Fig. 6

Noise test for robustness (Shazam vs ORB). In this figure, the experiment involved testing the ability to accurately identify the audio sample when it is subjected to the addition of noise. In the real world, this is a common case. Additive white Gaussian noise was added which would be used to simulate background noise. The similar test setup was done where a 20-s clip was taken from each song in the database, and the various degrees of noise were added to each sample and then tested for each system. It was seen that the Shazam algorithm was very robust to the noise as seen in the results. This would be due to the peaks being acquired using the spectrogram data in the audio fingerprinting extraction process. The peaks obtained will be more likely to survive the effects of noise as the song’s frequencies (from the audio sample) would have a higher amplitude when compared to the amplitude of frequencies of the noise. It was seen that the Shazam’s algorithm performed better in comparison to the ORB prototype although the ORB prototype performed well to an extent