Pathological Voices Research Articles

Grouping Intrinsic Mode Functions and Residue for Pathological Classifications via Electroglottograms

The electroglottogram (EGG) is a signal used for measuring the change of the relative contact area in the vocal cord during the throat production. In the recent years, the low cost and the non-invasive applications have been derived. Hence, the EGG has been applied in various science, engineering and medical fields such as in the basic voice science including the phonetics, the singing and the hearing as well as in the speech and the language therapy and the related clinical works including the voice production physiology, the swallowing and the psychology. However, the pathological classifications using the EGGs usually yield the poor performances. This is because the EGGs are required to decompose into the various components for extracting the features for performing the classifications. Nevertheless, the total numbers of the components decomposed by some time frequrncy representation such as the empirical mode decomposition (EMD) for different EGGs are different. Hence, the dimension of the feature vectors extracted from different EGGs is different. This introduces to the difficulty for building a machine learning model for performing the classification. This paper is to address this issue. This paper proposes a method for grouping the intrinsic mode functions (IMFs) and the residue obtained by applying the EMD to the EGGs for classifying between the healthy subjects and the pathological subjects. More precisely, this paper proposes a clustering based method to group the IMFs and the residue so that the total numbers of the grouped IMFs of different EGGs are the same. First, the IMFs and the residue of the EGGs are catergorized into a desired number of groups based on their correlation coefficients. Second, the IMFs or the residue in each group are summed together to obtain the grouped IMF. Third, the mean frequency and the first formant of each grouped IMF are computed. Finally, a random forest is employed for performing the classification. To our best knowledge, this joint EMD and clustering based method is firstly proposed to preform the pathological voice detection. The computer numerical simulations are conducted using the online availiable Saarbrücken voice database. Here, five cross validations have been performed. The mean accuracy, the mean specificity and the mean sensitivity among these five validations are 86.98, 79.92 and 91.57, respectively. The standard deviation of the accuracy, the specificity and the sensitivity among these five validations are ±2.00%, ±3.71% and ±2.13%, respectively. The simulation results show that our proposed method outperforms the common EGG or speech processing based methods. This paper proposes a clustering based method for grouping the IMFs and the residue for performing the pathological classifications via the EGGs. The grouping criterion is based on the correlation coefficients. It is found that our proposed method can achieve the highest classifications for the majority signal to noise ratios compared to the state of the arts methods. • This paper proposes a clustering based method for grouping the IMFs and the residue. • It applies to classify between the healthy and the pathological subjects. • The total numbers of the grouped IMFs of different EGGs are the same. • The mean frequency and the first formant of each grouped IMF are computed. • A random forest is employed for performing the classification.

Quantitative acoustical analysis of genetic syndromes in the number listing task

ObjectiveFor several disorders and diseases, quantitative acoustical analysis represents a non-invasive approach for assessing voice pathologies. Recently, methods were proposed to characterize the speech phenotype of some genetic syndromes, but the analysis was often limited to sustained vowels. Thus, more evidence is needed to confirm the usefulness of acoustical analysis in such cases. In this work, we evaluated if the acoustical analysis of a number listing task (numbers in ascending order from 1 to 10) may add helpful information for characterizing the speech phenotype related to the Italian language for three genetic syndromes: Down, Noonan, and Smith-Magenis. MethodsFor each syndrome, 24 subjects were analysed, along with 15 control subjects. Several acoustical features and entropy indexes were estimated using the BioVoice software and MATLAB routines. Differences between pathological and control cases and possible inter-syndromes differences were evaluated. ResultsStatistical results suggest that overall differences might exist between controls and subjects with genetic syndromes. Moreover, entropy indexes gave significant results also when the acoustical features did not detect differences. ConclusionPreliminary results confirmed the usefulness of both acoustical analysis and entropy techniques for the analyzed syndromes: for some syndromes, a specific speech phenotype exists that might support the clinician, highlighting the syndrome’s characteristics not yet exploited. SignificanceAn innovative approach is proposed to characterize the speech phenotype of some genetic syndromes based on both quantitative acoustical analysis and entropy measures of a speech task. The proposed non-invasive approach might be helpful for diagnosis, long-term monitoring, and follow-up of such syndromes.

