Musical Instrument Recognition in Isolated Notes and Solo Phrases
By Yuanbo Han and Wenyu Zhang, November 2018.
Methodology
-
Gammatone Filtering
-
CNN
Dataset
RWC Music Database: Musical Instrument Sound
Environment
- MATLAB R2018a
- Python 3.6
- scipy 1.1
- numpy 1.14
- torch 1.0
Note that these are only tested versions, not the least requirements.
Usage
Run gammatone/main.m to perform Gammatone filtering, but please first check the dataset path in line 5 and line 8. To get a grasp of the dependent functions, see gammatone_demo.m for a simple example.
After preprocessing, run cnn/read_data.py to read .MAT files as .NPY format. Then execute CNN.pyto train the neural network. There is an instance (mat_files/XY.mat) for testing CNN.py without above steps.