Scholarlink

This paper presents a transfer learning approach that leverages cross-linguistic transferability coupled with largely interpretable acoustic features to improve dysarthria detection in Parkinson’s disease. It uses features extracted from sustained phonation of /a/ and diadochokinetic exercises in Czech, Spanish, American English, and Italian. The approach addresses data sparsity in clinical settings and accounts for variability due to age and sex. Transfer learning models outperform monolingual classifiers (i.e. classifiers trained and tested on the same language) in most tests, demonstrating the effectiveness of this approach in overcoming data limitations and enhancing Parkinson-induced dysarthria detection.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhancing Detection of Parkinson-Induced Dysarthria with Cross-Lingual Transfer Learning

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Andreas Rouvalis,
  • Johannes Tröger,
  • Julius Steuer,
  • Juan Rafael Orozco Arroyave,
  • Jan Rusz,
  • Jouni Pohjalainen,
  • Hali Lindsay,
  • Bernd Möbius,
  • Dietrich Klakow

摘要

This paper presents a transfer learning approach that leverages cross-linguistic transferability coupled with largely interpretable acoustic features to improve dysarthria detection in Parkinson’s disease. It uses features extracted from sustained phonation of /a/ and diadochokinetic exercises in Czech, Spanish, American English, and Italian. The approach addresses data sparsity in clinical settings and accounts for variability due to age and sex. Transfer learning models outperform monolingual classifiers (i.e. classifiers trained and tested on the same language) in most tests, demonstrating the effectiveness of this approach in overcoming data limitations and enhancing Parkinson-induced dysarthria detection.