MMHFNet: Multi-modal and multi-layer hybrid fusion network for voice pathology detection

Automatic voice pathology detection using non-invasive techniques that utilize patients’ speech and electroglottograph (EGG) signals play a vital role in diagnosis and early medical intervention. In this paper, a novel deep Multi-Modal and Multi-Layer Hybrid Fusion Network (MMHFNet) is proposed to improve the performance of non-invasive voice pathology detection systems. MMHFNet simultaneously incorporates complementary information of different modalities (speech and EGG signals). It also vertically combines the low-level features, extracted from shallow layers, and high-level features, extracted from deep layers, to take the full advantage of spatio-spectral information of different layers for multi-layer fusion. The features extracted by MMHFNet are then fed into an LSTM classification network to diagnose the voice pathology. Comprehensive experiments are conducted on the publicly available Saarbruecken Voice Database (SVD) to evaluate the performance of the proposed MMHFNet. This dataset is used in two manners; one using its all samples and the other with selected samples to form the largest balanced SVD dataset. Experimental results demonstrated that the proposed MMHFNet achieves accuracy rates of 91% and 96.05% for datasets with all and balanced samples, respectively.

An Efficient SMOTE-Based Deep Learning Model for Voice Pathology Detection

The Saarbruecken Voice Database (SVD) is a public database used by voice pathology detection systems. However, the distributions of the pathological and normal voice samples show a clear class imbalance. This study aims to develop a system for the classification of pathological and normal voices that uses efficient deep learning models based on various oversampling methods, such as the adaptive synthetic sampling (ADASYN), synthetic minority oversampling technique (SMOTE), and Borderline-SMOTE directly applied to feature parameters. The suggested combinations of oversampled linear predictive coefficients (LPCs), mel-frequency cepstral coefficients (MFCCs), and deep learning methods can efficiently classify pathological and normal voices. The balanced datasets from ADASYN, SMOTE, and Borderline-SMOTE are used to validate and evaluate the various deep learning models. The experiments are conducted using model evaluation metrics such as the recall, specificity, G, and F1 value. The experimental results suggest that the proposed voice pathology detection (VPD) system integrating the LPCs oversampled by the SMOTE and a convolutional neural network (CNN) can effectively yield the highest accuracy at 98.89% when classifying pathological and normal voices. Finally, the performances of oversampling algorithms such as the ADASYN, SMOTE, and Borderline-SMOTE are discussed. Furthermore, the performance of SMOTE is superior to conventional imbalanced data oversampling algorithms, and it can be used to diagnose pathological signals in real-world applications.

Acoustic speech parameter relationships with voice disorders and phrase differences

While acoustic speech analysis is non-invasive, the utility has been mixed due to the range of voice types. For vocal health practitioners to efficiently and quickly assess and document voice changes, knowing which voice parameter would be sensitive to vocal change is crucial. Using a database of 296 individual voices including 8 voice pathology types and typical voice samples, the sensitivity of a range of acoustic speech parameters to differentiate common voice pathology types was investigated. Both traditional and contemporary acoustic speech metrics were estimated for the samples using a custom MATLAB script and PRAAT (e.g., jitter, shimmer HNR, CPPS, Alpha ratio, PPE). Analysis then evaluate the predictability value of the metrics to discriminate pathology type. From the pool of parameters, 11 were able to identify pathological voices from normal controls and several of the parameters were more sensitive to some pathology. For example, CPPs and jitter values could discriminate neuropathological voices whereas HNR and Shimmer cold discriminate muscle-based pathologies. These results indicate how the sensitivity of acoustic speech metrics to the voice pathology types can allow for the identification of individual metrics (or combinations of metrics) which could be used to track changes in vocal health.

