Conclution

We have seen the trends of machine learning, now is the turn of deep learning. Deep learning is used in a variety of field, and we can see every sign of it in our daily life. NPL, Computer Vision are not all, we have learned about deep learning in the audio domain which it can be used to compose music. Although, it is kind of like an entertainment. What we should see is the potential of it, if it was the start of the revolution of a new form of art–generate by machine. Our review is based on Sageev Oore, using MIDI file dataset and events that similar to MIDI events as the representation of music. The model they proposed is aiming to not only generate music but also music with expressive timing and dynamics. And the network structure is the LSTM network which is good with dealing sequences data. During our research, we have trained our own dataset also use piano performance. For the feature work, we consider using different types of music as the training set, see what kind of music would generate.

References

  1. Oore, S., Simon, I., Dieleman, S. et al. This time with feeling: learning expressive musical performance. Neural Comput & Applic 32, 955–967 (2020). https://doi.org/10.1007/s00521-018-3758-9
  2. File format description–(.mid) Standard MIDI File Format. http://faydoc.tripod.com/formats/mid.htm
  3. Ah Chung Tsoi, Andre Back, “Discrete time recurrent neural network architectures: A unifying review”, Neurocomputing, Volume 15, Issues 3–4, June 1997, Pages 183-223
  4. Curtis Hawthorne, Andriy Stasyuk, Adam Roberts, Ian Simon, Cheng-Zhi Anna Huang,Sander Dieleman, Erich Elsen, Jesse Engel, and Douglas Eck. ”Enabling Factorized Piano Music Modeling and Generation with the MAESTRO Dataset.” In International Conference on Learning Representations, 2019.
  5. Boehmer K (1967) Zur Theorie der offenen Form in der neuen Musik. Edition Tonos, Darmstadt
  6. Simon, Ian, and Sageev Oore. ”Performance rnn: Generating music with expressive timing and dynamics.” JMLR: Workshop and Conference Proceedings. Vol. 80. 2017.
  7. Walder, Christian, and Dongwoo Kim. ”Computer assisted composition with recurrent neural networks.” arXiv preprint arXiv:1612.00092 (2016).