The Research And Application Of Call Center Speech Quality Monitoring System Based On Multi-Modality

In recent years, because of the rapid development of call center, the position of call center in the company is more and more important. Call Center is the communication channel between the company and the customers. The quality of the speech service in the call center directly impacts on the relationship between the company and the customers. With the development of the economy and the changes of the consumption concept, the customers are more concerned about their interests and when their own interests are infringed or they don’t understand their own rights and interests, the customers will acquaint their rights and interests through the call center. It will be directly influence the relationship between the customers and the company if the operators’service quality is very poor. With the development of the call center and the changes of the consumption concept, the number of the call recordings will rapidly increase. Faced with the increase of the number of the voice, the traditional manual mode has not been satisfied with the current situation. Based on the above situation, using the technology of keywords spotting and emotion analysis realizes the automatic monitoring for the quality of speech service to improve the quality of speech service and ease the relationship between the company and the customers.This paper first introduces the two module of the speech quality system: keywords spotting and speech emotion analysis. The keywords spotting module mainly includes pre-process, feature extraction, acoustic model training, keywords spotting. The training of the acoustic model is adaptive training based on an existing acoustic model in order to get a better recognition rate using the trained acoustic model in call center field. Using the SphinxTrain tools train the adaptive acoustic model. The keywords spotting module use the PocketSphinx tools to detect the keywords and the tools are based on the filler model. The module of speech emotion recognition uses the artificial neural network to as an emotion classifier. This paper mainly introduces the algorithm of the model training and the algorithm of the emotion classification and mainly introduces the Back Propagation algorithm to train the neural network.This paper designs an evaluation index for the quality of the speech service based on the keywords spotting and the speech emotion recognition and designs an algorithm that evaluates the speech service quality and integrates it into the system. By checking the evaluation result of the speech service quality, the managers and operators in call center can quickly locate the advantage and disadvantage and then improve the quality of the speech service.