Pathological Voices Research Articles

The Impact of a Vocal Loading Task on Voice Characteristics of Female Speakers With Benign Vocal Fold Lesions

To examine the effect of a vocal loading task on measures of vocal structure and function in females with benign vocal fold lesions (BVFLs) and determine if change is observed in voice and lesion characteristics. Prospective cohort study. Twenty-eight (n = 28) female subjects with phonotraumatic BVFLs completed a vocal loading task of 30 minutes of reading aloud at 75-85 dBA. Multidimensional voice evaluation was completed pre- and post-load, including audio and videostroboscopy recordings and images for expert perceptual ratings and acoustic and aerodynamic evaluation. Subjects also scored themselves using a 10 cm visual analogue scale for Perceived Phonatory Effort, and completed the Evaluation of Ability to Voice Easily, a 12 item self-report scale of current perceived speaking voice function. An exploratory rather than confirmatory approach to data analysis was adopted. The direction and magnitude of the change scores (pre- to post-load) for each individual, across a wide variety of instrumental and self-report measures, were assessed against a Minimal Clinically Important Difference criteria. Observations of change and the direction of change in vocal response of individuals with BVFLs to 30 minutes of loud vocal load was variable. Minimal to no change was noted for participants pre- to post-load as rated perceptually, for auditory and videostroboscopy samples. For most instrumental measures, change was shown for many participants including an overall improvement in aerodynamic and acoustic measures of function and efficiency post-load for 20 participants (77%) and decline in function for 4 participants (15%). Self-reported effort and vocal function post-load was multidirectional with similar numbers of participants reporting no change, improved function or a decline. Subjects with BVFLs demonstrate change in vocal function following 30 minutes of vocal load. While this change can be variable and multidirectional, overall improvement was observed in instrumental measures of function and efficiency for most participants. Some participants perceived this change to be an increase in effort, some a reduction in effort and some perceived no change. Improved vocal function despite relative lesion stability can seemingly occur after loading in some pathological voices.

Significant pathological voice discrimination by computing posterior distribution of balanced accuracy

The ability to speak lucidly plays a key role in social relations. Consequently, the role of the larynx is quite important, and timely diagnosis of laryngeal diseases has proved to be crucial. In this study, a simple computational model for inverse of speech production model is employed to extract the glottal waveform using speech signal. This waveform has useful information about vocal folds performance in terms of providing evidence for distinguishing pathological disorders. Furthermore, obtaining the significance of classification results is important, because it leads to reliable inferences. This study utilizes the sustained vowel sound /a/ and a well-referenced database, namely MEEI. In this work, after extraction of six discriminating features by using appropriate signal modeling and processing methods and upon change of the feature space by using Kernel Principal Component Analysis (KPCA), a classifier consisting of Naïve Bayes and Fisher Linear Discriminant, is exerted on the feature sets. Regarding voice pathology detection, the proposed approach achieved a significant classification balanced accuracy 93.6%±0.03 with p-value <0.01 for normal and abnormal classification using the Beta distribution model for the posterior distribution of the average of the cross-validation results. The proposed features are also compared with some conventional features in this field. The results show significant improved performance for the proposed features in discriminating different types of pathological voices.

Development of a machine-learning based voice disorder screening tool

ObjectiveEarly recognition and referral are crucial for voice disorder management. Limited availability of subspecialists, poor primary care awareness, and the need for specialized equipment impede effective care. Thus, there is a need for a tool to improve voice pathology screening. Machine learning algorithms (MLAs) have shown promise in analyzing acoustic characteristics of phonation. However, few studies report clinical applications of MLAs for voice pathology detection. The objective of this study was to design and validate a MLA for detecting pathological voices. MethodsA MLA was developed for voice analysis. Audio samples converted into spectrograms were inputted into a pre-existing VGG19 convolutional neural network (CNN) and image-classifier. The resulting feature map was classified as either pathological or healthy using a Support Vector Machine (SVM) binary linear classifier. This combined MLA was “trained” with 950 sustained “/i/” vowel audio samples from the Saarbrucken Voice Database (SVD), which contains subjects with and without voice disorders. The trained MLA was “tested” with 406 SVD samples to determine sensitivity, specificity, and overall accuracy. External validation of the MLA was performed using clinical voice samples collected from patients attending a subspecialty voice clinic. ResultsThe MLA detected pathologies in SVD samples with 98.5% sensitivity, 97.1% specificity and 97.8% overall accuracy. In 30 samples obtained prospectively from voice clinic patients, the MLA detected pathologies with 100% sensitivity, 96.3% specificity and 96.7% overall accuracy. ConclusionsThis study demonstrates that a MLA using a simple audio input can detect diverse vocal pathologies with high sensitivity and specificity. Thus, this algorithm shows promise as a potential screening tool.

