MIR3
Music Instrument Recognition using convolutional neural networks
The repository contains the source code for music instrument recognition implemented in python/tensorflow v0.12. The model was trained using the MedleyDB dataset
Prediction
- After cloning the repository, run sh_download_model_1.sh
- Then run predict.py
Training
- Step 1. run data_chop_medley_db.py to chop long tracks into normalized 3-seconds chunks, excluding silence
- Step 2. run data_prepare_mini_experiment.py to prepare data subsets to be used for training and validation
- Step 3. run data_prepare_spectrograms.py to do resampling, FFT and MFCC
- Step 4. run train.py for training
Experiments
run MFCC.ipynb and MedleyExamples.ipynb