A machine learning approach for vocal fold segmentation and disorder classification based on ensemble method

In the healthcare domain, the essential task is to understand and classify diseases affecting the vocal folds (VFs). The accurate identification of VF disease is the key issue in this domain. Integrating VF segmentation and disease classification into a single system is challenging but important for precise diagnostics. Our study addresses this challenge by combining VF illness categorization and VF segmentation into a single integrated system. We utilized two effective ensemble machine learning methods: ensemble EfficientNetV2L-LGBM and ensemble UNet-BiGRU. We utilized the EfficientNetV2L-LGBM model for classification, achieving a training accuracy of 98.88%, validation accuracy of 97.73%, and test accuracy of 97.88%. These exceptional outcomes highlight the system's ability to classify different VF illnesses precisely. In addition, we utilized the UNet-BiGRU model for segmentation, which attained a training accuracy of 92.55%, a validation accuracy of 89.87%, and a significant test accuracy of 91.47%. In the segmentation task, we examined some methods to improve our ability to divide data into segments, resulting in a testing accuracy score of 91.99% and an Intersection over Union (IOU) of 87.46%. These measures demonstrate skill of the model in accurately defining and separating VF. Our system's classification and segmentation results confirm its capacity to effectively identify and segment VF disorders, representing a significant advancement in enhancing diagnostic accuracy and healthcare in this specialized field. This study emphasizes the potential of machine learning to transform the medical field's capacity to categorize VF and segment VF, providing clinicians with a vital instrument to mitigate the profound impact of the condition. Implementing this innovative approach is expected to enhance medical procedures and provide a sense of optimism to those globally affected by VF disease.

Classification of laryngeal diseases including laryngeal cancer, benign mucosal disease, and vocal cord paralysis by artificial intelligence using voice analysis

Voice change is often the first sign of laryngeal cancer, leading to diagnosis through hospital laryngoscopy. Screening for laryngeal cancer solely based on voice could enhance early detection. However, identifying voice indicators specific to laryngeal cancer is challenging, especially when differentiating it from other laryngeal ailments. This study presents an artificial intelligence model designed to distinguish between healthy voices, laryngeal cancer voices, and those of the other laryngeal conditions. We gathered voice samples of individuals with laryngeal cancer, vocal cord paralysis, benign mucosal diseases, and healthy participants. Comprehensive testing was conducted to determine the best mel-frequency cepstral coefficient conversion and machine learning techniques, with results analyzed in-depth. In our tests, laryngeal diseases distinguishing from healthy voices achieved an accuracy of 0.85-0.97. However, when multiclass classification, accuracy ranged from 0.75 to 0.83. These findings highlight the challenges of artificial intelligence-driven voice-based diagnosis due to overlaps with benign conditions but also underscore its potential.

Toward Real-World Voice Disorder Classification

OBJECTIVE

Voice disorders significantly compromise individuals' ability to speak in their daily lives. Without early diagnosis and treatment, these disorders may deteriorate drastically. Thus, automatic classification systems at home are desirable for people who are inaccessible to clinical disease assessments. However, the performance of such systems may be weakened due to the constrained resources and domain mismatch between the clinical data and noisy real-world data.

METHODS

This study develops a compact and domain-robust voice disorder classification system to identify the utterances of health, neoplasm, and benign structural diseases. Our proposed system utilizes a feature extractor model composed of factorized convolutional neural networks and subsequently deploys domain adversarial training to reconcile the domain mismatch by extracting domain-invariant features.

RESULTS

The results show that the unweighted average recall in the noisy real-world domain improved by 13% and remained at 80% in the clinic domain with only slight degradation. The domain mismatch was effectively eliminated. Moreover, the proposed system reduced the usage of both memory and computation by over 73.9%.

CONCLUSION

By deploying factorized convolutional neural networks and domain adversarial training, domain-invariant features can be derived for voice disorder classification with limited resources. The promising results confirm that the proposed system can significantly reduce resource consumption and improve classification accuracy by considering the domain mismatch.

SIGNIFICANCE

To the best of our knowledge, this is the first study that jointly considers real-world model compression and noise-robustness issues in voice disorder classification. The proposed system is intended for application to embedded systems with limited resources.

