Spectra456/silero-test

Run

1. Download dataset from Kaggle and place it inside repository directory.

2. Launch script for converting wav to jpg (by default it's using all your cpu threads, but you can change it inside code).

python wav2jpg.py

After this you got my_dataset.csv with jpg filename, target. Now you need to split it to train.csv and test.csv

3.1 For training CNN

python train.py

3.2 Make prediction for submission

python inference.py

4 Clusterization binary images with k-means. Got only 0.415. I think, we need to reduce number of features, here i have 1d array with size 128x72 = 9216. Maybe we can apply here linear transformation, just like in neural networks.

python kmeans_images.py