A Resilient Novel Compressed-Domain Audio Recognition Method for Anti-Linear Speed Change

Abstract

Audio fingerprint technology has been increasingly played an important role in audio content identification, audio information security, industrial process monitoring, etc. Due to compressed format has become the main way for audio files storage and transmission, it owns more practical significance that directly extracting audio fingerprint from compressed-domain. In general, existing compressed-domain audio fingerprint schemes are robust to common time-frequency-domain distortion, including noise, echo, band-pass filtering, 32Kbps@MP3 and others. But they are difficult to deal with large linear speed change distortion which is a frequent audio processing means in the field of television and broadcast. This paper proposes a novel compressed-domain audio recognition algorithm, which can resist linear speed change in the range of-10% to 10% (recognition rate is higher than 90%), via extracting fingerprint after do Fourier-Mellin transform for sub-band energy sequence of MDCT spectrum. This is enough to cope with almost all situations of audio acceleration/deceleration occurred in commercial application. In addition, it shows similarity in other performance compared with existing excellent compressed-domain audio recognition algorithms.