Voice emotion recognition method based on adversarial semantic erasure

The invention discloses a voice emotion recognition method based on adversarial semantic erasure. The method comprises the steps: (1) obtaining original voice data, and carrying out the preprocessingof the original voice data to obtain an input feature; (2) recognizing the input features by using a voice recognition model comprising a first feature extractor and a first recognizer to obtain recognition semantics, and extracting semantic features output by the feature extractor; and (3) extracting the input features by using a second feature extractor of a voice emotion recognition model comprising a second feature extractor and a second recognizer to obtain voice features, erasing the semantic features from the voice features to obtain emotion features, recognizing the input emotion features by using the second recognizer, and outputting a voice emotion category. According to the method, the voice emotion category can be quickly and accurately recognized based on the input voice data. 本发明公开了一种基于对抗语义擦除的语音情感识别方