A Metric Learning Based Solution for Non-Stationary Acoustic Source Classification

In this work, a metric learning-based approach is proposed for non-stationary acoustic source classification. A classic time-frequency representation of acoustic signals is adopted as input of a convolutional neural network in order to generate embedded features of reduced size. The embedding generation is optimized on similarity constraints in order to maximizes intra-class and minimize inter-class distances. Eight sources with different degrees of non-stationarity are selected for the acoustic source classification task. Experiments demonstrated that the proposed solution outperforms the baseline system for all individual acoustic sources, leading to an increment on the average balanced accuracy of mote than ten percentage points.