Automatic identification of hypernasality in normal and cleft lip and palate patients with acoustic analysis of speech

Abstract

Hypernasality is seen in cleft lip and palate patients who had undergone repair surgery as a consequence of velopharyngeal insufficiency. Hypernasality has been studied by evaluation of perturbation, noise measures, and cepstral analysis of speech. In this study, feature extraction and analysis were performed during running speech using six different sentences. Jitter, shimmer, Mel frequency cepstral coefficients, bionic wavelet transform entropy, and bionic wavelet transform energy were calculated. Support vector machines were employed for classification of data to normal or hypernasal. Finally, results of the automatic classification were compared with true labels to find accuracy, sensitivity, and specificity. Accuracy was higher when Mel frequency cepstral coefficients were combined with bionic wavelet transform energy feature. In the best case, accuracy of 85% with sensitivity of 82% and specificity of 85% was obtained. Results prove that acoustic analysis is a reliable method to find hypernasality in cleft lip and palate patients.

Current Applications of Artificial Intelligence in Cleft Care: A Scoping Review

“…The maximum use of AI was to study hypernasality. Three studies tried to detect its presence ( 40 , 41 , 43 ) and two classified it according to severity ( 44 , 48 ). They used extracted features of speech as inputs for the classifiers.…”

An Attention Model for Hypernasality Prediction in Children with Cleft Palate

“…Spectral features in voiced sounds such as first formant strength, the presence of anti-formants, and spectral flatness show some correlation with increased hypernasality [4,5,6]. More recently, hypernasality evaluation algorithms rely on machine learning; frame-level features such as MFCCs are extracted from the segmented regions to train classifiers like support vector machine and Gaussian mixture models [7,8,9]. In similar vein, convolutional neural networks and recurrent neural networks have also been used to detect hypernasality [10,11].…”

Objective Measures of Plosive Nasalization in Hypernasal Speech

“…While these methods have all demonstrated some effectiveness in measuring hypernasality, the complex spectral signature of nasalization is difficult to capture with a simple representation. There is also a body of work on measuring nasality using machine learning [17], [18], [19], but these methods are all trained on single-disorder data, so it is difficult to assess if they learn acoustics specific to hypernasality or other co-modulating variables.…”