Voice Pathology Detection Using the Adaptive Orthogonal Transform Method, SVM and MLP.

In this paper, an automatic voice pathology recognition system is realized. The special features are extracted by the Adaptive Orthogonal Transform method, and to provide their statistical properties we calculated the average, variance, skewness and kurtosis values. The classification process uses two models that are widely used as a classification method in the field of signal processing: Support Vector Machine (SVM) and Multilayer Perceptron (MLP). The proposed system is tested by using a German voice database: the Saarbruecken Voice Database (SVD). The experimental results show that the Adaptive Orthogonal Transform method works perfectly with the Multilayer Perceptron Neural Network, which achieved 98.87% accuracy. On the other hand, the combination of the Adaptive Orthogonal Transform method and Support Vector Machine reached 85.79% accuracy.

Synchronized and Desynchronized Dynamics Observed from Physical Models of the Vocal and Ventricular Folds

The ventricular folds, located superiorly to the vocal folds, do not usually vibrate during normal phonations. It has been shown, however, that they do vibrate together with the vocal folds under special circumstances such as voice pathology and singing voice. Towards understanding the effect of the ventricular fold oscillations on the vocal fold oscillations, the present study developed a synthetic model that takes into account anatomical features of the human ventricular folds. The synthetic model is made of flexible silicone compounds with material properties comparable to those of human ventricular fold tissues. In our experiment, an air-flow was injected into the vocal and ventricular fold models. As the distance between the left and right ventricular folds was reduced, the ventricular folds started to co-vibrate with the vocal folds. Depending upon the distance, various oscillation patterns of the vocal-ventricular folds were observed, e.g., synchronized dynamics with 1:1 or 1:2 frequency ratio and desynchronized chaotic dynamics. The observed chaotic dynamics might be related to voice pathology induced by the ventricular phonation. A computational model was further presented to elucidate the experimental findings.

Two-dimensional Scheme for Arranging Voice Disorders

Prevalent schemes that have been used for arranging voice pathologies have shaped theoretical and clinical views and the conceptualization of the pathologies and of the field as a whole. However, these available schemes contain inconsistencies and categorical overlaps. To develop and evaluate a new approach for arranging voice pathologies, using 2 continuous scales, organicity and tonicity, which were used to construct a 2-dimensional plane. This survey study was conducted among experts in the fields of laryngology and/or voice disorders from 10 countries. The survey was conducted using an online platform from March to May 2021. The data were analyzed in June 2021. Of the 45 experts who were initially approached, 39 (86.7%) completed the survey. The primary outcome measures were group ratings on 2 rating scales: organicity and tonicity. On the organicity scale, 0 represented nonorganic and 10 organic. On the tonicity scale, 0 represented hypotonic and 10 hypertonic. Participants included 16 laryngologists and 23 speech-language pathologists, of whom 27 (69.2%) were women and 12 (30.8%) men with a mean age of 55 years. The Cronbach α was high for organicity and tonicity (0.98 and 0.97, respectively). Interrater agreement (rwg) was moderate to very strong (rwg≥0.50) for most pathologies. The correlation between the 2 scales was moderate and negative (r = -0.38; P = .03). The pathologies were scattered across the full range of both scales and the 4 quadrants of the 2-dimensional plane, suggesting the continuity and bidimensionality of the new arrangement scheme. In addition, a latent profile analysis suggested that the 4-cluster solution is valid and roughly corresponded to the 4 quadrants of the constructed plane. The findings of this survey study suggest the potential use of a 2-dimensional plane that was based on 2 continuous scales as a new arrangement scheme for voice disorders. The results suggest that this approach provides a valid representation of the field based on 2 basic measures beyond the specific etiology of each laryngeal pathology or condition. This simple and comprehensive organization scheme has the potential to facilitate new insights on the nature of voice pathologies, considering the interpathology similarities and differences.

