PrefSQA: Pairwise Preference Prediction for Speech Quality Assessment and the Critical Role of High Quality Datasets

The article introduces PrefSQA, a novel approach for speech quality assessment that utilizes pairwise preference prediction to mitigate the limitations of mean opinion scores (MOS) affected by rater variability. The model incorporates uncertainty-aware logits, an impairment attention head, and a module for non-matching-reference comparisons, demonstrating improved performance on high-quality preference datasets compared to traditional methods. This advancement is significant for practitioners as it emphasizes the importance of high-quality datasets in training models for more reliable speech quality assessments.