Exploring speech characteristics for automatic pathological voice detection

This paper aims to explore speech characteristics for automatic pathological voice detection. First, the assumptions underlying the conducted experiments are presented, along with examples of databases containing pathological speech. Then, a selection process of features of the speech signal and algorithms used to distinguish between undisturbed and pathological speech are discussed. Two deep models are employed to perform binary classification of the speech signal. Their structure is presented along with the feature space chosen. The results of classifying undisturbed and pathological speech are compared with state-of-the-art literature sources. The accuracy values are comparable to those presented in the literature, i.e., they are within the range of 61%–71% for our research and 64%–98% for the literature data, depending on the algorithms and databases used. An application is also built to illustrate the speech recognition process. The experiments are summarized with conclusions, and the direction of possible future development of the research performed is given.

Determination of Harmonic Parameters in Pathological Voices—Efficient Algorithm

The harmonic parameters Autocorrelation, Harmonic to Noise Ratio (HNR), and Noise to Harmonic Ratio are related to vocal quality, providing alternative measures of the harmonic energy of a speech signal. They will be used as input resources for an intelligent medical decision support system for the diagnosis of speech pathology. An efficient algorithm is important when implementing it on low-power devices. This article presents an algorithm that determines these parameters by optimizing the window type and length. The method used comparatively analyzes the values of the algorithm, with different combinations of window and size and a reference value. Hamming, Hanning, and Blackman windows with lengths of 3, 6, 12, and 24 glottal cycles and various sampling frequencies were investigated. As a result, we present an efficient algorithm that determines the parameters using the Hanning window with a length of six glottal cycles. The mean difference of Autocorrelation is less than 0.004, and that of HNR is less than 0.42 dB. In conclusion, this algorithm allows extraction of the parameters close to the reference values. In Autocorrelation, there are no significant effects of sampling frequency. However, it should be used cautiously for HNR with lower sampling rates.

Voice Pathology Detection Using a Two-Level Classifier Based on Combined CNN–RNN Architecture

The construction of an automatic voice pathology detection system employing machine learning algorithms to study voice abnormalities is crucial for the early detection of voice pathologies and identifying the specific type of pathology from which patients suffer. This paper’s primary objective is to construct a deep learning model for accurate speech pathology identification. Manual audio feature extraction was employed as a foundation for the categorization process. Incorporating an additional piece of information, i.e., voice gender, via a two-level classifier model was the most critical aspect of this work. The first level determines whether the audio input is a male or female voice, and the second level determines whether the agent is pathological or healthy. Similar to the bulk of earlier efforts, the current study analyzed the audio signal by focusing solely on a single vowel, such as /a/, and ignoring phrases and other vowels. The analysis was performed on the Saarbruecken Voice Database,. The two-level cascaded model attained an accuracy and F1 score of 88.84% and 87.39%, respectively, which was superior to earlier attempts on the same dataset and provides a steppingstone towards a more precise early diagnosis of voice complications.

Combined Use of Nonlinear Measures for Analyzing Pathological Voices

Automatic voice pathology detection enables an objective assessment of pathologies that influence the voice production strategy. By utilizing the conventional pipeline model as well as the modern deep learning-centric end-to-end methodology, numerous pathological voice analyzing techniques have been developed. The conventional methodology is still a valid choice owing to the lack of enormous amounts of training data in the study region of pathological voice. In the meantime, obtaining higher precision, higher accuracy, and stability is still a complicated task. Therefore, by amalgamating the nonlinear measure, the pathological voices are analyzed to abate such risks. The viability of six nonlinear discriminating measures derived from the phase space realm, involving healthy and pathological voice signals, is studied in this work. The analyzed parameters are Singularity spectrum coefficients ([Formula: see text], [Formula: see text] and [Formula: see text]). Correlation entropy at optimum embedding dimension ([Formula: see text]) and correlation dimension at optimum embedding dimension ([Formula: see text]). Analyzing the pathological voices with better accuracy rates is the major objective of the proposed methodology. Here, the Support Vector Machine (SVM) was utilized as the classifier. Experimentations were performed on VOiceICarfEDerico (VOICED) databases subsuming 208 healthy, as well as pathological voices, amongst these 50 samples, were utilized. Here, the model obtained 97% of accuracy with 99% as of the classifier with Gaussian kernel function. Therefore, to differentiate normal as well as pathological subjects, the six proposed characteristics are highly beneficial; in addition, they will be supportive in pathology diagnosis.

