AUDIO ADVERTISEMENT IDENTIFICATION USING AUDIO FINGERPRINTING METHOD

• Main Author: Nur Ibrahim Santosa, Rahmat
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/51216
• Institution: Institut Teknologi Bandung

Audio identification has become an important research field as industrial needs such as broadcast monitoring evolve. One method that is widely used to identify audio is the audio fingerprinting method, which is the extraction of important features from the audio that can represent audio characteristics in a concise manner. This research focuses on the identification of audio advertisement because advertisement have different characteristics from music, specifically the presence of additional noise in the form of human speech. To identify audio advertisement, three main processes are used, consisting of fingerprint extraction, database storage, and fingerprint matching. Fingerprint extraction is done by taking the most significant peaksfrom audio wave frequencies. The peak frequency values along with its offsets time from the reference audio then are stored in the database. Fingerprint matching is done using sinusoidal pairing and time differences module. The score generated by the module must exceed the final threshold in order to be declared as the identification result of the system. The identification system created in this research resulting 0.833 recall score and 1.00 precision score for identifying audio advertisements using dataset consists of 62 audio advertisements from local radio broadcast. In addition of white noise to the audio with magnitude of 0.2 amplitude, the same recall and precision score is obtained. The identification system created is also able to identify several audio advertisements at one identification process, if there are more than one audio advertisements exists in an audio sample input.