Extraction and Utilization of Excitation Information of Speech: A Review

Speech production can be regarded as a process where a time-varying vocal tract system (filter) is excited by a time-varying excitation. In addition to its linguistic message, the speech signal also carries information about, for example, the gender and age of the speaker. Moreover, the speech signal includes acoustical cues about several speaker traits, such as the emotional state and the state of health of the speaker. In order to understand the production of these acoustical cues by the human speech production mechanism and utilize this information in speech technology, it is necessary to extract features describing both the excitation and the filter of the human speech production mechanism. While the methods to estimate and parameterize the vocal tract system are well established, the excitation appears less studied. This article provides a review of signal processing approaches used for the extraction of excitation information from speech. This article highlights the importance of excitation information in the analysis and classification of phonation type and vocal emotions, in the analysis of nonverbal laughter sounds, and in studying pathological voices. Furthermore, recent developments of deep learning techniques in the context of extraction and utilization of the excitation information are discussed.

Voice pathology detection and classification from speech signals and EGG signals based on a multimodal fusion method.

Automatic voice pathology detection and classification plays an important role in the diagnosis and prevention of voice disorders. To accurately describe the pronunciation characteristics of patients with dysarthria and improve the effect of pathological voice detection, this study proposes a pathological voice detection method based on a multi-modal network structure. First, speech signals and electroglottography (EGG) signals are mapped from the time domain to the frequency domain spectrogram via a short-time Fourier transform (STFT). The Mel filter bank acts on the spectrogram to enhance the signal's harmonics and denoise. Second, a pre-trained convolutional neural network (CNN) is used as the backbone network to extract sound state features and vocal cord vibration features from the two signals. To obtain a better classification effect, the fused features are input into the long short-term memory (LSTM) network for voice feature selection and enhancement. The proposed system achieves 95.73% for accuracy with 96.10% F1-score and 96.73% recall using the Saarbrucken Voice Database (SVD); thus, enabling a new method for pathological speech detection.

Voice Pathologies Classification Using GMM And SVM Classifiers

In this paper we investigate the proprieties of automatic speaker recognition (ASR) to develop a system for voice pathologies detection, where the model does not correspond to a speaker but it corresponds to group of patients who shares the same diagnostic. One of essential part in this topic is the database (described later), the samples voices (healthy and pathological) are chosen from a German database which contains many diseases, spasmodic dysphonia is proposed for this study. This problematic can be solved by statistical pattern recognition techniques where we have proposed the mel frequency cepstral coefficients (MFCC) to be modeled first, with gaussian mixture model (GMM) massively used in ASR then, they are modeled with support vector machine (SVM). The obtained results are compared in order to evaluate the more preferment classifier. The performance of each method is evaluated in a term of the accuracy, sensitivity, specificity. The best performance is obtained with 12 coefficientsMFCC, energy and second derivate along SVM with a polynomial kernel function, the classification rate is 90% for normal class and 93% for pathological class.This work is developed under MATLAB

Convolutional neural network ensemble for Parkinson's disease detection from voice recordings

