Voice metrics for discourse quality analysis

In this paper, we introduce new discourse quality metrics and an evaluation method, and furthermore, we provide a pilot implementation and evaluate it in a specific use case - that of public speaking. Voice analysis and discourse quality metrics are important topics as they can solve various problems of modern-day life. In public speaking trainings, it is important to be able to analyze voice and discourse, to offer the speaker valuable feedback and information on how to improve when speaking in public. We identified several papers that study this issue, in which rhythm, tone, fluency and clarity, as well as biological signals, are analyzed, to detect emotions or anxiety. We defined a set of voice quality metrics that can be implemented based on any speech recognition system. Our pilot implementation showcases the difference of using various Speech-to-Text (STT) APIs over the proposed metrics and how these affect the discourse evaluation process.