Deep learning-based speech analysis for Alzheimer's disease detection: a literature review

BACKGROUND

Alzheimer's disease has become one of the most common neurodegenerative diseases worldwide, which seriously affects the health of the elderly. Early detection and intervention are the most effective prevention methods currently. Compared with traditional detection methods such as traditional scale tests, electroencephalograms, and magnetic resonance imaging, speech analysis is more convenient for automatic large-scale Alzheimer's disease detection and has attracted extensive attention from researchers. In particular, deep learning-based speech analysis and language processing techniques for Alzheimer's disease detection have been studied and achieved impressive results.

METHODS

To integrate the latest research progresses, hundreds of relevant papers from ACM, DBLP, IEEE, PubMed, Scopus, Web of Science electronic databases, and other sources were retrieved. We used these keywords for paper search: (Alzheimer OR dementia OR cognitive impairment) AND (speech OR voice OR audio) AND (deep learning OR neural network).

CONCLUSIONS

Fifty-two papers were finally retained after screening. We reviewed and presented the speech databases, deep learning methods, and model performances of these studies. In the end, we pointed out the mainstreams and limitations in the current studies and provided a direction for future research.

Different Performances of Machine Learning Models to Classify Dysphonic and Non-Dysphonic Voices

OBJECTIVE

To analyze the performance of 10 different machine learning (ML) classifiers for discrimination between dysphonic and non-dysphonic voices, using a variance threshold as a method for the selection and reduction of acoustic measurements used in the classifier.

METHOD

We analyzed 435 samples of individuals (337 female and 98 male), with a mean age of 41.07 ± 13.73 years, of which 384 were dysphonic and 51 were non-dysphonic. From the sustained /ε/ vowel sample, 34 acoustic measurements were extracted, including traditional perturbation and noise measurements, cepstral/spectral measurements, and measurements based on nonlinear models. The variance method was used to select the best set of acoustic measurements. We tested the performance of the best-selected set with 10 ML classifiers using precision, sensitivity, specificity, accuracy, and F1-Score measurements. The kappa coefficient was used to verify the reproducibility between the two datasets (training and testing).

RESULTS

The naive Bayes (NB) and stochastic gradient descent classifier (SGDC) models performed best in terms of accuracy, AUC, sensitivity, and specificity for a reduced dataset of 15 acoustic measures compared to the full dataset of 34 acoustic measures. SGDC and NB obtained the best performance results, with an accuracy of 0.91 and 0.76, respectively. These two classifiers presented moderate agreement, with a Kappa of 0.57 (SGDC) and 0.45 (NB).

CONCLUSION

Among the tested models, the NB and SGDC models performed better in discriminating between dysphonic and non-dysphonic voices from a set of 15 acoustic measures.

Pathological Voice Detection Based on Phase Reconstitution and Convolutional Neural Network

The nonlinear dynamic features can effectively describe the acoustic characteristics of normal and pathological voice. In this paper, the phase space reconstruction and convolution neural network are used to classify the normal and pathological voice. The phase space information of normal and pathological voice is reconstructed using delay time and embedding dimension, the one-dimensional signal is converted to a two-dimensional matrix, and the reconstructed trajectory graph sample of the signal is generated. The trajectory graph samples are used as the input of the VGG-like convolutional neural network, and the graphical features are extracted to achieve a classification of normal and pathological voice. In order to overcome the lack of clinical data, a data enhancement scheme is used. The experiment which classifies the normal and pathological voice is carried out on three pathological databases respectively, i.e. the Massachusetts eye and ear infirmary (MEEI) database, Saarbrücken voice database (SVD) database, and a clinical database collected by the authors. Five-fold cross validation is used and the average recognition rates on the three databases are 99.42%, 97.30% and 95.88% respectively. The average recognition rates are 96.04% and 92.27% for normal, vocal fold paralysis and vocal fold non-paralysis voice in MEEI database and SVD database. The experimental results show that the method has high classification recognition rate and good robustness, and has certain universal applicability for the recognition of the normal and pathological voice.

Development of a machine-learning based voice disorder screening tool

OBJECTIVE

Early 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.

METHODS

A 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.

RESULTS

The 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.

CONCLUSIONS

This 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.

