A fast speech feature extraction method based on perceptual hashing

Aiming at the existing problems of speech feature extraction algorithms based on perceptual hashing, such as problems with low efficiency and weak security, we present a fast speech perceptual hashing feature extraction method based on modified discrete cosine transform (MDCT) and compressed sensing (CS). Firstly, the speech signal is conducted with MDCT transform after processed by applying preprocessing, framing, adding window, and then the MDCT coefficients are regarded as the perceptual feature value. Secondly the measurement matrix in CS is applied to reduce the dimension of perceptual feature values. Finally, the feature value is used to conduct a hashing structure process to produce the perceptual hashing sequence. Experimental results show that the proposed method has a high efficiency in terms of time consumption, security and distinction.