Quest for Speech Enhancement Method in the Analysis of Pathological Voices

Quest for Speech Enhancement Method in the Analysis of Pathological Voices

A Preliminary Investigation of the Reliability of Acoustic Parameters of Voice through Smartphone Recordings in Individuals with Dysphonia.

Telepractice offers the opportunity to receive care at home without risk of exposure to healthcare acquired infections, especially during a pandemic. Hence, establishing the reliability of the diagnosis of dysphonia via a smartphone is fundamental to providing an alternative service delivery model. A total of 20 participants participated in the study. Recordings of sentence-based voice samples were done using a standardized microphone and the software used in labs and on smartphones. Comparisons were made of acoustic and perceptual voice in real-time and recorded samples speech in persons with typical vs pathological voice. Results revealed no significant differences perceptually between real-time voice and recorded voice in individuals with typical and pathological voices. In acoustic analysis, there was no significant difference in Fundamental frequency (F0) and Auditory Voice Quality Index (AVQI) between real-time voice and recorded voice in individuals with typical and pathological voice.

The accuracy of an Online Sequential Extreme Learning Machine in detecting voice pathology using the Malaysian Voice Pathology Database

BackgroundA multidimensional voice quality assessment is recommended for all patients with dysphonia, which requires a patient visit to the otolaryngology clinic. The aim of this study was to determine the accuracy of an online artificial intelligence classifier, the Online Sequential Extreme Learning Machine (OSELM), in detecting voice pathology. In this study, a Malaysian Voice Pathology Database (MVPD), which is the first Malaysian voice database, was created and tested.MethodsThe study included 382 participants (252 normal voices and 130 dysphonic voices) in the proposed database MVPD. Complete data were obtained for both groups, including voice samples, laryngostroboscopy videos, and acoustic analysis. The diagnoses of patients with dysphonia were obtained. Each voice sample was anonymized using a code that was specific to each individual and stored in the MVPD. These voice samples were used to train and test the proposed OSELM algorithm. The performance of OSELM was evaluated and compared with other classifiers in terms of the accuracy, sensitivity, and specificity of detecting and differentiating dysphonic voices.ResultsThe accuracy, sensitivity, and specificity of OSELM in detecting normal and dysphonic voices were 90%, 98%, and 73%, respectively. The classifier differentiated between structural and non-structural vocal fold pathology with accuracy, sensitivity, and specificity of 84%, 89%, and 88%, respectively, while it differentiated between malignant and benign lesions with an accuracy, sensitivity, and specificity of 92%, 100%, and 58%, respectively. Compared to other classifiers, OSELM showed superior accuracy and sensitivity in detecting dysphonic voices, differentiating structural versus non-structural vocal fold pathology, and between malignant and benign voice pathology.ConclusionThe OSELM algorithm exhibited the highest accuracy and sensitivity compared to other classifiers in detecting voice pathology, classifying between malignant and benign lesions, and differentiating between structural and non-structural vocal pathology. Hence, it is a promising artificial intelligence that supports an online application to be used as a screening tool to encourage people to seek medical consultation early for a definitive diagnosis of voice pathology.