Recognition of Emotional States of Children with down Syndrome by Facial Expression: Perceptual and Automatic Analysis of Dynamic Images

The study is devoted to the investigation of the recognition of the emotional state of children with Down syndrome (DS) by their facial expression. For this purpose, a series of perceptual experiments involving adults (n=75) and automatic analysis of the facial expressions of children (n=35, aged 5—16 years) were carried out using the FaceReader program. The ability of adults to recognize the emotional states of children: joy — neutral (calm state) — sadness — anger, by open faces and faces with masks over the eyes and mouth is shown. Better recognition of the state of joy and neutral state under the condition of an open face and a decrease in recognition accuracy in a mask in the eye area compared to the absence of a mask and a mask in the mouth area were found. Automatic recognition of the states of joy and neutral states is better than the states of sadness and anger, if the face is open and the mask in the mouth area of the child. The conditions for use the automatic recognition of facial expression in children with DS and for applying the method of perceptual analysis for identifying the specificity of the child emotional sphere development 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.

Ten Years of Research on Automatic Voice and Speech Analysis of People With Alzheimer's Disease and Mild Cognitive Impairment: A Systematic Review Article

Background: The field of voice and speech analysis has become increasingly popular over the last 10 years, and articles on its use in detecting neurodegenerative diseases have proliferated. Many studies have identified characteristic speech features that can be used to draw an accurate distinction between healthy aging among older people and those with mild cognitive impairment and Alzheimer's disease. Speech analysis has been singled out as a cost-effective and reliable method for detecting the presence of both conditions. In this research, a systematic review was conducted to determine these features and their diagnostic accuracy.

Methods: Peer-reviewed literature was located across multiple databases, involving studies that apply new procedures of automatic speech analysis to collect behavioral evidence of linguistic impairments along with their diagnostic accuracy on Alzheimer's disease and mild cognitive impairment. The risk of bias was assessed by using JBI and QUADAS-2 checklists.

Results: Thirty-five papers met the inclusion criteria; of these, 11 were descriptive studies that either identified voice features or explored their cognitive correlates, and the rest were diagnostic studies. Overall, the studies were of good quality and presented solid evidence of the usefulness of this technique. The distinctive acoustic and rhythmic features found are gathered. Most studies record a diagnostic accuracy over 88% for Alzheimer's and 80% for mild cognitive impairment.

Conclusion: Automatic speech analysis is a promising tool for diagnosing mild cognitive impairment and Alzheimer's disease. The reported features seem to be indicators of the cognitive changes in older people. The specific features and the cognitive changes involved could be the subject of further research.

Voice Disorder Identification by using Hilbert-Huang Transform (HHT) and K Nearest Neighbor (KNN)

OBJECTIVES

Clinical evaluation of dysphonic voices involves a multidimensional approach, including a variety of instrumental and noninstrumental measures. Acoustic analyses provide an objective, noninvasive and intelligent measures of voice quality. Based on sound recordings, this paper proposes a new classification method of voice disorders with HHT and KNN.

METHODS

In this research, 12 features of each sample is calculated by HHT. Based on the algorithm of Linear Prediction Coefficient (LPCC), a sample can be characterized by 9 features. After each sample is expressed by 21 features, the classifier is constructed based on KNN. In addition, classifier based on KNN was further compared with random forest and extra trees classifiers in relation to their classification performance of voice disorder.

RESULTS

The experiment results revel that classifier based on KNN showed better performance than other two classifiers with accuracy rate of 93.3%, precision of 93%, recall rate of 95%, F1-score of 94% and the area of receiver operating characteristic curve is 0.976.

CONCLUSIONS

The method put forward in this paper can be effectively used to classify voice disorders.

Vocal Acoustic Analysis

Vocal acoustic analysis is becoming a useful tool for the classification and recognition of laryngological pathologies. This technique enables a non-invasive and low-cost assessment of voice disorders, allowing a more efficient, fast, and objective diagnosis. In this work, ANN and SVM were experimented on to classify between dysphonic/control and vocal cord paralysis/control. A vector was made up of 4 jitter parameters, 4 shimmer parameters, and a harmonic to noise ratio (HNR), determined from 3 different vowels at 3 different tones, with a total of 81 features. Variable selection and dimension reduction techniques such as hierarchical clustering, multilinear regression analysis and principal component analysis (PCA) was applied. The classification between dysphonic and control was made with an accuracy of 100% for female and male groups with ANN and SVM. For the classification between vocal cords paralysis and control an accuracy of 78,9% was achieved for female group with SVM, and 81,8% for the male group with ANN.

