Accurate audio-segment classification using feature extraction matrix

Naoki Nitanda, Miki Haseyama, Hideo Kitajima

Abstract: This paper proposes an accurate audio signal classification method using a feature extraction matrix. The proposed method classifies the segments of the audio signal into the following five audio classes: silence, speech, music, speech with music background, and speech with noise background. In this classification, a diagonal matrix, which is called the feature extraction matrix, is utilized in order to extract the effective audio features for the classification. By using this feature extraction matrix, the five audio classes are clearly separated each other in the feature space, and thereby highly precise classification can be attained. Experimental results performed by applying the proposed method to real audio signals are shown to verify its high performance.