The computerized detection of Parkinson's disease (PD) will facilitate population screening and frequent monitoring and provide a more objective measure of symptoms, benefiting both patients and healthcare providers. Dysarthria is an early symptom of the disease and examining it for computerized diagnosis and monitoring has been proposed. Deep learning-based approaches have advantages for such applications because they do not require manual feature extraction, and while this approach has achieved excellent results in speech recognition, its utilization in the detection of pathological voices is limited. In this work, we present an ensemble of convolutional neural networks (CNNs) for the detection of PD from the voice recordings of 50 healthy people and 50 people with PD obtained from PC-GITA, a publicly available database. We propose a multiple-fine-tuning method to train the base CNN. This approach reduces the semantical gap between the source task that has been used for network pretraining and the target task by expanding the training process by including training on another dataset. Training and testing were performed for each vowel separately, and a 10-fold validation was performed to test the models. The performance was measured by using accuracy, sensitivity, specificity and area under the ROC curve (AUC). The results show that this approach was able to distinguish between the voices of people with PD and those of healthy people for all vowels. While there were small differences between the different vowels, the best performance was when/a/was considered; we achieved 99% accuracy, 86.2% sensitivity, 93.3% specificity and 89.6% AUC. This shows that the method has potential for use in clinical practice for the screening, diagnosis and monitoring of PD, with the advantage that vowel-based voice recordings can be performed online without requiring additional hardware.

Acoustic Analysis of Voice Signal of Patients with Unilateral Laryngeal Paralysis a View to Objective Evaluation after Rehabilitation

We propose in this study an acoustic analysis called objective, based on physical measurements, to extract the acoustic characteristics of the voice of patients with Unilateral Laryngeal Paralysis. Our experiments were made at the Otorhinolaryngology service of the hospital of Bab El Oued. Algiers. (Algeria). In this work, an acoustic analysis of the vocal signal is based on measurements of the instability of the amplitude and frequency of the vibrations of the vocal cords. The results obtained are compared with those of a group of reference of subjects who normally speak. The study of the results obtained by the acoustic analysis of the pathological voice during the rehabilitation phase shows a strong correlation of the acoustic parameters between the pathological voice and the reference one. The exclusive use of hearing to evaluate the effect of voice rehabilitation in the Algerian hospital environment remains insufficient. It is important to correlate the perceptual information with the interpreted acoustic measurements, in a manner to be able to develop a therapeutic project appropriate to the patient’s expectations and difficulties.

Quality of life and phonatory and morphological outcomes in cognitively unimpaired adolescents with Pierre Robin sequence: a cross-sectional study of 72 patients

BackgroundPierre Robin sequence (PRS) is a heterogeneous condition involving retro(micro)gnathia, glossoptosis and upper airway obstruction, very often with posterior cleft palate. Patients with PRS, either isolated or associated with Stickler syndrome have good intellectual prognosis. Nevertheless, the quality of life in adolescence and the phonatory and morphological outcomes are rarely analysed. We assessed the phonatory and morphological outcomes of 72 cognitively unimpaired adolescents with PRS, studied their oral (COHIP-SF19), vocal (VHI-9i) and generic quality of life (QoL; KIDSCREEN-52), and searched for determinants of these outcomes.ResultsTwo-thirds of our adolescents retained low or moderate phonation difficulties, but risk factors were not identified. For 14%, morphological results were considered disharmonious, with no link to neonatal retrognathia severity. Only one vs two-stage surgery seemed to affect final aesthetic results. The oral QoL of these adolescents was comparable to that of control patients and was significantly better than that of children with other craniofacial malformations (COHIP-SF19 = 17.5, 15.4 and 25.7, respectively). The oral QoL of the adolescents with non-isolated PRS was significantly worse (COHIP-SF19 = 24.2) than that of control patients and close to that of children with other craniofacial malformations. The vocal QoL of the adolescents (mean [SD] VHI-9i = 7.5 [5.4]) was better than that of patients with other voice pathologies and better when phonation was good. The generic QoL of the adolescents was satisfactory but slightly lower than that of controls, especially in dimensions concerning physical well-being, relationships and autonomy. QoL results were lower for adolescents with non-isolated than isolated PRS. Only non-isolated PRS and low oral QoL affected generic QoL.ConclusionMorphological or phonatory impairments remain non-rare in adolescents with PRS but do not seem to be directly responsible for altered QoL. These adolescents, especially those with non-isolated PRS, show self-confidence and social-relation fragility. We must focus on long-term functional and psychological results for PRS patients and improve therapy protocols and follow-up, notably those affecting the oral aspects of the disease.

