README

This is sourcecode of DCASE Challenge 2016 Task 1 of paper
http://www.cs.tut.fi/sgn/arg/dcase2016/documents/challenge_technical_reports/Task1/Kong_2016_task1.pdf

THIS CODE IS MIT LISCENCED. 
AUTHOR: QIUQIANG KONG, q.kong@surrey.ac.uk
CVSSP, UNIVERSITY OF SURREY
Created: 2016.06.24
Latest modified: 2017.05.02
======================================

Pre-requiste python packages:
pip install hat [--upgrade]	# Deep learning toolbox, https://qiuqiangkong.github.io/Hat

Usage:
------ Preparation ------
* modify the paths of dataset in config.py
* python prepare_data.py --calculate_logmel 	# Calculate features

------ Development Cross Validation ------
* python main_dnn.py --dev_train	# Train model on cross validation of development data. 
* python main_dnn.py --dev_recognize	# Get predictions and print out statistics. 

------ Testing of Private Data ------
* python main_dnn.py --eva_train	# Train model on whole development data
* python main_dnn.py --eva_recognize	# Get predictions of private data and print statistics.