Audio Source Identification using Delta-Delta MFCC Features

With the increase in cybercrime involving audio, there is a growing demand for forensic audio analysis. Source microphone identification, a sub-field of audio forensics, is crucial in identifying the microphone used to record a particular audio file. This paper introduces a novel source microphone identification method based on the Delta-Delta mel-frequency cepstral coefficients (DD-MFCC). A multi-class linear discriminant analysis (LDA) is used for classification. The existing audio source identification methods are tested mostly on synthetic audio or recordings in controlled, noise-free environments. The performance of the proposed method is evaluated using real audio recorded in various indoor and outdoor environments. Several experiments conducted on these datasets demonstrate that the proposed method outperforms the existing audio source identification techniques with an average identification accuracy of above 94%.