Artificial Intelligence Techniques for the Non-invasive Detection of COVID-19 Through the Analysis of Voice Signals.

Healthcare sensors represent a valid and non-invasive instrument to capture and analyse physiological data. Several vital signals, such as voice signals, can be acquired anytime and anywhere, achieved with the least possible discomfort to the patient thanks to the development of increasingly advanced devices. The integration of sensors with artificial intelligence techniques contributes to the realization of faster and easier solutions aimed at improving early diagnosis, personalized treatment, remote patient monitoring and better decision making, all tasks vital in a critical situation such as that of the COVID-19 pandemic. This paper presents a study about the possibility to support the early and non-invasive detection of COVID-19 through the analysis of voice signals by means of the main machine learning algorithms. If demonstrated, this detection capacity could be embedded in a powerful mobile screening application. To perform this important study, the Coswara dataset is considered. The aim of this investigation is not only to evaluate which machine learning technique best distinguishes a healthy voice from a pathological one, but also to identify which vowel sound is most seriously affected by COVID-19 and is, therefore, most reliable in detecting the pathology. The results show that Random Forest is the technique that classifies most accurately healthy and pathological voices. Moreover, the evaluation of the vowel /e/ allows the detection of the effects of COVID-19 on voice quality with a better accuracy than the other vowels.

Office Based Voice Evaluation in Assessing Outcome of Treatment in Laryngeal Disorders.

Rapidly evolving global economic status has changed professional behavior. Of late, there is a drastic increase in professional voice users and thereby increase in the patients with voice disorders. Clinically, the changes in the voice can only be perceived but cannot be quantified and documented. It is therefore necessary to have a gadget/tool which is simple and with which we can document and quantify the voice change objectively. This is a prospective case control study with 50 normal control subjects (25 male and 25 female subjects) who had no voice complaints and 50 patients(cases) who presented with complaints of change in voice due to various voice pathologies, the voices were recorded from the control group and also from the cases group both pre and posttreatment and were analyzed. The disease specific frequency in various pathologies of voice pre and post treatment were documented. The degree of change in voice frequency pre and post treatment were compared with each other and with the average frequency obtained from the voices in the control group. Office based acoustic analysis can definitely be used as a simple tool for documenting voice changes on outpatient basis and it provides a dependable documented evidence with which we can even compare the results following treatment.

Acoustic spaces for normal and pathological voices

What does it mean for a voice to sound normal? Our recent work comparing listener judgements of talkers with and without a diagnosed voice disorder showed that judgments of “normal” versus “not normal” and strategies for estimating how much a given voice deviates from normal depended on the listener, the context, the purpose of the judgement, and other factors as well as on the voice. F0, F1 and F2, and F0 variability were significant predictors of both categorical judgments and of scalar normalness ratings, but between-listener inconsistency in the perceptual models was observed. The present study examined acoustic spaces for the same normal and pathological voices by performing principal component analysis on scaled values of F0, formant frequencies, spectral noise, source spectral shape, and their variability. Acoustic spaces for normal and disordered talkers had similar structures, and included measures related to source spectral shape, F1, and higher formant frequencies. Overall, variability in noise was only important for pathological voices. The ways in which acoustic voice spaces are related to listener perceptions of a voice are discussed. [Work supported by NIH/NSF.]

Classification of Parkinson disease based on analysis and synthesis of voice signal

The most important application of voice profiling is pathological voice detection. Parkinson's disease is a chronic neurological degenerative disease affecting the central nervous system responsible for essentially progressive evolution movement disorders. 70% to 90% of Parkinson’s disease (PD) patients show an affected voice. This paper proposes a methodology for PD based on acoustic, glottal, physical, and electrical parameters. The results show that the acoustic parameter is more important in the case of Parkinson’s disease as compared to glottal and physical parameters. The authors achieved 97.2% accuracy to differentiate Parkinson and healthy voice using jitter to pitch ratio proposed algorithm. The Authors also proposed an algorithm of poles calculation of the vocal tract to find formants of the vocal tract. Further, formants are used for finding the transfer function of vocal tract filter. In the end, the authors suggested parameters of the electrical vocal tract model are also changed in the case of PD voices.

