PathBench: Speech Intelligibility Benchmark for Automatic Pathological Speech Assessment
About
Automatic speech intelligibility assessment is crucial for monitoring speech disorders and therapy efficacy. However, existing methods are difficult to compare: research is fragmented across private datasets with inconsistent protocols. We introduce PathBench, a unified benchmark for pathological speech assessment using public datasets. We compare reference-free, reference-text, and reference-audio methods across three protocols (Matched Content, Extended, and Full) representing how a linguist (controlled stimuli) versus machine learning specialist (maximum data) would approach the same data. We establish benchmark baselines across six datasets, enabling systematic evaluation of future methodological advances, and introduce Dual-ASR Articulatory Precision (DArtP), achieving the highest average correlation among reference-free methods.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Pathological speech intelligibility assessment | COPAS Word | PCC0.6 | 36 | |
| Pathological speech intelligibility assessment | COPAS Sentence | PCC0.36 | 36 | |
| Pathological speech intelligibility assessment | UASpeech Word | PCC0.98 | 24 | |
| Pathological speech intelligibility assessment | TORGO Sentence | PCC0.92 | 24 | |
| Pathological speech intelligibility assessment | NeuroVoz Sentence | PCC0.79 | 24 | |
| Pathological speech intelligibility assessment | EasyCall Word | PCC0.71 | 24 | |
| Pathological speech intelligibility assessment | TORGO Word | PCC0.57 | 24 | |
| Pathological speech intelligibility assessment | EasyCall Sentence | PCC0.62 | 24 | |
| Pathological speech intelligibility assessment | YouTube (YT) Sentence | PCC0.78 | 10 |