Speech recognition attack defense method based on PSO algorithm

Abstract

The invention discloses a speech recognition attack defense method based on a PSO algorithm. The method comprises the following steps of (1) preparing an original audio data set, and dividing the original audio data set into a pre-training data set, a test data set and a disturbance data set used for generating an adversarial sample; (2) training a speech recognition model, wherein the speech recognition model is built, related parameters of the model are initialized, the speech recognition model is trained by using the pre-training data set, and the test data set is used to test the recognition accuracy of the model; (3) attacking the speech recognition model, wherein an attack method based on the PSO algorithm is built, a fitness function and related parameters of the PSO algorithm are set, and an optimal adversarial sample generated by the attack method can be recognized by mistake and is not recognized by the human ear; and (4) performing adversarial training on the speech recognition model, wherein the adversarial sample generated in the step (3) is added into the pre-training data set, and the speech recognition model is re-trained, so that the speech recognition model has the capability of defending attack of the adversarial sample and the safety and stability of the model are improved.

Description

technical field [0001] The invention relates to a speech recognition attack defense method based on a PSO algorithm. Background technique [0002] With the rapid development of machine learning in recent years, speech recognition technology has gradually become one of the key technologies in people's lives. By extracting relevant information from the input voice and audio and processing it, the smart device can transcribe the correct text content and even realize communication and interaction with people. With the unremitting efforts of researchers, speech recognition technology has gradually matured, the recognition accuracy has been continuously improved, and the application range has been continuously expanded, which has brought great convenience to people's production and life. However, the latest scientific research shows that the speech recognition model based on deep learning has certain defects and is easily interfered by adversarial examples. Attackers cause speec...

AI Technical Summary

Problems solved by technology

[0006] Aiming at the security problem that the speech recognition model in the prior art is vulnerable to adversarial attacks, the present invention provides a speech recognition attack defense method based on the PSO algorithm, which can efficiently generate adversarial audio samples with transcription errors, and through the adversarial Training improves the model's ability to defend against adversarial examples

Images