PathBench: Speech Intelligibility Benchmark for Automatic Pathological Speech Assessment

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Halpern, Bence Mark, Tienkamp, Thomas, Abur, Defne, Toda, Tomoki
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866908873185558528
author Halpern, Bence Mark
Tienkamp, Thomas
Abur, Defne
Toda, Tomoki
author_facet Halpern, Bence Mark
Tienkamp, Thomas
Abur, Defne
Toda, Tomoki
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2603_08097
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle PathBench: Speech Intelligibility Benchmark for Automatic Pathological Speech Assessment
Halpern, Bence Mark
Tienkamp, Thomas
Abur, Defne
Toda, Tomoki
Sound
I.2.7
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.
title PathBench: Speech Intelligibility Benchmark for Automatic Pathological Speech Assessment
topic Sound
I.2.7
url https://arxiv.org/abs/2603.08097