Research On Digital Audio Judicial Authentication Technology

With the popularity of digital technology products such as audio and video in people’s daily life, and due to the widely use of kinds of audio and video editing software at the same time, the safety of digital information has attracted more and more attention. Especially the audio files, as a kind of legitimate audio-visual evidence material, have emerged in the modern judicial cases with its characteristics of easy access and storage. Therefore, recognizing the speaker in the audio (identity authentication) and authenticating the authenticity of the recording (authenticity authentication) become an important issue in the court proof, and digital audio forensics technology increasingly presents its theoretical significance and practical value.Combining with the actual audio judicial authentication cases, the two aspects of the audio judicial authentication technology—identity and authenticity authentication, are both studied in this paper. For audio identity authentication, the method of phonetics and automatic voiceprint recognition are both used. And among which, the method of phonetics analysis is based on formant frequency extraction, which can be more accurate and reliable compared with the previous method that only rely on the observation of formant position and direction. For the automatic voiceprint recognition, a multi-channel voiceprint recognition speech database—VPRSD has been built in order to improve the accuracy of the voiceprint recognition system based on GMM-UBM. And experiments are carried out from the following aspects:The type of voice channel, the number of voice channel, and the number of the background speakers. The results indicate that, the VPRSD gives great help to improve the performance of the voiceprint recognition system. For audio authenticity authentication, this paper adopts a method based on Electric Network Frequency (ENF). This method converts audio signal detection to electric network signal (single and constant sinusoidal signal) detection by extracting the electric network signal attached in the recording files, and then estimates the phase of the electric network signal. Finally, we will check whether there is a mutation of the phase to decide whether the audio has been manipulated. Besides, we also use the artificial review and spectrum analysis method to make more reliable conclusion for audio authenticity authentication.Experiments show that the proposed authentication technology and scheme can satisfy the common digital audio judicial authentication.