A Survey on Machine Learning Approaches for Automatic Detection of Voice Disorders

The human voice production system is an intricate biological device capable of modulating pitch and loudness. Inherent internal and/or external factors often damage the vocal folds and result in some change of voice. The consequences are reflected in body functioning and emotional standing. Hence, it is paramount to identify voice changes at an early stage and provide the patient with an opportunity to overcome any ramification and enhance their quality of life. In this line of work, automatic detection of voice disorders using machine learning techniques plays a key role, as it is proven to help ease the process of understanding the voice disorder. In recent years, many researchers have investigated techniques for an automated system that helps clinicians with early diagnosis of voice disorders. In this paper, we present a survey of research work conducted on automatic detection of voice disorders and explore how it is able to identify the different types of voice disorders. We also analyze different databases, feature extraction techniques, and machine learning approaches used in these research works.

Detection of Pathological Voice Using Cepstrum Vectors: A Deep Learning Approach

OBJECTIVES

Computerized detection of voice disorders has attracted considerable academic and clinical interest in the hope of providing an effective screening method for voice diseases before endoscopic confirmation. This study proposes a deep-learning-based approach to detect pathological voice and examines its performance and utility compared with other automatic classification algorithms.

METHODS

This study retrospectively collected 60 normal voice samples and 402 pathological voice samples of 8 common clinical voice disorders in a voice clinic of a tertiary teaching hospital. We extracted Mel frequency cepstral coefficients from 3-second samples of a sustained vowel. The performances of three machine learning algorithms, namely, deep neural network (DNN), support vector machine, and Gaussian mixture model, were evaluated based on a fivefold cross-validation. Collective cases from the voice disorder database of MEEI (Massachusetts Eye and Ear Infirmary) were used to verify the performance of the classification mechanisms.

RESULTS

The experimental results demonstrated that DNN outperforms Gaussian mixture model and support vector machine. Its accuracy in detecting voice pathologies reached 94.26% and 90.52% in male and female subjects, based on three representative Mel frequency cepstral coefficient features. When applied to the MEEI database for validation, the DNN also achieved a higher accuracy (99.32%) than the other two classification algorithms.

CONCLUSIONS

By stacking several layers of neurons with optimized weights, the proposed DNN algorithm can fully utilize the acoustic features and efficiently differentiate between normal and pathological voice samples. Based on this pilot study, future research may proceed to explore more application of DNN from laboratory and clinical perspectives.

Modal and non-modal voice quality classification using acoustic and electroglottographic features

The goal of this study was to investigate the performance of different feature types for voice quality classification using multiple classifiers. The study compared the COVAREP feature set; which included glottal source features, frequency warped cepstrum and harmonic model features; against the mel-frequency cepstral coefficients (MFCCs) computed from the acoustic voice signal, acoustic-based glottal inverse filtered (GIF) waveform, and electroglottographic (EGG) waveform. Our hypothesis was that MFCCs can capture the perceived voice quality from either of these three voice signals. Experiments were carried out on recordings from 28 participants with normal vocal status who were prompted to sustain vowels with modal and non-modal voice qualities. Recordings were rated by an expert listener using the Consensus Auditory-Perceptual Evaluation of Voice (CAPE-V), and the ratings were transformed into a dichotomous label (presence or absence) for the prompted voice qualities of modal voice, breathiness, strain, and roughness. The classification was done using support vector machines, random forests, deep neural networks and Gaussian mixture model classifiers, which were built as speaker independent using a leave-one-speaker-out strategy. The best classification accuracy of 79.97% was achieved for the full COVAREP set. The harmonic model features were the best performing subset, with 78.47% accuracy, and the static+dynamic MFCCs scored at 74.52%. A closer analysis showed that MFCC and dynamic MFCC features were able to classify modal, breathy, and strained voice quality dimensions from the acoustic and GIF waveforms. Reduced classification performance was exhibited by the EGG waveform.

Development of the Arabic Voice Pathology Database and Its Evaluation by Using Speech Features and Machine Learning Algorithms

A voice disorder database is an essential element in doing research on automatic voice disorder detection and classification. Ethnicity affects the voice characteristics of a person, and so it is necessary to develop a database by collecting the voice samples of the targeted ethnic group. This will enhance the chances of arriving at a global solution for the accurate and reliable diagnosis of voice disorders by understanding the characteristics of a local group. Motivated by such idea, an Arabic voice pathology database (AVPD) is designed and developed in this study by recording three vowels, running speech, and isolated words. For each recorded samples, the perceptual severity is also provided which is a unique aspect of the AVPD. During the development of the AVPD, the shortcomings of different voice disorder databases were identified so that they could be avoided in the AVPD. In addition, the AVPD is evaluated by using six different types of speech features and four types of machine learning algorithms. The results of detection and classification of voice disorders obtained with the sustained vowel and the running speech are also compared with the results of an English-language disorder database, the Massachusetts Eye and Ear Infirmary (MEEI) database.