Pathological voice classification based on multi-domain features and deep hierarchical extreme learning machine

The intelligent data-driven screening of pathological voice signals is a non-invasive and real-time tool for computer-aided diagnosis that has attracted increasing attention from researchers and clinicians. In this paper, the authors propose multi-domain features and the hierarchical extreme learning machine (H-ELM) for the automatic identification of voice disorders. A sufficient number of sensitive features are first extracted from the original voice signal through multi-domain feature extraction (i.e., features of the time domain and the sample entropy based on ensemble empirical mode decomposition and gammatone frequency cepstral coefficients). To eliminate redundancy in high-dimensional features, neighborhood component analysis is then applied to filter out sensitive features from the high-dimensional feature vectors to improve the efficiency of network training and reduce overfitting. The sensitive features thus obtained are then used to train the H-ELM for pathological voice classification. The results of the experiments showed that the sensitivity, specificity, F1 score, and accuracy of the H-ELM were 99.37%, 98.61%, 99.37%, and 98.99%, respectively. Therefore, the proposed method is feasible for the initial classification of pathological voice signals.

A Highly Accurate Dysphonia Detection System Using Linear Discriminant Analysis

The recognition of pathological voice is considered a difficult task for speech analysis. Moreover, otolaryngologists needed to rely on oral communication with patients to discover traces of voice pathologies like dysphonia that are caused by voice alteration of vocal folds and their accuracy is between 60%–70%. To enhance detection accuracy and reduce processing speed of dysphonia detection, a novel approach is proposed in this paper. We have leveraged Linear Discriminant Analysis (LDA) to train multiple Machine Learning (ML) models for dysphonia detection. Several ML models are utilized like Support Vector Machine (SVM), Logistic Regression, and K-nearest neighbor (K-NN) to predict the voice pathologies based on features like Mel-Frequency Cepstral Coefficients (MFCC), Fundamental Frequency (F0), Shimmer (%), Jitter (%), and Harmonic to Noise Ratio (HNR). The experiments were performed using Saarbrucken Voice Database (SVD) and a privately collected dataset. The K-fold cross-validation approach was incorporated to increase the robustness and stability of the ML models. According to the experimental results, our proposed approach has a 70% increase in processing speed over Principal Component Analysis (PCA) and performs remarkably well with a recognition accuracy of 95.24% on the SVD dataset surpassing the previous best accuracy of 82.37%. In the case of the private dataset, our proposed method achieved an accuracy rate of 93.37%. It can be an effective non-invasive method to detect dysphonia.

Automatic Voice Disorder Detection Using Self-Supervised Representations

Many speech features and models, including Deep Neural Networks (DNN), are used for classification tasks between healthy and pathological speech with the Saarbruecken Voice Database (SVD). However, accuracy values of 80.71% for phrases or 82.8% for vowels /aiu/ are the highest reported for audio samples in SVD when the evaluation includes the wide amount of pathologies in the database, instead of a selection of some pathologies. This paper targets this top performance in the state-of-the-art Automatic Voice Disorder Detection (AVDD) systems. In the framework of a DNN-based AVDD system we study the capability of Self-Supervised (SS) representation learning for describing discriminative cues between healthy and pathological speech. The system processes the SS temporal sequence of features with a single feed-forward layer and Class-Token (CT) Transformer for obtaining the classification between healthy and pathological speech. Furthermore, there is evaluated a suitable data extension of the training set with out-of-domain data is also evaluated to deal with the low availability of data for using DNN-based models in voice pathology detection. Experimental results using audio samples corresponding to phrases in the SVD dataset, including all pathologies available, show classification accuracy values until 93.36%. This means that the proposed AVDD system achieved accuracy improvements of 4.1% without the training data extension, and 15.62% after the training data extension compared to the baseline system. Beyond the novelty of using SS representations for AVDD, the fact of obtaining accuracies over 90% in these conditions and using the whole set of pathologies in the SVD is a milestone for voice disorder-related research. Furthermore, the study on the amount of in-domain data in the training set related to the system performance show guidance for the data preparation stage. Lessons learned in this work suggest guidelines for taking advantage of DNN, to boost the performance in developing automatic systems for diagnosis, treatment, and monitoring of voice pathologies.

