The classifier judged this item not relevant to AI security, so it is left out of lists, feeds, the API and dataset releases. If that is wrong, report a correction on this page.
InfoResearchPeer-reviewed
Detecting Partially Spoofed Utterances Without Segment Annotation Through Fake Segment Mining-Based Graph Neural Networks
- Published
- Record updated
Summary
Researchers address partially spoofed speech detection, where an utterance mixes fake and bona fide segments and no segment-level labels exist. They propose FMG, a Fake segment Mining based Graph neural network that models inter-segment temporal dependence through ATD and ATC branches and identifies representative fake segments with an FSM module using attentive pooling and prototype and entropy losses. Experiments on spoofed-speech datasets show it outperforms compared models.