Counterfeit voice detection method based on visual Transform

Abstract

The invention relates to the technical field of voice detection, in particular to a fake voice detection method based on visual Transform, and the method comprises the steps: carrying out the fine adjustment of a target pre-training model through employing marking data, obtaining a self-supervised learning front-end feature extractor, extracting the middle voice feature of a to-be-detected voice, inputting the middle voice feature into a preset rear-end classification network, and carrying out the recognition of the middle voice feature of the to-be-detected voice; performing random dropout processing on a plurality of overlapped blocks of a target size so as to add a classification token and a distillation token to the plurality of processed overlapped blocks, wherein the plurality of overlapped blocks are divided into a plurality of overlapped blocks of a target size and the plurality of overlapped blocks are subjected to random dropout processing; and inputting an average value of the classification token and the distillation token into a classification layer of a Transformer encoder for classification prediction to obtain artifacts in the voice to be detected, and outputting a final voice detection result according to the artifacts. The method still has good detection performance in mismatch scenes possibly encountered in various practical applications, the robustness is high, and good interpretability is achieved.

Images