Acoustic Scene Classification Using Deep Audio Feature and BLSTM Network

Although acoustic scene classification has been received great attention from researchers in the field of audio signal processing, it is still a challenging and unsolved task to date. In this paper, we present our work of acoustic scene classification for the challenge of the Detection and Classification of Acoustic Scenes and Events 2017, i.e., DCASE2017 challenge, using a feature of Deep Audio Feature (DAF) for acoustic scene representation and a classifier of Bidirectional Long Short Term Memory (BLSTM) network for acoustic scene classification. We first use a deep neural network to generate the DAF from Mel frequency cepstral coefficients, and then adopt a network of BLSTM fed by the DAF for acoustic scene classification. When evaluated on the official datasets of the DCASE2017 challenge, the proposed system outperforms the baseline system in terms of classification accuracy.