Scholarlink

<p>In this paper, the authors focused on recognizing tonsillectomy cases using statistical and acoustic measures together with different feature extraction techniques. The centroid, spread, skewness, kurtosis, jitter, shimmer, and F0 range analysis showed considerable differences between the pathological case and the normal case. Pathological cases in tonsillectomy showed noteworthy features that affect the speech signal, considered as an effect of the surgical process. Different feature extraction methods were examined in the study, including the linear prediction coding (LPC) and discrete wavelet transform (DWT), which was referred to as LPCFDWT, formants with DWT, which was termed as FADWT, and Mel-frequency cepstral coefficient (MFCC). The use of LPCFDWT with a fine Gaussian SVM classifier has a remarkable tonsillectomy recognition accuracy of 96.90% on Arabic vowels ('A' and 'O'). The results outperform several published machine learning methods investigated in this study. An original database of Arabic sentences and words was recorded and involved in the investigation, showing less ability to model the tonsillectomy effect compared to vowels. The word accuracy was 79.10, and sentence accuracy was 81.70, meaning that there was a notable difference compared to the vowel level accuracy. Furthermore, the method of principal component analysis (PCA) was used for feature selection, which proves that tonsillectomy effects can be captured with a small number of features by the presented method. The study sought to compare various classifiers against the fine Gaussian SVM, which affirmed the method’s supremacy over all the methods that were tested. The study’s contributions lie in evaluating several feature extraction methods, classifiers, and original recordings of the cases of tonsillectomy, as well as in showing the efficiency difference of vowels, words, and sentences. As for further work, more cases are planned to be organized to enlarge the research database with more detailed outcomes, excellent diagnoses, and clinical innovation.</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

Analyzing the impact of tonsillectomy on vocal signals using statistical and acoustic methods

  • 期刊论文

  • 发表日期：

    2026年3月9日

查看全文
  • K. Daqrouq,
  • R. A. Alharbey,
  • A. Alkhateeb,
  • E. Nöth

摘要

In this paper, the authors focused on recognizing tonsillectomy cases using statistical and acoustic measures together with different feature extraction techniques. The centroid, spread, skewness, kurtosis, jitter, shimmer, and F0 range analysis showed considerable differences between the pathological case and the normal case. Pathological cases in tonsillectomy showed noteworthy features that affect the speech signal, considered as an effect of the surgical process. Different feature extraction methods were examined in the study, including the linear prediction coding (LPC) and discrete wavelet transform (DWT), which was referred to as LPCFDWT, formants with DWT, which was termed as FADWT, and Mel-frequency cepstral coefficient (MFCC). The use of LPCFDWT with a fine Gaussian SVM classifier has a remarkable tonsillectomy recognition accuracy of 96.90% on Arabic vowels ('A' and 'O'). The results outperform several published machine learning methods investigated in this study. An original database of Arabic sentences and words was recorded and involved in the investigation, showing less ability to model the tonsillectomy effect compared to vowels. The word accuracy was 79.10, and sentence accuracy was 81.70, meaning that there was a notable difference compared to the vowel level accuracy. Furthermore, the method of principal component analysis (PCA) was used for feature selection, which proves that tonsillectomy effects can be captured with a small number of features by the presented method. The study sought to compare various classifiers against the fine Gaussian SVM, which affirmed the method’s supremacy over all the methods that were tested. The study’s contributions lie in evaluating several feature extraction methods, classifiers, and original recordings of the cases of tonsillectomy, as well as in showing the efficiency difference of vowels, words, and sentences. As for further work, more cases are planned to be organized to enlarge the research database with more detailed outcomes, excellent diagnoses, and clinical innovation.