Robust speech endpoint detection based on MP3 file in various noisy environments

In audio retrieval, endpoint detection is a crucial step of front processing. This paper presents a MP3-based approach for robust endpoint detection under different types of noises. Firstly, the modified discrete cosine transform (MDCT) coefficients and scale factors (SCF) are extracted from MP3 file. Then, combined with the selection of SCF, a set of MDCT coefficients of white noise are used to enhance the speech signals. Finally, the spectral entropy-based method is applied to MDCT coefficients to gain the proposed parameter. Experimental results show that this algorithm has excellent performance in various adverse environments. Furthermore, for using the intermediate date of MP3 decoding, the proposed one has much less computational complexity.