A Simple Neural Network Based Countermeasure for Replay Attack
Wenfeng Pang, Qian He · 2017
Speaker verification has been applied broadly, but the characteristic of being spoofed easily is one of the obstructions for expanding its applications. Among various form of spoofing, replay is an efficient method, even humans can not easily classify the replay audio from the legal ones. In this paper, CNN-RNN neural network is proposed to detect replay spoofing, furthermore, simpler architectures are investigated by pruning and retraining. Experiments evaluated with accuracy, FAR and FRR showed that the proposed approach can work effectively for detecting replay spoofing.