Voice Quality Assessment in Communication Services using Deep Learning

Abstract

IP networks have provided many services, such as telephone communications. However, the Packet Loss Rate (PLR) can occur on IP networks and it can affect the users Quality of Experience (QoE). In this case, it is important to perform the assessment of the speech quality and to implement a methodology to predict speech quality. Thus, this paper presents a nonintrusive speech quality model based on Hybrid Discriminative Restricted Boltzmann Machines (HDRBM), in order to identify speech quality classes. A speech database with different PLRs was built and a quality index was found for each file. The experimental results of the performance assessment showed that the proposed model based on HDRBM overcame the ITU-T recommendation P.563. Subjective tests presented 97.11% of precision using the proposed Speech Quality Classifier performed by the HDRBM approach.