Article
Dissecting Speech Quality: Investigating the Interplay between Overall Quality and Dimensional Perceptions (en)
* Presenting author
Abstract:
This study explores the intricate relationship between overall speech quality and its constituent dimensions: noisiness, discontinuity, coloration, and loudness. Employing statistical methodologies and Python-based machine learning models, diverse databases with varying degradation conditions are exhaustively analyzed. The study dissects the dependencies and interactions among overall quality scores and dimensional perceptions. Through detailed modeling, the underlying factors shaping human auditory perception are elucidated, offering crucial insights into the hierarchy and relative significance of these quality dimensions. Notably, nuanced patterns emerge: not all quality dimensions equally impact the overall quality score. Additionally, the diverse nature of degradation conditions influences individual quality dimensions disparately. Clear relationships are identified and established between the four dimensions and the overall speech quality, providing valuable insights into their interplay. These findings have significant implications for tailored audio processing algorithms and communication system designs, enhancing user experience across applications. This research not only advances speech quality assessment methodologies but also substantially contributes to the theoretical foundations of psychoacoustics, enriching the academic discourse in the field.