An integrated convolutional neural network with a fusion attention mechanism for acoustic scene classification

摘要

In human-computer interaction, acoustic scene classification(ASC) is one of the relevant research domains. In real life, therecorded audio may include a lot of noise and quiet clips, making it hardfor earlier ASC-based research to isolate the crucial scene information insound. Furthermore, scene information may be scattered across numerousaudio frames; hence, selecting scene-related frames is crucial for ASC. Inthis context, an integrated convolutional neural network with a fusion attentionmechanism (ICNN-FA) is proposed for ASC. Firstly, segmentedmel-spectrograms as the input of ICNN can assist the model in learningthe short-term time-frequency correlation information. Then, the designedICNN model is employed to learn these segment-level features. In addition,the proposed global attention layer may gather global information byintegrating these segment features. Finally, the developed fusion attentionlayer is utilized to fuse all segment-level features while the classifier classifiesvarious situations. Experimental findings using ASC datasets fromDCASE 2018 and 2019 indicate the efficacy of the suggested method