Encrypted Domain Mel-Frequency Cepstral Coefficient and Fragile Audio Watermarking

摘要

Audio has become increasingly important in modern social communication and mobile Internet, e.g., the employment of voice messaging in mobile social Apps. In the scenario of cloud computing, we need to consider the privacy protection of the audio content and the integrity of the audio simultaneously. In this paper, we study the application of an encrypted audio fragile watermarking in a privacy-preserving audio authentication system in the cloud. We firstly implement the computation of Mel-frequency cepstral coefficient in the encrypted domain (MFCC-ED) with additive homomorphic encryption. By performing MFCC-ED on the encrypted audio, we extract an encrypted fragile watermark message. The encrypted fragile watermark is then embedding into the encrypted discrete wavelet transform domain. Our experimental results show that the proposed watermarked scheme is fragile that is sensitive to common audio attacks. We also present the localization method that is able to well localize the temper positions.