Current research on speech content identification aim primarily at raw wideband speech signals, which are generally transmitted in a compressed format. This makes it unable to meet the demand of speech content identification in compressed domain. This paper proposes a new speech perceptual hashing algorithm for speech content identification with compressed domain based on MFCC (Mel Frequency Cepstral Coefficient), to solve problems of real-time speech content identification and large quantity of voice message information over the mobile Internet. This algorithm extracts MFCC feature based on the raw wideband method. The process begins by extracting the MDCT coefficients, which are the intermediately decoded results of compressed speeches in MP3 format. These coefficients are translated to MFCC parameters and the binary hashing values are then generated from these parameters, combined with human auditory features. This algorithm uses highly compressed data to realize fast identification for speech content. Experimental results show that the proposed algorithm can realize tampering localization and increase 5% in efficiency when compared with raw wideband algorithms, with the precondition of robustness and discrimination.