OSQA-SI: A Lightweight Non-Intrusive Analysis Model for Speech Intelligibility Prediction
Hsing-Ting Chen, Po-Hsun Sung
This report proposes a non-intrusive speech intelligibility prediction model named Objective Sound Quality Analysis for Speech Intelligibility (OSQA-SI). The model adopts a simple sequential architecture, trained with a minimal number of parameters, and its performance is compared across two types of input acoustic features. Due to its extremely low parameter count, the model is suitable for real-time speech intelligibility assessment in real-world environments on mobile devices.