/multi_resolution_classification

mra classification approach for ESC audio data set

Primary LanguagePython

Multi-resolution Analysis (MRA) based classification for audio data.

Utilizes auto-weka and LSTM on untransformed and 4 approximation / detail representations. 

Install weka and auto-weka 
(https://www.cs.waikato.ac.nz/~ml/weka/, https://www.cs.ubc.ca/labs/beta/Projects/autoweka/)
Install tensorflow and keras via pip3 utility

Usage:
(process data: extract features, normalize; build mra classifier and test) 
The processed feature files for the ESC-10 data set (https://github.com/karoldvl/ESC-50) are supplied in the data directory. 

setup paths as necessary (e.g. to cuda for tensorflow-gpu)
python3 lstm1_sound.py (uses LSTM)
./run_mra_voting.pl (uses auto-weka + LSTM)


Sergey Voronin, 2018