Gated Recurrent Networks Applied to Acoustic Scene Classification and Acoustic Event Detection

We present two resource efficient frameworks for acoustic scene classification and acoustic event detection. In particular, we combine gated recurrent neural networks (GRNNs) and linear discriminant analysis (LDA) for efficiently classifying environmental sound scenes of the IEEE Detection and Classification of Acoustic Scenes and Events challenge (DCASE2016). Our system reaches an overall accuracy of 79.1% on DCASE 2016 task 1 development data, resulting in a relative improvement of 8.34% compared to the baseline GMM system. By applying GRNNs on DCASE2016 real event detection data using a MSE objective, we obtain a segment-based error rate (ER) score of 0.73 – which is a relative improvement of 19.8% compared to the baseline GMM system. We further investigate semi-supervised learning applied to acoustic scene analysis. In particular, we evaluate the effects of a hybrid, i.e. generativediscriminative, objective function.