Performance Analysis of Discrete Wavelet Transforms for Acoustic Scene Classification with DCASE Dataset

To control the chaos during an acoustic scene, it is crucial to recognize each particular sound class in the surrounding. This paper discusses the development of a model for acoustic scene classification, incorporating Fast Dynamic Time Warping (DTW) and reference audio. The main goal of the model is to correctly classify various sounds that are produced in environments, including airport, metro stations, parks, public square, bus etc. The model successfully identifies similarities across the extensive dataset of 23035 audio samples gathered from the Detection and Classification of Acoustic Scenes and Events (DCASE) 2021 Challenge Task 1 by employing the Fast DTW approach. In light of the significance of handling extensive audio data, the integration of reference audio diminishes processing complexity. Additionally, the suggested model investigates the effectiveness of features from several Discrete Wavelet Transforms, including Meyer, Symlets, and Daubechies, to improve classification accuracy. The acquired results, which demonstrate the model’s efficacy and breakthroughs in acoustic scene classification, are compared with the DCASE 2021 baseline model and other technical papers presented in the same year. With the Meyer wavelet characteristic of audio data, the suggested model attains an accuracy of 79.92%.