Speech speed multiplication attack detection method based on rhythm features and random forest classifier

Abstract

The invention discloses a voice speed multiplication attack detection method based on rhythm features and a random forest classifier, and belongs to the technical field of voice recognition and safety. Acquiring an audio data set, wherein the audio data set comprises normal audio and double-speed confrontation audio; extracting jitter features, tremolo features and harmonic noise ratio features of all audios in the audio data set to form feature vectors; and training a random forest classifier by using the feature vectors of the normal audio and the multi-speed confrontation audio, and carrying out voice multi-speed attack detection by using the trained random forest classifier. The voice double-speed spoofing attack can be efficiently detected through the existing microphone and voice hardware of the voice recognition system, and the method has the advantages of being low in cost and high in attack detection accuracy, can be used for safety protection of the voice recognition system on intelligent equipment such as a mobile phone and has wide requirements and application prospects.

Description

technical field [0001] The invention belongs to the field of speech recognition technology and security technology, and in particular relates to a speech double-speed attack detection method based on prosodic features and random forest classifiers. Background technique [0002] Automatic Speech Recognition (ASR) systems can recognize speech and output speech recognition text. Existing popular ASR systems include open source systems (such as Kaldi and DeepSpeech) and commercial systems (such as Google Cloud Speech-to-Text, Baidu ASR, and iFlytek). For the input audio, the ASR system first performs signal processing to reduce noise and remove irrelevant frequency components; then divide the processed audio signal into short segments to extract features such as Mel Frequency Cepstral Coefficients (MFCC); finally, use the extracted features , the most likely word sequence is inferred by a pre-trained speech recognition model. [0003] Time-scale Modification (TSM) refers to an...

AI Technical Summary

Problems solved by technology

[0005] At present, many related researches have protected against voice attacks by detecting the noise and distortion introduced during the generation of voice adversarial samples. However, this type of detection method is difficult to detect voice double-speed attacks without adding noise.

Images