Robust perceptual audio hashing

Robust audio hashing can be used for audio content identification in broadcast monitoring and database indexing applications. We propose novel robust audio hashing techniques in which small changes in the source data (such as filtering, compression, AD/DA conversion, etc.) do not cause any alteration to the hash sequences. In one scheme, we use the periodicity property of the audio signals measured with correlation-based techniques. In a second scheme, the audio signal is converted into matrix form through Mel frequency cepstral coefficients (MFCC). The singular values of this matrix constitute the hash values of the signal of interest. Experiments performed involving several attacks have proven the robustness and specificity of the proposed methods.