Estimation of Hypernasality Scores from Cleft Lip and Palate Speech

Abstract

Hypernasality refers to the perception of excessive nasal resonances in vowels and voiced consonants. Existing speech processing based approaches concentrate only on the classification of speech into normal or hypernasal, which do not give the degree of hypernasality in terms of continuous values like nasometer. Motivated by the functionality of nasometer, in this work, a method is proposed for the evaluation of hypernasality. Speech signals representing two extremely opposite cases of nasality are used to develop the acoustic models, where oral sentences (rich in vowels, stops, and fricatives) of normal speakers and nasal sentences (rich in nasals and nasalized vowels) of moderate-severe hypernasal speakers represent the groups with minimum and maximum attainable degrees of nasality, respectively. The acoustic features derived from glottal activity regions are used to model the maximum and minimum nasality classes using Gaussian mixture model and deep neural network approaches. The posterior probabilities obtained for nasal sentence class are referred to as hypernasality scores. The scores show a significant correlation (p < 0.01) with respect to perceptual ratings of hypernasality, provided by expert speechlanguage pathologists. Further, hypernasality scores are used for the detection of hypernasality, and the results are compared with the nasometer based approach.

Robust Estimation of Hypernasality in Dysarthria with Acoustic Model Likelihood Features

“…This suggests that these features are a robust measure of hypernasality, relatively invariant to the disease-specific co-modulating variables that hinder the performance of the baselines on the same task. The nasalization features in the NAP, by virtue of being trained on a large corpus of healthy speech, and targeting a specific perceptual quality are simultaneously more robust to both the diseasespecific overfitting expected from NN methods such as [55] and speaker-to-speaker variances discussed in the design of the formant-based A1P0 and related features in [66], [68]. Articulatory precision features are robust in a similar way.…”

“…Mel-frequency cepstral coefficients (MFCCs) and other spectral transformations [37], [38], [36], [39], [40], [41], [42], [43], [44], glottal source related features (jitter and shimmer) [45], [46], difference between the low-pass and bandpass profile of the Teager Energy Operator (TEO) [47], [48], and non-linear features [49], [50] have all been proposed as model input features. Gaussian mixture models (GMM), support vector machines, and deep neural networks have been used in conjunction with these features for hypernasality evaluation from word and sentence level data [51], [52], [53], [54]. Recently, end-to-end neural networks taking MFCC frames as input and producing hypernasality assessments as output have also been proposed [55].…”

“…Gaussian mixture models (GMM), support vector machines, and deep neural networks have been used in conjunction with these features for hypernasality evaluation from word and sentence level data [51], [52], [53], [54]. Recently, end-to-end neural networks taking MFCC frames as input and producing hypernasality assessments as output have also been proposed [55].…”

“…The results of this model are labeled FF-Linear, FF-Additive, and FF-KNN in Table II. In addition, we implemented the neural network proposed in [55], consisting of three feed-forward layers with sigmoid activations. The inputs to the networks are 39-dimensional Mel Frequency Cepstral Coefficients (MFCC) computed with a 20 ms window length and no overlap.…”

Automatic speaker independent dysarthric speech intelligibility assessment system

“…A very similar approach of using i-vector for regression task has been adopted in both [43] and [44]. GMM and DNN based models trained using MFCC features were employed for the task of hypernasality estimation in [45]. Two acoustic models trained on a large corpus of healthy speech, one to measure the nasal resonance from voiced sounds and another to measure the articulatory imprecision from unvoiced sounds was used in [46] to estimate hypernasality in dysarthric subjects.…”