METHOD FOR DETECTING AUDIO RESISTANCE ATTACKS FOR VOICE INPUT PROCESSING BY AUTOMATED VOICE RECOGNITION SYSTEM, CORRESPONDING DEVICE, COMPUTER PROGRAM PRODUCT AND

The present disclosure relates to a method and apparatus for detecting audio antagonism attacks for voice input (VI) processed by an automatic voice recognition system (ASR). The method comprises: obtaining (12) a transcript (T) obtained by processing an audio signal of a speech input by an automatic speech recognition system; converting (13) the transcript (T) into a synthetic audio signal (SAS); the method comprises the steps of receiving an input audio signal and a synthetic audio signal, extracting (15, 15 ') acoustic features from the input audio signal and the synthetic audio signal and converting them into a sequence of feature vectors (sFV1, sFV2), calculating (17) a dynamic time-warping distance (D) between the converted sequence of feature vectors, and deriving (18) a data segment representing the detection of an audio antagonism attack as a function of a comparison between the dynamic time-warping distance and a predetermined threshold. 本公开涉及一种用于针对由自动语音识别系统(ASR)处理的语音输入(VI)检测音频对抗性攻击的方法和设备。该方法包括：获得(1