基于声纹大数据的防电信诈骗应用研究

AbstractNowadays,the number of telecommunication network fraud cases isincreasing day by day,which seriously endangers the property security ofthe people.With the continuous development of communication means,the amount of communication data is also increasing.It is a seriouschallenge to the real-time and accuracy of the telecommunication fraudprevention system.The traditional anti fraud system,which uses neuralnetwork as support technology,has limited resolution accuracy forfraudulent phones,and it is difficult to meet the needs of a large numberof voice samples.In this paper,we propose an anti fraud framework based onvoiceprint big data.The voiceprint feature library is formed by collectingand extracting voiceprint features from voice files.When receiving thevoiceprint authentication request,the distance between the voiceprint tobe authenticated and the voiceprint feature in the voiceprint featurelibrary is calculated to determine whether the current authenticatedvoiceprint is from the fraud phone,so as to implement early warning forthe telecom fraud.The method proposed in this paper improves theprocessing speed and running efficiency of the program in the case oflarge amount of data by parallel extraction of voiceprint features.Key words:Telecommunications Fraud;Deep Learning:VoiceprintRecognition