Acoustic scene classification by fusing lightgbm and vgg-net multichannel predictions

This report provides a solution for the task 1 of DCASE 2017 challenge. We build two parallel audio scene classification systems – LightGBM and VGG-net. Their prediction scores are output respectively from the multichannel version of the TUT Acoustic Scenes 2017 dataset. We perform a linear logistic regression method to fuse the (1) LightGBM, (2) VGG-net and (3) LightGBM+VGG-net multichannel scores. Finally, three outputs from the fused systems are submitted for the challenge. The evaluation is done on the development set.