Audio classification based on audio WSOLA and CNN algorithm

Abstract

This paper presents an audio classification algorithm based on the WSOLA and CNN techniques to address the problem of data imbalance in audio classification. Audio classification involves categorizing audio signals into different labels or categories, and is crucial in speech recognition, music classification, sound time detection, and sound quality evaluation. In this study, we propose a method that utilizes the WSOLA algorithm to enhance the audio data with fewer categories in the dataset, which can help to improve the accuracy and stability of classification when the dataset is unbalanced. This approach can also prevent the model from focusing too much on categories with large data volumes while neglecting other categories.By mitigating the issue of audio data imbalance, the model can better learn the characteristics of all categories, thereby enhancing the model's performance. We conducted experiments on the UrbanSound8K dataset, where we enhanced the audio data using the WSOLA method and then classified the audio using the CNN classifier. Our results indicate that the overall classification accuracy and stability were significantly improved, demonstrating that the proposed approach can reasonably classify the audio dataset using the WSOLA and CNN techniques.