ASVspoof 2019: Future Horizons in Spoofed and Fake Audio Detection

Abstract

ASVspoof, now in its third edition, is a series of community-led challenges\nwhich promote the development of countermeasures to protect automatic speaker\nverification (ASV) from the threat of spoofing. Advances in the 2019 edition\ninclude: (i) a consideration of both logical access (LA) and physical access\n(PA) scenarios and the three major forms of spoofing attack, namely synthetic,\nconverted and replayed speech; (ii) spoofing attacks generated with\nstate-of-the-art neural acoustic and waveform models; (iii) an improved,\ncontrolled simulation of replay attacks; (iv) use of the tandem detection cost\nfunction (t-DCF) that reflects the impact of both spoofing and countermeasures\nupon ASV reliability. Even if ASV remains the core focus, in retaining the\nequal error rate (EER) as a secondary metric, ASYspoof also embraces the\ngrowing importance of fake audio detection. ASVspoof 2019 attracted the\nparticipation of 63 research teams, with more than half of these reporting\nsystems that improve upon the performance of two baseline spoofing\ncountermeasures. This paper describes the 2019 database, protocols and\nchallenge results. It also outlines major findings which demonstrate the real\nprogress made in protecting against the threat of spoofing and fake audio.\n