A Speech Quality Classifier based on Signal Information that Considers Wired and Wireless Degradations

Author(s)

Militani, Davi ; Begazo, Dante Coaquira ; Rosa, Renata ; Rodriguez, Demostenes Z. ; Begusic, D ; Rozic, N ; Radic, J ; Saric, M

• Document type: Journal article
• Source: 2019 27TH INTERNATIONAL CONFERENCE ON SOFTWARE, TELECOMMUNICATIONS AND COMPUTER NETWORKS (SOFTCOM); v. N/A, p. 6-pg., 2019-01-01.

Abstract

There are many factors that can affect the users' quality of experience (QoE) in Voice over IP (VoIP) services, especially in wireless networks. In speech communication systems there are different impairment factors because physical phenomena that occur in the wired and wireless transmission channel. In this context, a non-intrusive speech quality classifier based on Discriminative Restricted Boltzmann Machines (DRBM) is proposed, which considers speech signals degradations caused by wired and wireless network degradations. To accomplish this goal, a test scenario is implemented, in which the speech signal is coded by the AMR-WB and transmitted using both a lossy wired channel and the Rayleigh fading model. Also, different modulation schemes and channel degradations, such as packet loss rate, signal-to-noise ratio and Doppler shifts are implemented. As a result, a speech database is built that is used to train different machine learning algorithms. Experimental results demonstrated that the DRBM reached the best results. Performance assessment results show that the proposed classifier based on DRBM overcomes the current standardized algorithm described in ITU-T Rec. P.563. (AU)

• FAPESP's process: 15/24496-0 - Evaluation of the service of communication operators using the voice Quality Index
• Grantee: Demostenes Zegarra Rodriguez
• Support Opportunities: Regular Research Grants
• FAPESP's process: 18/26455-8 - Audio-Visual Speech Processing by Machine Learning
• Grantee: Miguel Arjona Ramírez
• Support Opportunities: Regular Research Grants