Two-Stage Frequency-Temporal Filtering for Effective Audio Fingerprinting of Sound Recordings

Abstract

Sound recordings are commonly distorted by channel and background noise. The performance of audio identification is mainly degraded by that noise. For an audio fingerprinting system, Haitsma and Kalker introduced a robust and efficient audio hashing scheme applying high-pass filtering (differentiation) to the frequency-temporal sequence of perceptual filter-bank energies. However, the robustness of the audio fingerprinting scheme is still important in real noisy environments. This paper introduces some alternatives of frequency-temporal filtering for effective audio fingerprinting of sound recordings in real environments. As the alternative to frequency filtering, a type of band-pass filter, instead of a high-pass filter, is used to enhance robustness to background noise in a real situation. As the alternative to temporal filtering, RASTA, instead of a high-pass filter, is used for normalizing sound recording conditions. As well, this paper introduces a two-stage audio fingerprinting scheme to achieve synergy of the combination of frequency-temporal filtering. Experimental results show that the proposed method is effective for sound recordings in real environments.