Adversarial sample attack-resistant speech recognition model training method

Abstract

The invention discloses an adversarial sample attack-resistant speech recognition model training method. The method comprises the following steps of A, selecting a data set consisting of speech files; B, sampling the speech files to obtain a collection tensor; C, extracting a feature tensor from the acquired tensor; D, inputting the feature tensor, the Chinese characters corresponding to the speech files and the training parameters into a convolutional neural network model, mapping the output tensor into the probability of outputting different Chinese characters at each time step by using a CTC function in combination with an acoustic model and a language model, and then selecting the maximum value of the probability to transcribe the corresponding Chinese character as an actual output value; finally, calculating a deviation value between the target value and the actual output value and recording the deviation value as loss; and E, using a back propagation method to update the weight parameter of the convolutional neural network model, and updating the training parameter input into the convolutional neural network model while updating the weight parameter. The method can reduce the training calculation amount and the training time, and improves the capability of resisting malicious attacks.

Description

technical field [0001] The invention relates to the field of speech recognition, in particular to a speech recognition model training method for resisting adversarial sample attacks. Background technique [0002] With the development of deep learning and neural network research, automatic recognition application scenarios are becoming more and more extensive. In the field of speech, deep learning is reshaping the way we interact with machines, such as the personal assistants that are widely used in smartphones (Apple's Siri, Google's Assistant). These systems recognize and execute user commands by running speech recognition models. In fact, the research on automatic speech recognition predates the advent of computers, and speech synthesis and recognition technology can be traced back to the original vocoder. After the new century, artificial intelligence network research has emerged in the field of speech recognition. Most artificial intelligence networks use multi-layer p...

AI Technical Summary

Problems solved by technology

[0003] While recent studies have shown that neural networks are vulnerable to adversarial attacks

This problem also exists in the field of speech recognition. The attacker adds a slight disturbance to the audio, which will cause the neural network to input completely different values, but the human ear cannot recognize the slight disturbance.

As the voice confrontation attacks that have appeared in recent years have become more and more aggressive and more and more types, the security issues in the field of voice recognition have become prominent, and it has also laid certain hidden dangers for the large-scale commercialization of voice recognition technology.

Images