Scholarlink

One of the main obstacles facing the research in the automatic detection of voice pathology is the lack of data. In this work, we try to tackle this problem by the use of self-supervised, and weakly supervised, pre-trained models, as feature extractors. We also perform cross-database experiments to test the ability of the proposed system to generalise across databases. UA-Speech and TORGO databases are used to carry the experiments, and the results are compared to baseline features. Two classifiers were used: SVM and a neural network. The pre-trained models achieved the best results on both databases when the neural network was used as classifier.

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

Using Embeddings of Pre-trained Models for Cross-Database Dysarthria Detection: Supervised vs. Self-supervised Approach

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Sally Ismail,
  • Margarita Anastassova,
  • Mehdi Boukallel,
  • Christian Bolzmacher,
  • Mehdi Ammi

摘要

One of the main obstacles facing the research in the automatic detection of voice pathology is the lack of data. In this work, we try to tackle this problem by the use of self-supervised, and weakly supervised, pre-trained models, as feature extractors. We also perform cross-database experiments to test the ability of the proposed system to generalise across databases. UA-Speech and TORGO databases are used to carry the experiments, and the results are compared to baseline features. Two classifiers were used: SVM and a neural network. The pre-trained models achieved the best results on both databases when the neural network was used as classifier.