On the Use of LSTM-RNN for Detecting Audio Spoofing Attacks
Lian Fen Huang, Jinhong Zhao · 2022
This paper proposes a long short-term memory recurrent neural network (LSTM-RNN) for detecting audio spoofing attacks. Many algorithms, including deep learning technology, have made a lot of achievements in this field. However, the automatic speaker verification (ASV) system is still vulnerable to a spoofing attack. Therefore, in this paper, we analyze the architecture of LSTM-RNN and apply the LSTM-RNN method. Considering the application of LSTM-RNN in speech recognition, we believe that LSTM-RNN can also play a prominent role in this application. To verify our proposed approach, we take the ASVspoof 2015 dataset for the test. We use two kinds of features in the experiment: Mel-scale Frequency Cepstral Coefficients (MFCC) and Log-FBank. Among them, the Log-FBank feature system can achieve better performance. The experimental results show that the ASV system can be used in practice with the LSTM method.