Replay Attack Anaysis Based on Acoustic Parameters of Overall Voice Quality

摘要

The replay attack detection has played an important role in the automatic speaker verification (ASV). From the perspective of the pronunciation mechanism, this paper proposes seven acoustic parameters reflecting the overall voice quality to improve the effectiveness of the replay attacks detection, including slope, SL, mACF, HNR, NHR, GNE and CPPS. The set of continuous speech samples from 108 subjects taken from three kinds of text content are used to evaluate the applicability and multi-condition sensitivity analysis of these parameters. The result confirms that five acoustic parameters including SL, mACF, HNR, NHR and CPPS provide reliable measurements for discriminating authentic recording and its playback versions.