Design And Implementation Of Voice Classification System Based On Telephone Audio Big Data

With the global popularity of the Internet and the large-scale construction of 5G networks in China,more comprehensive network coverage makes network activities more convenient and brings greater risks of information leakage.Speculators will use the obtained user contact information to carry out telecommunications fraud and other illegal activities to the mainland people from non-China's mainland areas through communication services such as carrier telephone network systems and VoIP telephone servers around the world.Therefore,the classified management of this part of the telephone audio big data and the analysis of the behavior patterns of telephone audio callers are of great help to intervening in the improper behavior of speculators.This thesis proposes an overall system architecture scheme that can support telephony audio big data scenarios.Aiming at the characteristics of the current telephone audio acquisition system's increasing data level by tens of millions and the audio expression by means of call description record data,a complete system architecture solution that can support business is designed.With the help of the shared-nothing architecture method and virtualization technology,the voice classification system implemented using this architecture solution not only meets the requirements in the current scenario but also presents the characteristics of high availability and easy operation and maintenance.This thesis proposes a classification task scheduling model based on the Go language thread scheduling GPM model.By referring to the GPM model,design classification task scheduling global queues,local queues,and other abstract component models to reduce lock operations during concurrent task scheduling;design abstract methods such as task multi-level distribution and task stealing to improve the load balance of parallel scheduler runtime;Combined with the production-consumer design model,the task scheduling concurrent ability is improved to reduce the overall task execution time.Finally,a voice classification task scheduling model is constructed based on the above design,so that the voice classification task can be scheduled and executed asynchronously,concurrently,and efficiently.This thesis proposes a method for mining and analyzing caller behavior patterns based on the results of voice classification.According to the call description record is mostly used to describe the characteristics of the voice in the call time,call the geographic location,and other meta-information,by analyzing the audio call description record data obtained by the voice classification system,the data is aggregated and analyzed at the time and space level.Obtain the temporal and spatial distribution characteristics of the caller's behavior pattern.Then,by comparing the classification results obtained by the voice classification system with the call audio call description record data items,the characteristics of the migration pattern of the caller's calling area are obtained.Using the above-mentioned behavioral pattern mining and analysis methods,this thesis proposes a total of four caller behavior patterns and displays the data laws behind the call description record data on the voice classification system platform side.Based on the design of the above scheme and model,this thesis implements and tests a voice classification system based on telephone audio big data.By using the classification task scheduling model to construct a task scheduling module,the voice classification task is executed asynchronously.At the same time,it provides users with a management and control platform,including audio data classification results screening,caller behavior pattern analysis results viewing,and other functions.Finally,design and implement the function and performance test plan for the voice classification system implemented in this thesis.