Stand-alone music creation using machine learning
(2017) In Master's Theses in Mathematical Sciences FMA820 20171Mathematics (Faculty of Engineering)
- Abstract
- Creating music have always been a hard thing to do, as time have passed there have been much research done on the topic of creating high quality music. New approaches and theories are appearing on how to generate music, the latest one (as writing this paper) is based on deep neural networks. In this paper Nottinghams abc notation database will be the main training set used as training data for the classifier. This paper will mainly investigating Recurrent neural networks (RNN) capability to generate music as they are created for pattern recognition in sequences of data. Will see evaluated different RNN models such as bidirectional RNN and seq2seq.
This thesis shows that deep neural networks are able to generate music how-
ever... (More) - Creating music have always been a hard thing to do, as time have passed there have been much research done on the topic of creating high quality music. New approaches and theories are appearing on how to generate music, the latest one (as writing this paper) is based on deep neural networks. In this paper Nottinghams abc notation database will be the main training set used as training data for the classifier. This paper will mainly investigating Recurrent neural networks (RNN) capability to generate music as they are created for pattern recognition in sequences of data. Will see evaluated different RNN models such as bidirectional RNN and seq2seq.
This thesis shows that deep neural networks are able to generate music how-
ever improving the accuracy with different size, depth, RNN cells and other means to the networks is much harder. Through all networks tested there where never any large difference to the accuracy to the validation set. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/8939229
- author
- Nilsson, Adam LU
- supervisor
-
- Karl Åström LU
- organization
- course
- FMA820 20171
- year
- 2017
- type
- H2 - Master's Degree (Two Years)
- subject
- keywords
- Music generation, machine learning, music improvisation, deep learning, LSTM, Recurrent Neural Networks
- publication/series
- Master's Theses in Mathematical Sciences
- report number
- LUTFMA-3330-2017
- ISSN
- 1404-6342
- other publication id
- 2017:E56
- language
- English
- id
- 8939229
- date added to LUP
- 2018-05-30 17:44:08
- date last changed
- 2018-05-30 17:44:08
@misc{8939229, abstract = {{Creating music have always been a hard thing to do, as time have passed there have been much research done on the topic of creating high quality music. New approaches and theories are appearing on how to generate music, the latest one (as writing this paper) is based on deep neural networks. In this paper Nottinghams abc notation database will be the main training set used as training data for the classifier. This paper will mainly investigating Recurrent neural networks (RNN) capability to generate music as they are created for pattern recognition in sequences of data. Will see evaluated different RNN models such as bidirectional RNN and seq2seq. This thesis shows that deep neural networks are able to generate music how- ever improving the accuracy with different size, depth, RNN cells and other means to the networks is much harder. Through all networks tested there where never any large difference to the accuracy to the validation set.}}, author = {{Nilsson, Adam}}, issn = {{1404-6342}}, language = {{eng}}, note = {{Student Paper}}, series = {{Master's Theses in Mathematical Sciences}}, title = {{Stand-alone music creation using machine learning}}, year = {{2017}}, }