IAML_2018
Industrial Application of Machine Learning , Fall 2018, Seoul National University.
Project 1
Genre classification
- Requirements
- 30초 길이의 노래가 주어졌을 때 해당 곡의 장르 맞추기
- CNN 계열의 모델 사용
Project 2
Hit prediction
- Requirements
- 30초 길이의 노래가 주어졌을 때 해당 곡의 재생 횟수 맞추기
- 5개의 수준으로 나누어 classification 문제로 접근
- RNN 계열의 모델 사용
Project 3
Chord prediction
- Requirements
- 12초 길이의 wav 파일의 chord 맞추기
- Results
- Implemented the CRNN structure suggested in Choi. et al (2017) with some modifications to address to this specific task.
- The model showed 98% accuracy on the training set and 93% accuracy on the validation set, on average.
- The performance could be improved via other models such as the Transformer, but this haven't been tried yet.
- The raw music files aren't provided in this repository.
Project 4
Music Generation
TBA