发表机构
The Scott-Morgan Foundation; Tecnológico de Monterrey; SMF Labs（斯科特·摩根基金会; 蒙特雷理工学院; SMF实验室）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
本研究针对多语种构音障碍严重度评估，提出基于病因的对比嵌入模型，结合音韵伪标注，在CP、PD和ALS上均优于混合基线，显著提升宏F1分数。
AI 中文摘要
大多数多语种构音障碍严重度系统要么在单一病因-语言对上训练，要么将异质病因合并到一个标签空间中。我们通过四个匹配的HuBERT-base对比嵌入模型，在共享骨干网络、训练方案、语料库注册表和留出评估下测试了这一合并假设：一个混合病因基线和三个针对脑性瘫痪（CP）、帕金森病（PD）和肌萎缩侧索硬化（ALS）的病因特异性模型。训练结合了临床标注语音和来自无训练音韵学剖析方法[1]、[2]的序数伪标签。在说话人不相交、泄漏过滤的留出子集上，病因特异性模型在所有三个目标病因上均优于混合基线：CP（宏F1 0.829对0.676，相对提升+22.6%）、PD（0.715对0.511，+40.0%）和ALS（0.788对0.596，+32.3%）。在CP上，添加144个SAP和44个CDSD伪标签说话人将宏F1从仅临床CP模型的0.786提升至0.829（+4.3个百分点）。每个病因的训练数据涵盖三到七种语言。我们将此定位为标签空间设计选择的受控比较，并讨论伪标签校准、分割卫生和置信度阈值部署作为未来工作的重要局限性。
英文摘要
Most multilingual dysarthria-severity systems either train on a single aetiology-language pair or pool heterogeneous aetiologies into one label space. We test that pooling assumption with four matched HuBERT-base contrastive embedding models under a shared backbone, training recipe, corpus registry and held-out evaluation: one mixed-aetiology baseline and three aetiology-specific models for cerebral palsy (CP), Parkinson's disease (PD) and amyotrophic lateral sclerosis (ALS). Training combines clinically labelled speech with ordinal pseudo-labels from a training-free phonological profiling method [1], [2]. On speaker-disjoint, leakage-filtered held-out subsets, the per-aetiology models outperform the mixed baseline across all three target aetiologies: CP (macro F1 0.829 vs 0.676, +22.6 % relative), PD (0.715 vs 0.511, +40.0 %) and ALS (0.788 vs 0.596, +32.3 %). On CP, adding 144 SAP and 44 CDSD pseudo-labelled speakers lifts macro F1 from 0.786 to 0.829 over a clinical-only CP model (+4.3 percentage points). Training data span three to seven languages per aetiology. We position this as a controlled comparison of label-space design choices and discuss pseudo-label calibration, split hygiene, and confidence-thresholded deployment as important limitations for future work.
CommentsAccepted at IEEE SLT 2026, 13-16 December 2026, Palermo, Sicily. v2: corrected pseudo-labelled subset language count from 'seven' to 'five' in Sec. 2 and Table II caption (arithmetic in the paper unchanged; the enumerated list already had five languages)
