Detection of vocal fold nodules in children based on multi-band cepstral features of sustained vowels
10.3969/j.issn.1006-7299.2025.04.001
- VernacularTitle:基于持续元音多波段倒谱特征的儿童声带小结检测
- Author:
Jianhan LEI
1
;
Yang LIU
;
Boquan LIU
;
Hengxin LIU
Author Information
1. 上海交通大学人文学院(上海 200030)
- Publication Type:Journal Article
- Keywords:
Vocal fold nodules;
Pediatric voice disorders;
Acoustic features;
Mel-frequency cepstral co-efficients
- From:
Journal of Audiology and Speech Pathology
2025;33(4):307-311
- CountryChina
- Language:Chinese
-
Abstract:
Objective To propose an effective objective acoustic voice assessment method for detecting vocal fold nodules in children.Methods Continuous vowel sounds/a/from 48 children with vocal fold nodules and 40 healthy chil-dren were analyzed using multi-band cepstral analysis.Thirteen Mel-frequency cepstral coefficients(MFCC),five cepstral peak values,and six cepstral distances were extracted as subband features.An independent sample t-test was conducted to compare the acoustic feature parameters between the two groups.Features with statistically significant differences were fur-ther analyzed using the receiver operating characteristic(ROC)curve to identify the optimal acoustic features with the high-est discriminative power for detecting vocal fold nodule disorders.Results The MFCC2,MFCC3,MFCC5,MFCC11,MFCC12,DQP,EP1,and EP2 feature values were significantly higher in the vocal fold nodule group compared to the nor-mal children group(P＜0.05,P＜0.001),while MFCC1,MFCC6,MFCC8,MFCC13,and EEP were significantly lower(P＜0.05).ROC curve analysis of these features showed that the combined ROC curve area under the curve(AUC)for MFCC1,MFCC2,MFCC3,MFCC5,MFCC6,MFCC8,MFCC11,MFCC12,MFCC13,DQP,EP1,EP2,and EEP was 0.98.The individual AUCs for MFCC1,MFCC2,MFCC3,MFCC5,MFCC6,MFCC8,MFCC11,MFCC12,DQP,and EP2 were all greater than 0.7,indicating a certain level of accuracy.Among these,MFCC2 and MFCC3 had AUCs of 0.85 and 0.87,respectively,indicating that these features have high diagnostic value for vocal fold nodule-related voice seg-ments in children.Conclusion The specific combination of acoustic parameters derived from multiband cepstral features of sustained vowels,including Mel-frequency cepstral coefficients(MFCC1,MFCC2,MFCC3,MFCC5,MFCC6,MFCC8,MFCC11,MFCC12,MFCC13)and cepstral peak values(DQP,EP1,EP2,EEP),can enhance the accuracy of detecting vocal fold nodule-related voice disorders in children.Particularly,MFCC2 and MFCC3 demonstrate high sensitivity and specificity,enabling more accurate identification of vocal fold nodule-related voice disorders in children.
