Automatic voice quality evaluation method of IVR service in call center based on Stacked Auto Encoder

Abstract

Abstract The basic features extracted by traditional methods for speech quality evaluation are not clear, which leads to the small correlation coefficient of subjective and objective evaluation value. Therefore, an automatic voice quality evaluation method for IVR service in call center based on stackable automatic encoder is proposed. All kinds of devices are used to simulate the real use of IVR service voice of call center and collect IVR service voice of call center. According to the process of sampling quantization frame pre emphasis window processing, the IVR service voice data of call center is pre processed. Based on the structure of stackable automatic encoder, the reconstruction process of coding and decoding is designed to extract the basic features of business speech. BP neural network is introduced to establish an automatic speech evaluation model to evaluate speech quality automatically. Experimental results: compared with the traditional method, the average correlation coefficients of subjective evaluation value and objective evaluation value are 0.023517 and 0.02258 respectively, and the average deviation of correlation coefficient is 0.048775 and 0.03485 respectively.