The PartialSpoof Database and Countermeasures for the Detection of Short Fake Speech Segments Embedded in an Utterance

Automatic speaker verification is susceptible to various manipulations and spoofing, such as text-to-speech synthesis, voice conversion, replay, tampering, adversarial attacks, so on. We consider a new spoofing scenario called "Partial Spoof" (PS) in which synthesized or transformed speech segments are embedded into bona fide utterance. While existing countermeasures (CMs) can detect fully spoofed utterances, there need for their adaptation extension the PS scenario. propose improvements construct significantly more accurate CM that locate short-generated at finer temporal resolutions. First, we introduce newly developed self-supervised pre-trained models enhanced feature extractors. Second, extend our PartialSpoof database by adding segment labels Since short be attackers of variable length, six different resolutions considered, ranging from 20 ms large 640 ms. Third, enables simultaneous use segment-level well utterance-level execute utterance- detection same time. also show proposed capable detecting utterance level with low error rates related logical access (LA) The equal on ASVspoof 2019 LA were 0.77 0.90%, respectively.