Mixed Watermarking-Fingerprinting Approach for Integrity Verification of Audio Recordings

We introduce a method for audio-integrity verification based on a combination of watermarking and fingerprinting. An audio fingerprint is a perceptual digest that holds content information of a recording and allows one to identify it from other recordings. Integrity verification is performed by embedding the fingerprint into the audio signal itself by means of a watermark. The original fingerprint is reconstructed from the watermark and compared with a new fingerprint extracted from the observed signal. If they are identical, the signal has not been modified; if not, the system is able to determine the approximate locations where the signal has been corrupted. The watermarked signal may undergo content preserving transformations, such as resampling or D/A and A/D conversion, without triggering the corruption alarm.