Classification of Parkinson disease based on analysis and synthesis of voice signal

The most important application of voice profiling is pathological voice detection. Parkinson's disease is a chronic neurological degenerative disease affecting the central nervous system responsible for essentially progressive evolution movement disorders. 70% to 90% of Parkinson’s disease (PD) patients show an affected voice. This paper proposes a methodology for PD based on acoustic, glottal, physical, and electrical parameters. The results show that the acoustic parameter is more important in the case of Parkinson’s disease as compared to glottal and physical parameters. The authors achieved 97.2% accuracy to differentiate Parkinson and healthy voice using jitter to pitch ratio proposed algorithm. The Authors also proposed an algorithm of poles calculation of the vocal tract to find formants of the vocal tract. Further, formants are used for finding the transfer function of vocal tract filter. In the end, the authors suggested parameters of the electrical vocal tract model are also changed in the case of PD voices.

Deep learning approach for voice pathology detection and classification

A non-invasive cum robust voice pathology detection and classification architecture is proposed in the current manuscript. In place of the conventional feature-based machine learning techniques, a new architecture is proposed herein which initially performs deep learning-based filtering of the input voice signal, followed by a decision-level fusion of deep learning and a non-parametric learner. The efficacy of the proposed technique is verified by performing a comparative study with very recent work on the same dataset but based on different training algorithms.The proposed architecture has five different stages.The results are recorded in terms of nine (9) different classification score indices which are – mean average Precision, sensitivity, specificity, F1 score, accuracy, error, false-positive rate, Matthews Correlation Coefficient, and the Cohen’s Kappa index. The experimental results have shown that the use of machine learning classifier can get at most 96.12% accuracy, while the proposed technique achieved the highest accuracy of 99.14% in comparison to other techniques.

Deep learning approach for voice pathology detection and classification

A non-invasive cum robust voice pathology detection and classification architecture is proposed in the current manuscript. In place of the conventional feature-based machine learning techniques, a new architecture is proposed herein which initially performs deep learning-based filtering of the input voice signal, followed by a decision-level fusion of deep learning and a non-parametric learner. The efficacy of the proposed technique is verified by performing a comparative study with very recent work on the same dataset but based on different training algorithms.The proposed architecture has five different stages.The results are recorded in terms of nine (9) different classification score indices which are – mean average Precision, sensitivity, specificity, F1 score, accuracy, error, false-positive rate, Matthews Correlation Coefficient, and the Cohen’s Kappa index. The experimental results have shown that the use of machine learning classifier can get at most 96.12% accuracy, while the proposed technique achieved the highest accuracy of 99.14% in comparison to other techniques.

Gammatone spectral latitude features extraction for pathological voice detection and classification

To improve the performance of pathological voice detection and classification, gammatone spectral latitude (GTSL) features were proposed. GTSL features are inspired by the nonlinear phenomena produced from the human phonation, presenting explicit physiological meaning. The features combine with human auditory perception characteristics. GTSL features quantify the turbulent noise by the nonlinear compression of peak value and dynamic range of the spectrums in each frequency channel. For pathological voice detection, gammatone spectral latitude (GTSL) features fitted better with traditional machine learning algorithms than traditional nonlinear features and gammatone ceptral coefficients (GTCCs). In the classification between healthy, neuromuscular and structural voices, the proposed features achieved average accuracy of 99.6% in the Massachusetts Eye and Ear Infirmary (MEEI) database, which is 35.6% higher than other gammatone features. The accuracies in other database, Saarbruecken Voice Database (SVD) and Hospital Universitario Prłncipe de Asturias (HUPA), were 89.9% and 97.4% respectively. The experimental results indicate that, GTSL features can provide objective evaluation of voice diseases with low computational complexity and database dependency.