Realization Of Call Center Based On ASR And Research On Customer Classification

Along with the process of applying Customer Relationship Management system for medium or small companies, Building a functional call center is an important way to enhance the competition of the company as call center is the unified contacts of the companies and customers. The Interactive Voice Response (IVR) system is an important part of the call center.In order to strengthen the expandability of the system, Interactive Voice Response (IVR) system adopts the development proposal that is irrelevant with the specific working flow to build a system design framework. The framework adopted TreeView component to display the working flow, the component is made up of nodes and each node represents one operation in the working flow. The structure design module realizes the adding and deleting of the nodes. The content design module appoints the operation type and parameters of the nodes. The running module transforms the flow path into the controlling code of the speech card and the management module manages several working flow path. The function of registering, checking out, quering and modifying the message of the hotel is realized by analying the function of the hotel message management system. Under the former framework, the function of the hotel call center is realized, which contains telephone leaving message, minibar register, waking up service and updating of the room state.As there are shortcomings of the IVR system that use the keypress of the telephone, the automatic speech recognition algorithm for isolated words based on Discreted Hidden Markov Model (DHMM) is researched. In the research of the endpoint detection, the distinguishment of voiced and unvoiced signal of the maximum of the short autocorrelation function and the distance of the LPC is analysised, which shows that the former can distinguish the voiced and unvoiced signal better, so a double threshold endpoint detection method based on short autocorrelation function is proposed. The extraction methods of the Linear Prediction Cepstrum Coefficient (LPCC) and the Mel-Frequency Cepstrum Coefficient (MFCC) are researched. The training and recognition processes of the recognition system for isolated words based on Discrete HMM (DHMM) are analysised. At last the experiment results and the analysis are given. The ASR algorithm is also used in the IVR system which realizes the IVR system which has the function of speech recognition. The core of the Customer Relationship Management (CRM) is the guidance of the customers, which means providing personal services for customers. The customer classification method is the key technology for carrying out the CRM. The call center has collected and accumulated a lot of data of the customers. The customers can be divided into different types using currentily available data. This paper proposes an improved BP neural network algorithm based on rough set and genetic algorithm. First this algorithm carries attribute reduction on original data sets, which optimizes the input variable of the BP neural network; then makes use of the global searching characteristic of genetic algorithm, which optimizes the original weight and threshold of the BP neural network. At last the improved BP neural network is applied to the customer classification and good classification result has been gotton.