摘要
Abstract
Speech synthesis technology provides convenience for people's lives,but at the same time there is a risk of being used by criminals,which may cause trust and security issues,and even disrupt social order.Therefore,research is of great signifi-cance to the authenticity identification of speech information.Aiming at the forged speech of various synthesis methods,a variety of speech features is used to explore the impact of different features on the performance of the model,the accuracy of speech forgery is improved by using densely connected residual networks,and an attention mechanism is introduced to improve the model to increase the performance of key features.The proportion of weights can further improve the performance of the model.Experiments have veri-fied that the algorithm achieves 98.3%and 93.4%accuracy in known attacks and unknown attacks,respectively.When using 50%of the training data to train,the algorithm can achieve an accuracy of 92.2%,indicating a densely connected network with attention.It can effectively identify the authenticity of speech and has high performance.关键词
深度学习/语音鉴伪/密集连接网络/注意力机制Key words
deep learning/speech forgery detection/densely connected network/attention mechanism分类
信息技术与安全科学