Hierarchical Classification and System Combination for Automatically Identifying Physiological and Neuromuscular Laryngeal Pathologies

OBJECTIVES

Speech signal processing techniques have provided several contributions to pathologic voice identification, in which healthy and unhealthy voice samples are evaluated. A less common approach is to identify laryngeal pathologies, for which the use of a noninvasive method for pathologic voice identification is an important step forward for preliminary diagnosis. In this study, a hierarchical classifier and a combination of systems are used to improve the accuracy of a three-class identification system (healthy, physiological larynx pathologies, and neuromuscular larynx pathologies).

METHOD

Three main subject classes were considered: subjects with physiological larynx pathologies (vocal fold nodules and edemas: 59 samples), subjects with neuromuscular larynx pathologies (unilateral vocal fold paralysis: 59 samples), and healthy subjects (36 samples). The variables used in this study were a speech task (sustained vowel /a/ or continuous reading speech), features with or without perceptual information, and features with or without direct information about formants evaluated using single classifiers. A hierarchical classification system was designed based on this information.

RESULTS

The resulting system combines an analysis of continuous speech by way of the commonly used sustained vowel /a/ to obtain spectral and perceptual speech features. It achieved an accuracy of 84.4%, which represents an improvement of approximately 9% compared with the stand-alone approach. For pathologic voice identification, the accuracy obtained was 98.7%, and the identification accuracy for the two pathology classes was 81.3%.

CONCLUSIONS

Hierarchical classification and system combination create significant benefits and introduce a modular approach to the classification of larynx pathologies.

An Investigation of Multidimensional Voice Program Parameters in Three Different Databases for Voice Pathology Detection and Classification

BACKGROUND AND OBJECTIVE

Automatic voice-pathology detection and classification systems may help clinicians to detect the existence of any voice pathologies and the type of pathology from which patients suffer in the early stages. The main aim of this paper is to investigate Multidimensional Voice Program (MDVP) parameters to automatically detect and classify the voice pathologies in multiple databases, and then to find out which parameters performed well in these two processes.

MATERIALS AND METHODS

Samples of the sustained vowel /a/ of normal and pathological voices were extracted from three different databases, which have three voice pathologies in common. The selected databases in this study represent three distinct languages: (1) the Arabic voice pathology database; (2) the Massachusetts Eye and Ear Infirmary database (English database); and (3) the Saarbruecken Voice Database (German database). A computerized speech lab program was used to extract MDVP parameters as features, and an acoustical analysis was performed. The Fisher discrimination ratio was applied to rank the parameters. A t test was performed to highlight any significant differences in the means of the normal and pathological samples.

RESULTS

The experimental results demonstrate a clear difference in the performance of the MDVP parameters using these databases. The highly ranked parameters also differed from one database to another. The best accuracies were obtained by using the three highest ranked MDVP parameters arranged according to the Fisher discrimination ratio: these accuracies were 99.68%, 88.21%, and 72.53% for the Saarbruecken Voice Database, the Massachusetts Eye and Ear Infirmary database, and the Arabic voice pathology database, respectively.

Investigation of Voice Pathology Detection and Classification on Different Frequency Regions Using Correlation Functions

OBJECTIVES AND BACKGROUND

Automatic voice pathology detection and classification systems effectively contribute to the assessment of voice disorders, which helps clinicians to detect the existence of any voice pathologies and the type of pathology from which patients suffer in the early stages. This work concentrates on developing an accurate and robust feature extraction for detecting and classifying voice pathologies by investigating different frequency bands using correlation functions. In this paper, we extracted maximum peak values and their corresponding lag values from each frame of a voiced signal by using correlation functions as features to detect and classify pathological samples. These features are investigated in different frequency bands to see the contribution of each band on the detection and classification processes.

MATERIAL AND METHODS

Various samples of sustained vowel /a/ of normal and pathological voices were extracted from three different databases: English, German, and Arabic. A support vector machine was used as a classifier. We also performed a t test to investigate the significant differences in mean of normal and pathological samples.

