Content area
Abstract
This thesis deals with the subjective testing of both speech quality and speech intelligibility, investigates the existing methods, record their main features, as well as advantages and disadvantages. The work also compares different tests in terms of various parameters and provides a modern solution for existing subjective testing methods.
The first part of the research deals with the repeatability of subjective speech quality tests provided in perfect laboratory conditions. Such repeatability tasks are performed using Pearson correlations, pairwise comparison, and other mathematical analyses, and are meant to prove the correctness of procedures of provided subjective tests. For that reason, four subjective speech quality tests were provided in three different laboratories. The obtained results confirmed that the provided tests were highly repeatable, and the test requirements were strictly followed.
Another research was done to verify the significance of speech quality and speech intelligibility tests in communication systems. To this end, more than 16 million live call records over VoIP telecommunications networks were analyzed. The results confirmed the primary assumption that better user experience brings longer call durations. However, alongside the main results, other valuable conclusions were made.
The next step of the thesis was to investigate the parallel task technique, existing approaches, their advantages, and disadvantages. It turned out that the majority of parallel tasks used in tests were either physically or mentally oriented. As the subjects in most cases are not equally trained or intelligent, their performances during the tasks are not equal either, so the results could not be compared correctly.
In this thesis, a novel approach is proposed where the conditions for all subjects are equal. The approach presents a variety of tasks, which include a mix of mental and physical tasks (laser-shooting simulator, car driving simulator, objects sorting, and others.). Afterward, the methods were used in several subjective speech quality and speech intelligibility tests. The results indicate that the tests with parallel tasks have more realistic values than the ones provided in laboratory conditions.
Based on the research, experience, and achieved results, a new standard was submitted to the European Telecommunications Standards Institute with an overview, examples, and recommendations for providing subjective speech quality and speech intelligibility tests. The standard was accepted and published under the number ETSI TR 103 503.