Analysis of audio fingerprinting techniques
Siva Sankaran, Satish Kumar
MetadataShow full item record
The goal of this thesis is to compare various audio fingerprinting algorithms under a common framework. An audio fingerprint is a compact content-based signature that uniquely summarizes an audio recording. In this thesis, acoustic fingerprints are based on prominent peaks extracted from the spectrogram of the audio signal in question. A spectrogram is a visual representation of the spectrum of frequencies in an audio signal as it varies with time. Some of the applications of audio fingerprinting include but are not limited to music identification, advertisement detection, channel identification in TV and radio broadcasts. Currently, there are several fingerprinting techniques that employ different fingerprinting algorithms. However, there is no study or concrete proof that suggests one algorithm is better in comparison with the other algorithms. In this thesis, some of the feasible techniques employed in audio fingerprint extraction such as Same-Band Frequency analysis, Cross-Band Frequency analysis, use of Mel Frequency Banks, and use of Mel Frequency Cepstral Coefficients (MFCC) are analyzed and compared under the same framework.