When an audio recording is used as evidence in litigation and forensic investigations, it needs to be checked thoroughly for authenticity and integrity in order to be admissible, compelling, and decisive evidence in a court of law. An audio recording can be subject to tampering attacks with easy-to-use editing and signal processing tools, thereby undermining its legal value. Artifacts embedded in an audio recording can provide valuable clues about the acoustic environment in` which the audio was recorded and allow for the detection of tampering. This paper presents findings of two parallel methodologies: (1) where the features are extracted from the room impulse response and (2) where features are extracted directly from the reverberated recordings. These methods focus on extracting parameters from audio recordings that helped distinguish different auditory scenes. Experiments employing an exhaustive set of machine learning classifiers along with different acoustic features were conducted for the classification of auditory environments. A comparative analysis has been carried out to assess the performance of each classifier and relative performance impact of each feature set in terms of the accuracy of classification. A two-layer Artificial Neural Network (ANN) provided an accuracy of 98.7% using room impulse responses and an accuracy of 99.5% when the reverberated audio recordings were trained.
(See document for exact affiliation information.)
https://aes2.org/publications/elibrary-page/?id=20705
(403KB)
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member Join the AES. If you need to check your member status, login to the Member Portal.
