Vinícius Jefferson Dias Vieira

Accuracy of Acoustic Analysis Measurements in the Evaluation of Patients With Different Laryngeal Diagnoses

OBJECTIVE This study aims to investigate the accuracy of acoustic measures in discriminating between patients with different laryngeal diagnoses. STUDY DESIGN The study design is descriptive, cross-sectional, and retrospective. METHODS A total of 279 female patients participated in the research. Acoustic measures of the mean and standard deviation (SD) values of the fundamental frequency (F0), jitter, shimmer, and glottal to noise excitation (GNE) were extracted from the emission of the vowel /ε/. RESULTS Isolated acoustic measures do not demonstrate adequate performance in discriminating patients with and without laryngeal alteration. The combination of GNE, SD of the F0, jitter, and shimmer improved the ability to classify patients with and without laryngeal alteration. In isolation, the SD of the F0, shimmer, and GNE presented acceptable performance in discriminating individuals with different laryngeal diagnoses. The combination of acoustic measurements caused discrete improvement in performance of the classifier to discriminate healthy larynx vs vocal polyp (SD of the F0, shimmer, and GNE), healthy larynx vs unilateral vocal fold paralysis (SD of the F0 and jitter), healthy larynx vs vocal nodules (SD of the F0 and jitter), healthy larynx vs sulcus vocalis (SD of the F0 and shimmer), and healthy larynx vs voice disorder due to gastroesophageal reflux (F0 mean, jitter, and shimmer). CONCLUSIONS Isolated acoustic measures do not demonstrate adequate performance in discriminating patients with and without laryngeal alteration, although they present acceptable performance in classifying different laryngeal diagnoses. Combined acoustic measures present an acceptable capacity to discriminate between the presence and the absence of laryngeal alteration and to differentiate several laryngeal diagnoses.

Pathological voice assessment by recurrence quantification analysis

In this paper, the performance of quantification measures of recurrence plots is evaluated in the task of discriminating pathological voices from the healthy ones. In order to classify these signals as healthy, edema or paralysis, seven recurrence quantification measures are used: Determinism (DET), Maximum length of the diagonal structures (Lmax), Entropy (ENTR), Slope of line-of-best-fit (TREND), Laminarity (LAM), Length of longest vertical line segment (Vmax), and mean vertical line length or trapping time (TT). A discriminant analysis method is applied to each feature individually, using two discriminant functions: Linear and quadratic. The performance of individual classifiers are improved when combining the measures two by two. Results show that the method employed can be used as a highly reliable method for pathological voice assessment.

Análise da dinâmica do sistema de produção vocal na presença de patologias da laringe

Este trabalho trata da avaliacao do uso de medidas obtidas a partir da analise dinâmica nao linear na discriminacao entre vozes saudaveis e vozes afetadas por patologias laringeas (edemas de Reinke, nodulos e paralisia nas pregas vocais). As medidas empregadas, obtidas por meio da analise de quantificacao de recorrencia, sao: transitividade, comprimento medio das linhas diagonais e o tempo de recorrencia dos tipos 1 e 2. Os resultados obtidos no processo de classificacao indicam que as medidas avaliadas apresentam potencial discriminativo entre os dois grupos de sinais de voz considerados, podendo ser empregadas em sistemas de auxilio ao diagnostico de patologias na laringe.

Effectiveness of Recurrence Quantification Measures in Discriminating Subjects With and Without Voice Disorders

The objective of this study was to analyze the accuracy of recurrence quantification measurements (RQMs) in discriminating between individuals with and without voice disorders. This study consisted of a total of 541 recorded voice samples from normal and dysphonic subjects. All subjects recorded a sustained vowel /Ɛ/ and underwent a laryngoscopic examination of the larynx. Twelve RQMs and three parameters related to the topology of the phonatory system were extracted from the samples, for a total of 15 measures. The classification used quadratic discriminant analysis and includes the measures of accuracy, sensitivity, and specificity. Single measurements such as Shannons entropy, average diagonal length, and transitivity had only acceptable performance ratings (≥70%) in discriminating between individuals with and without voice disorders. The combination of the parameters average diagonal length, Shannons entropy, trapping time, length of the longest vertical line, tau, imbedding dimension, neighborhood radius, and transitivity produced the highest accuracy in discrimination (83.27%). Therefore, the performance of RQMs related to the formation of diagonal lines in classifying individuals with and without voice disorders was acceptable at ≥70%. A combination of RQMs showed good performance in discriminating between the study groups, with higher sensitivity and specificity.

Feature selection based on binary particle swarm optimisation and neural networks for pathological voice detection

Laryngeal pathologies directly affect the quality of the voice. In the last years, digital signal processing techniques have been applied for detection of vocal fold pathologies through speech signal analysis. In this work, a binary Particle Swarm Optimization (PSO) algorithm using Multilayer Perceptron (MLP) neural network is employed for the selection of the most significative features in a pathological voice detection system. The discrimination between healthy and pathological speech signals are obtained by 52 Haralick texture features, extracted from two-dimensional wavelet coefficients of the speech signals recurrence plots. The fitness function adopted is based in the maxima accuracy rate. Experimental results show that the use of PSO increased the accuracy rates with a minimum number of features.

Avaliação de Desempenho na Classificação de Patologias Laríngeas por Análise LPC de Sinais de Voz e Redes Neurais MLP

This article deals with the application of Linear Predictive Coding (LPC) and MLP neural networks in classifying healthy and voices affected by laryngeal pathologies (vocal fold edema, nodules and paralysis). The system performance is evaluated by varying the order of the LPC coefficients from 8 to 32. Classification rates above 98% were obtained in the healthy and pathological voice discrimination. Among pathologies, the accuracy values were higher than 90%. Keywords— Processing of Speech Signals, Linear Predictive Coding, Neural Network MLP, Laryngeal Pathologies.

Feature selection based on binary particle swarm optimization and neural networks for pathological voice detection

Laryngeal pathologies directly affect the quality of the voice. In the last years, digital signal processing techniques have been applied for detection of vocal fold pathologies through speech signal analysis. In this work, a binary Particle Swarm Optimization (PSO) algorithm using Multilayer Perceptron (MLP) neural network is employed for the selection of the most significative features in a pathological voice detection system. The discrimination between healthy and pathological speech signals are obtained by 52 Haralick texture features, extracted from two-dimensional wavelet coefficients of the speech signals recurrence plots. The fitness function adopted is based in the maxima accuracy rate. Experimental results show that the use of PSO increased the accuracy rates with a minimum number of features.

Voice pathology assessment using wavelet based on texture analysis of recurrence plots

This work proposes an efficient texture classification strategy performed in the wavelet domain in order to characterize healthy and pathological speech signals from recurrence plots (RP). The two-dimensional wavelet transform is applied to the recurrence plots at one resolution level. Thirteen Haralick texture features are obtained from each approximation and detail subband coefficients. In classification, multilayer perceptron (MLP) neural networks with cross validation are employed. Classification accuracy is improved and the number of features is reduced by particle swarm optimization (PSO). Results suggest that this method may be useful for pathological voice discrimination.