Scholarlink

<p>Parkinson’s disease (PD) is often diagnosed following a complete physical examination by a clinician that includes the patient’s medical history, neurological examination, evaluation of motor symptoms, and additional supportive tests with precise diagnostic criteria. However, there is no proven approach to detect PD. Approximately 89% of individuals with PD encounter difficulties with their ability to speak. This study investigates the use of a combined approach involving empirical wavelet transform (EWT) and Hilbert transform (HT) to analyze voice tremor in patients with PD. The speech signal is represented in the time–frequency (TF) domain using EWT and HT. From this TF representation, energy and entropy features are extracted from each frequency component. The effectiveness of these features is evaluated using vowel and word samples from the PC-GITA dataset. The optimal feature set is selected through a genetic algorithm, and classification is performed using a support vector machine (SVM). The proposed method achieved a classification accuracy of 93% using isolated words /atleta/ and /petaka/.</p>

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

Parkinson’s disease detection from speech using combination of empirical wavelet transform and Hilbert transform

  • 期刊论文

  • 发表日期：

    2025年2月27日

查看全文
  • Sachin Anap,
  • Satish Jondhale,
  • Balasaheb Agarkar,
  • Sachin Chaudhari

摘要

Parkinson’s disease (PD) is often diagnosed following a complete physical examination by a clinician that includes the patient’s medical history, neurological examination, evaluation of motor symptoms, and additional supportive tests with precise diagnostic criteria. However, there is no proven approach to detect PD. Approximately 89% of individuals with PD encounter difficulties with their ability to speak. This study investigates the use of a combined approach involving empirical wavelet transform (EWT) and Hilbert transform (HT) to analyze voice tremor in patients with PD. The speech signal is represented in the time–frequency (TF) domain using EWT and HT. From this TF representation, energy and entropy features are extracted from each frequency component. The effectiveness of these features is evaluated using vowel and word samples from the PC-GITA dataset. The optimal feature set is selected through a genetic algorithm, and classification is performed using a support vector machine (SVM). The proposed method achieved a classification accuracy of 93% using isolated words /atleta/ and /petaka/.