Audio Classification Method Based on Machine Learning

Audio classification has very large theoretical and practical values in both pattern recognition and artificial intelligence. In this paper, we propose a novel audio classification method based on machine learning technique. Firstly, we illustrate the hierarchical structure of audio data, which is made up of four layers: 1) Audio frame, 2) Audio clip, 3) Audio shot, and 4) Audio high level semantic unit. Secondly, three types of audio data feature are extracted to construct feature vector, including 1) Short time energy, 2) Zero crossing rate and 3) Mel-Frequency cepstral coefficients. Thirdly, we discuss how to classify audio data using the SVM classifier with Gaussian kernel. Finally, experimental results demonstrate that the proposed method is able to achieve higher audio classification accuracy.