CLASSIFICATION OF AUDIO SOURCE CHANNELS USING TIME-FREQUENCY ANALYSIS AND DECISION TREE CLASSIFIER

Abstract

This paper presents the development and classification of audio source separation algorithms utilizing timefrequency analysis techniques, specifically Wigner-Ville distributions (WVDs), compact kernel distribution (CKD) and multi-directional distribution (MDD). These methods are integrated with a decision tree classifier to enhance the separation of audio sources in noisy environments. The algorithms were tested on the AI-generated audio signals across various signal-to-noise ratio (SNR) levels. The CKD method demonstrated exceptional performance, achieving a classification accuracy of 100% at 0dB to 15dB SNR for the multichannel AI-generated audio signals. This study highlights the effectiveness of the advanced time-frequency analysis techniques CKD and MDD in improving audio source separation and their potential for real-time audio processing applications. The results indicate that these techniques can significantly enhance the clarity and quality of separated audio signals, providing an effective solution for tasks such as
music source separation, speech enhancement, and environmental sound separation.