performance of similar methods but instead applied in music generation. WaveNET, a state of the art neural network in audio synthesis and generation is trained using Generative Adversarial Networks to produce piano music. Two different methods for controlling the generation are presented, named HARD and SOFT. The HARD method fails to produce music of the same quality as without control. The SOFT method generates music of the same perceptual quality as without control but fails to control the output of the network. Finally, a discussion why this might be, and ideas regarding other methods for controlling the generation of music, and sequences in general are presented.

Abstract [sv]

Att kontrollera vad ett neuralt nätverk genererar har haft stor framgång när det applicerats på bilddomänen. Detta examensarbete undersöker hur liknande metoder fungerar i musikgenerering. Med hjälp av generativa motverkande nätverk tränar vi WaveNET, ett toppmodernt nätverk i ljudsyntes och generation, för att producera pianomusik. Två olika metoder för att styra genereringen presenteras, vid namn HARD och SOFT. Metoden HARD misslyckas med att producera musik av samma kvalitet som utan kontroll. Metoden SOFT skapar musik av samma perceptuella kvalitet som utan kontroll, men misslyckas med att styra genereringen. Vi diskuterar varför det här kan vara och presenterar idéer för andra metoder för att styra genereringen av musik och sekvenser i allmänhet.

Place, publisher, year, edition, pages
2019. , p. 44
Series
TRITA-EECS-EX ; 2019:486
National Category
Computer and Information Sciences
Identifiers
URN: urn:nbn:se:kth:diva-260961OAI: oai:DiVA.org:kth-260961DiVA, id: diva2:1356187
Supervisors
Examiners
Available from: 2019-10-08 Created: 2019-10-01 Last updated: 2019-10-08Bibliographically approved

Open Access in DiVA

fulltext(2763 kB)2 downloads
File information
File name FULLTEXT01.pdfFile size 2763 kBChecksum SHA-512
3aebb3f2be27ce365a88d74e1d9b6b85b8607098f2091d9b6655915b0e13c5917b7c016e808c6a27a963288fbd1489538c839718c0c0677df118f1008329bb51
Type fulltextMimetype application/pdf

By organisation
School of Electrical Engineering and Computer Science (EECS)
Computer and Information Sciences

Search outside of DiVA

GoogleGoogle Scholar
Total: 2 downloads
The number of downloads is the sum of all downloads of full texts. It may include eg previous versions that are now no longer available

urn-nbn

Altmetric score

urn-nbn
Total: 17 hits
CiteExportLink to record
Permanent link

Direct link
Cite
Citation style
  • apa
  • ieee
  • modern-language-association-8th-edition
  • vancouver
  • Other style
More styles
Language
  • de-DE
  • en-GB
  • en-US
  • fi-FI
  • nn-NO
  • nn-NB
  • sv-SE
  • Other locale
More languages
Output format
  • html
  • text
  • asciidoc
  • rtf