Clipping detection of audio signals based on kernel Fisher discriminant

In this paper, a kind of clipping detection method for audio signal is proposed based on kernel Fisher discriminant (KFD) in MDCT domain. The kernel method and the Fisher linear discriminant analysis (FLDA) are introduced to the proposed method. First, the clipping and non-clipping feature parameters are extracted using MDCT coefficients of audio signals. Next, the optimal projection vector and the classification threshold are obtained by training KFD based on the extracted feature parameters. Finally, the decision rule of KFD is employed to detect the clipping distortion. The test results indicate that the proposed algorithm yields good clipping detection result, the performance is better than the reference method.