Automatic scene change detection for composed speech and music sound under low snr in compressed domain

摘要

With the amount of MP3 compressed data increasing, automatic scene change detection is becoming more and more important. Several studies have proposed some interesting approaches. However, none of these techniques analyze the audio signals in a low SNR noisy environment in compressed domain. In this paper, a new scene-change detection algorithm is proposed to detect the composed speech and music scene change automatically in a low SNR noisy environment in compressed domain. From the Modified Discrete Cosine Transform (MDCT) matrix, three robust features are extracted which include the Compressed MDCT Average Envelope (CMAE) parameters, the Compressed Possibilities Density Ratio (CPDR), and the Compressed Possibilities density Ratio local Average (CPRA). A statistical activity detection model in compressed domain employs the CMA to segment the audio signal into pure noise and noisy audio segments for compressed data sets. The automatic scene detection for composed speech and music sound with noise under low SNR in compressed domain is implemented. The experimental results show that the proposed methods can detect the pure noise, noisy speech and noisy music effectively even when SNR is as low as 0dB.