Enhancing the Performance of Pathological Voice Quality Assessment System Through the Attention-Mechanism Based Neural Network

Doctors, nowadays, primarily use auditory-perceptual evaluation, such as the grade, roughness, breathiness, asthenia, and strain scale, to evaluate voice quality and determine the treatment. However, the results predicted by individual physicians often differ, because of subjective perceptions, and diagnosis time interval, if the patient's symptoms are hard to judge. Therefore, an accurate computerized pathological voice quality assessment system will improve the quality of assessment. This study proposes a self_attention-based system, with a deep learning technology, named self_attention-based bidirectional long-short term memory (SA BiLSTM). Different pitches [low, normal, high], and vowels [/a/, /i/, /u/], were added into the proposed model, to make it learn how professional doctors evaluate the grade, roughness, breathiness, asthenia, and strain scale, in a high dimension view. The experimental results showed that the proposed system provided higher performance than the baseline system. More specifically, the macro average of the F1 score, presented as decimal, was used to compare the accuracy of classification. The (G, R, and B) of the proposed system were (0.768±0.011, 0.820±0.009, and 0.815±0.009), which is higher than the baseline systems: deep neural network (0.395±0.010, 0.312±0.019, 0.321±0.014) and convolution neural network (0.421±0.052, 0.306±0.043, 0.3250±0.032) respectively. The proposed system, with SA BiLSTM, pitches, and vowels, provides a more accurate way to evaluate the voice. This will be helpful for clinical voice evaluations and will improve patients' benefits from voice therapy.

A Novel Voice Feature AVA and its Application to the Pathological Voice Detection Through Machine Learning

A Novel Voice Feature AVA and its Application to the Pathological Voice Detection Through Machine Learning

Hierarchical Multi-Class Classification of Voice Disorders Using Self-Supervised Models and Glottal Features

Previous studies on the automatic classification of voice disorders have mostly investigated the binary classification task, which aims to distinguish pathological voice from healthy voice. Using multi-class classifiers, however, more fine-grained identification of voice disorders can be achieved, which is more helpful for clinical practitioners. Unfortunately, there is little publicly available training data for many voice disorders, which lowers the classification performance on data from unseen speakers. Earlier studies have shown that the usage of glottal source features can reduce data redundancy in detection of laryngeal voice disorders. Another approach to tackle the problems caused by scarcity of training data is to utilize deep learning models, such as wav2vec 2.0 and HuBERT, that have been pre-trained on larger databases. Since the aforementioned approaches have not been thoroughly studied in the multi-class classification of voice disorders, they will be jointly studied in the present work. In addition, we study a hierarchical classifier, which enables task-wise feature optimization and more efficient utilization of data. In this work, the aforementioned three approaches are compared with traditional mel frequency cepstral coefficient (MFCC) features and one-vs-rest and one-vs-one SVM classifiers. The results in a 3-class classification problem between healthy voice and two laryngeal disorders (hyperfunctional dysphonia and vocal fold paresis) indicate that all the studied methods outperform the baselines. The best performance was achieved by using features from wav2vec 2.0 LARGE together with hierarchical classification. The balanced classification accuracy of the system was 62.77% for male speakers, and 55.36% for female speakers, which outperformed the baseline systems by an absolute improvement of 15.76% and 6.95% for male and female speakers, respectively.

A mini-review of pathological voice recognition

The aim of this review is to examine various pathological conditions that impact the voice and how these features can be used in their diagnosis. An electronic search of PubMed and Google Scholar was performed for the articles published between January 2000 and July 2022 using the keywords found in the Medical Subject Headings database along with PubMed regarding diseases affecting voice. Our preliminary search result identified 608 articles using the keywords mentioned below. Among those, 12 articles met the inclusion criteria set for this review. Voice analysis can prove to be the missing link in the study and early detection of diseases. Using multiple voice attributes to cross-reference and diagnose conditions has excellent potential to fasten the process and significantly improve the diagnosis and treatment of various diseases.

A Modular Deep Learning Architecture for Voice Pathology Classification

A Modular Deep Learning Architecture for Voice Pathology Classification