Design And Implementation Of Telephone Audio Classification System

In recent years,China’s telecommunications industry has developed rapidly,and with it,a large amount of telephone audio data has been generated.The use of speech technology to process telephone audio can uncover useful information in it,which is of great significance in areas such as national defense security,voice services,and the prevention and control of telecommunication fraud.As a kind of actual audio,telephone audio often contains many different audio types:normal call speech,abnormal answer tone and music,etc.The direct use of existing speech technology to analyze it cannot achieve satisfactory results,and will bring unnecessary time and arithmetic overhead.This thesis designs and implements a convenient and efficient telephone audio classification system,which can accurately classify telephone audio,guide the subsequent process to adopt the corresponding speech technology,and improve the performance of the speech system.The specific research of this thesis is as follows:The design of the telephone audio classification system was carried out.the overall design and sub-module design of the telephone audio classification system is carried out.A hierarchical classification module is designed to classify telephone audio hierarchically by a support vector machine classifier and a template matching classifier.A new fusion feature is designed to train the support vector machine;a new multi-sample standard template library construction process is designed,which is combined with the time regularization technique to reduce the system timeoverhead.In addition,the data storage layer,data transmission layer,and system management layer are designed to achieve secure data storage and stable system operation.In order to provide users with better access to the system,this thesis also provides a detailed design of the system visualization layer.The implementation of a telephone audio classification system was carried out.Each sub-layer in this system was implemented in turn,and the system was thoroughly tested and analyzed.The classification accuracy of the support vector machine using the new fusion features as classification features reached 96.10%,which is 4.8%higher than the industry average;the template matching classifier was able to reduce the time overhead by 75%and achieve a classification accuracy of 93.61%.After testing,the classification accuracy of this system reached 93.30%and was able to complete the telephone audio classification task,which is a comprehensive,advanced and user-friendly telephone audio classification system.