RESULTS

The best achieved accuracies in both detection and classification were varied depending on the band, the correlation function, and the database. The most contributive bands in both detection and classification were between 1000 and 8000 Hz. In detection, the highest acquired accuracies when using cross-correlation were 99.809%, 90.979%, and 91.168% in the Massachusetts Eye and Ear Infirmary, Saarbruecken Voice Database, and Arabic Voice Pathology Database databases, respectively. However, in classification, the highest acquired accuracies when using cross-correlation were 99.255%, 98.941%, and 95.188% in the three databases, respectively.

Voice Pathology Detection Using Modulation Spectrum-Optimized Metrics

There exist many acoustic parameters employed for pathological assessment tasks, which have served as tools for clinicians to distinguish between normophonic and pathological voices. However, many of these parameters require an appropriate tuning in order to maximize its efficiency. In this work, a group of new and already proposed modulation spectrum (MS) metrics are optimized considering different time and frequency ranges pursuing the maximization of efficiency for the detection of pathological voices. The optimization of the metrics is performed simultaneously in two different voice databases in order to identify what tuning ranges produce a better generalization. The experiments were cross-validated so as to ensure the validity of the results. A third database is used to test the optimized metrics. In spite of some differences, results indicate that the behavior of the metrics in the optimization process follows similar tendencies for the tuning databases, confirming the generalization capabilities of the proposed MS metrics. In addition, the tuning process reveals which bands of the modulation spectra have relevant information for each metric, which has a physical interpretation respecting the phonatory system. Efficiency values up to 90.6% are obtained in one tuning database, while in the other, the maximum efficiency reaches 71.1%. Obtained results also evidence a separability between normophonic and pathological states using the proposed metrics, which can be exploited for voice pathology detection or assessment.

Modulation Spectra Morphological Parameters: A New Method to Assess Voice Pathologies according to the GRBAS Scale

Disordered voices are frequently assessed by speech pathologists using perceptual evaluations. This might lead to problems caused by the subjective nature of the process and due to the influence of external factors which compromise the quality of the assessment. In order to increase the reliability of the evaluations, the design of automatic evaluation systems is desirable. With that in mind, this paper presents an automatic system which assesses the Grade and Roughness level of the speech according to the GRBAS perceptual scale. Two parameterization methods are used: one based on the classic Mel-Frequency Cepstral Coefficients, which has already been used successfully in previous works, and other derived from modulation spectra. For the latter, a new group of parameters has been proposed, named Modulation Spectra Morphological Parameters: MSC, DRB, LMR, MSH, MSW, CIL, PALA, and RALA. In methodology, PCA and LDA are employed to reduce the dimensionality of feature space, and GMM classifiers to evaluate the ability of the proposed features on distinguishing the different levels. Efficiencies of 81.6% and 84.7% are obtained for Grade and Roughness, respectively, using modulation spectra parameters, while MFCCs performed 80.5% and 77.7%. The obtained results suggest the usefulness of the proposed Modulation Spectra Morphological Parameters for automatic evaluation of Grade and Roughness in the speech.

Automatic Voice Pathology Detection With Running Speech by Using Estimation of Auditory Spectrum and Cepstral Coefficients Based on the All-Pole Model

BACKGROUND AND OBJECTIVE

Automatic voice pathology detection using sustained vowels has been widely explored. Because of the stationary nature of the speech waveform, pathology detection with a sustained vowel is a comparatively easier task than that using a running speech. Some disorder detection systems with running speech have also been developed, although most of them are based on a voice activity detection (VAD), that is, itself a challenging task. Pathology detection with running speech needs more investigation, and systems with good accuracy (ACC) are required. Furthermore, pathology classification systems with running speech have not received any attention from the research community. In this article, automatic pathology detection and classification systems are developed using text-dependent running speech without adding a VAD module.

METHOD

A set of three psychophysics conditions of hearing (critical band spectral estimation, equal loudness hearing curve, and the intensity loudness power law of hearing) is used to estimate the auditory spectrum. The auditory spectrum and all-pole models of the auditory spectrums are computed and analyzed and used in a Gaussian mixture model for an automatic decision.

RESULTS

In the experiments using the Massachusetts Eye & Ear Infirmary database, an ACC of 99.56% is obtained for pathology detection, and an ACC of 93.33% is obtained for the pathology classification system. The results of the proposed systems outperform the existing running-speech-based systems.

DISCUSSION

The developed system can effectively be used in voice pathology detection and classification systems, and the proposed features can visually differentiate between normal and pathological samples.