Research on Speech Playback Spoof Detection Based on ASV Spoof 2021

Speech playback spoofing refers to the use of pre-recorded and authenticated speech of the target speaker, through some playback devices to play the recorded speech and send it to the ASV system. As a result, the error acceptance rate of ASV system is significantly increased. Based on this, this paper designs a speech playback spoofing detection system model based on deep neural network to distinguish real audio from replayed audio. Reduce the error acceptance rate of ASV systems. To be specific, firstly, the audio features of the original speech are extracted by the convolutional neural network and combined with the MFCC features. Then a speech playback detection model is designed based on the cyclic neural network and a speech playback detection system model is built. Experimental results show that, compared with the existing algorithms, The voice playback spoofing detection system based on ASV spoof 2021 can well distinguish real voice from replayed voice and effectively avoid replayed voice from entering the ASV system.