Date of Award
Masters of Science (Research)
Institute of Technology Tralee
Dr. Pat Doody
Communication systems are undergoing constant and rapid innovation, both at the design stage and in the field. This in turn has led to an inereasing need for fast, efficient, portable and economic methods for the testing of these systems. For voice carrying communication systems the quality of the transmitted voice that the system produces is a large factor in the overall performance rating of the system. This measure is known as the ‘Quality of Voice’ (QoV) and can be evaluated either subjectively or objectively.
Speech quality is a complex subjective phenomenon that can be best quantified by subjective testing. A subject QoV measurement requires a ‘listener’ to rate a sample of speech produced by the system. To achieve aceurate results an average rating, or Mean Opinion Score (MOS), must be found from a large panel of listeners. This results in subjective QoV testing being a highly expensive and time consuming process to conduct.
Objective methods of QoV estimation attempt to predict the results a panel of listeners would produee when presented with a given sample of speech. Objective QoV estimation teehniques comprises both Intrusive and Non-Intrusive methods. Non-intrusive QoV estimation methods involve an automated algorithm taking account of various speech impairment factors based on the operational parameters of the system under test. These parameters are then used predict the distortion levels introduced to the speech during transmission by the system and thereby an estimate of the QoV eapabilities of the system can be made.
Intrusive QoV estimation methods involve a comparison between an original speech sample and a resulting speech sample which has been degraded by transmission through the system. By performing a distance measure between the original and degraded speech samples, an estimation of the QoV capabilities of the system can be made.
This project aims to create an objective method for the estimation of the voice transmission capabilities of a system using Artificial Intelligence (A.I.) techniques. Algorithms for both Intrusive and Non-intrusive objective QoV estimation through machine learning will be investigated during this thesis. It is hoped that the application of A.I. techniques to objective QoV estimation algorithms will improve the efficiency and economy of communication system testing.
Riordan, Daniel, "Lightweight Objective Quality of Voice Estimation Through Machine Learning" (2008). Theses [online].
Available at: https://sword.cit.ie/allthe/323