A Hybrid Approach Using Binaural i-vectors and Deep Convolutional Neural Networks

This report describes the 4 submissions for Task 1 (Audio scene classification) of the DCASE-2016 challenge of the CP-JKU team. We propose 4 different approaches for Audio Scene Classification (ASC). First, we propose a novel i-vector extraction scheme for ASC using both left and right audio channels. Second, we propose a Deep Convolutional Neural Network (DCNN) architecture trained on spectrograms of audio excerpts in end-to-end fashion. Third, we use a calibration transformation to improve the performance of our binaural i-vector system. Finally, we propose a late-fusion of our binaural i-vector and the DCNN. We report the performance of our proposed methods on the provided cross-validation setup for the DCASE-2016 challenge. Using the late-fusion approach, we improve the performance of the baseline by 17 percentage point in accuracy. Our submissions achieved ranks first and second among 49 submissions in the audio scene classification task of DCASE-2016 challenge.