An audio retrieval method based on chromagram and distance metrics

In this paper, a content-based audio retrieval method is proposed, which can quickly detect and locate similar sound in audio database. We extract a chroma-based audio feature: chromagram, a variation on time-frequency distributions, which represents the spectral energy at each of 12 pitch classes. Compared with traditional feature MFCC (Mel Frequency Cesptral Coefficient), chromagram is better when using correlation distance as audio similarity measurement. Then we choose Jonathan Foote's music retrieval database to do experiments and final results show that the retrieval accuracy can reach over 96.7% using chromagram as features even when the signal-to-noise ratio is 0 dB.