Method of training grouping model for voice grouping, and voice noise reduction method

Abstract

The embodiment of the invention provides a method for training a grouping model for voice grouping, and a voice noise reduction method. The voice noise reduction method comprises the steps: B1, obtaining an input audio, and extracting the grouping features of the input audio; B2, inputting the grouping features of the input audio into a grouping model, and predictively outputting the label of theinput audio; and B3, according to the label of the input audio, taking the transcription text obtained by the audio processed by the optimal noise reduction strategy of the group corresponding to thelabel as the transcription text of the input audio. According to the technical scheme provided by the embodiment of the invention, the group to which the input audio belongs can be predicted accordingto the anti-disturbance of the input audio, and the noise reduction strategy suitable for the group is selected for noise reduction, so that the transcription error rate can be reduced, and the transcription quality of a clean sample cannot be influenced while the anti-transcription is recovered to the original transcription.

Description

technical field [0001] The present invention relates to the field of voice recognition, in particular to the field of voice recognition related to anti-attack processing, more specifically, to a method for training a grouping model for voice grouping and a voice noise reduction method. Background technique [0002] Deep neural networks (DNNs) have achieved impressive results in various artificial intelligence applications, including image classification, natural language processing, and speech recognition. In some domains, the performance of DNNs has reached or even surpassed human performance. Therefore, DNN is widely used in some safety-sensitive tasks that require high robustness of the model. [0003] However, in recent years, deep neural networks have been seriously threatened by adversarial attacks. Adversarial attacks generate adversarial samples by adding subtle noise to legitimate samples. While humans cannot recognize adversarial examples, they can make deep neu...

AI Technical Summary

Problems solved by technology

Existing defense strategies treat all speech adversarial samples in the same way, but are inappropriate because the added perturbations vary in magnitude

Images