Acoustic assessment of the voices of children using nonlinear analysis: proposal for assessment and vocal monitoring.

OBJECTIVE: To analyze the accuracy of recurrence measurements, both isolated and combined, to assess the intensity of vocal disorders in children. METHOD: A total of 93 children of both sexes (48 girls and 45 boys), aged between 3 and 10 years, participated. The vocal-deviation intensity was evaluated by the consensus of three speech therapists from the pronunciation of vowel /ε/ using the visual analog scale. In the acoustic analysis, eight recurrence plot characteristics were evaluated and extracted with neighborhood radius values that maintained the recurrence rate at 1%, 2%, 3%, 4%, and 5%. The classification was performed using quadratic discriminant analysis applied for individual and combined measurements. The performance was evaluated by measuring the accuracy, which related the cases correctly classified to all the analyzed cases. RESULTS: In the classification cases concerning individual measure performance, the trapping time and maximum length of the diagonal lines showed the best classification potential to discriminate between healthy and disturbed voices, with accuracy rates above 80%. In the healthy and mild deviation cases, the trend (TREND) measure was also relevant. For the mild versus moderate deviation classification, the best performance was obtained by the TREND measure (85.00% ± 7.64%). A gain was obtained in the classification rate when the measures of recurrence were combined, reaching an accuracy of 95.00% ± 5.00%, for discriminating between healthy voices and those with mild deviation. CONCLUSIONS: The measures of recurrence, either alone or combined, may be useful in detecting healthy and disturbed voices and in differentiating the intensity of vocal disorders in children.

Classificação de sinais de vozes saudáveis e patológicas por meio da combinação entre medidas da análise dinâmica não linear e codificação preditiva linear / Healthy and pathological voice assessment by means of nonlinear dynamic analysis measures and linear predictive coding

Análise acústica tem sido sugerida como uma ferramenta auxiliar não invasiva e de baixo custo ao diagnóstico de patologias laríngeas. Diversas são as técnicas empregadas, entre as quais técnicas baseadas no modelo linear de produção da fala e na análise dinâmica não linear de sinais de vozes. O primeiro método é baseado na Teoria fonte-filtro, em que a fonte é a laringe e o filtro é o trato vocal. Nesse modelo, os sons surdos são modelados por uma fonte de ruído aleatório e os sons sonoros por um trem de impulsos na frequência fundamental do locutor. Na abordagem não linear são considerados aspectos da voz humana, não explorados no modelo linear, tais como: variação temporal da forma do trato vocal, as ressonâncias associadas à sua fisiologia, as perdas devido ao atrito nas paredes internas do trato vocal, a radiação do som nos lábios, o acoplamento nasal e o comportamento dinâmico associado à vibração das pregas vocais. Neste trabalho são associadas as duas abordagens e avaliado o desempenho na classificação com as características, de forma individual, e a partir da combinação das mesmas. São empregadas oito medidas oriundas da análise dinâmica não linear (dimensão de correlação, quatro medidas de entropia, expoente de Hurst, maior expoente de Lyapunov e o primeiro mínimo da função de informação mútua), além de coeficientes LPC, obtidos a partir da análise preditiva linear. Os resultados sugerem a viabilidade da técnica empregada para a discriminação entre vozes saudáveis e patológicas em geral, como também entre vozes afetadas por patologias laríngeas distintas como edema, nódulos e paralisia nas pregas vocais.

Acoustic analysis has been suggested as a noninvasive aiding and low cost tool for laryngeal disease diagnosis. Several techniques are employed using either the linear model of speech production, or the nonlinear dynamic analysis of voice signals. The first method is based on source-filter theory, in which the source is the larynx and the filter is the vocal tract. In this model, the unvoiced sounds are modeled by a random noise source and the voiced ones by impulse train at the speaker fundamental frequency. In nonlinear approach, aspects of the human voice are considered, not explored in the linear model, such as temporal variation of the vocal tract shape, resonances associated with its physiology, losses due to friction in the vocal tract inner walls, sound radiation in the lips, nose coupling and dynamic behavior associated with vocal fold vibration. This work combines the two approaches and evaluates the performance in classifying the features individually, and from their combination. Eight measures are employed derived from the nonlinear dynamic analysis (correlation dimension, four entropy measures, Hurst exponent, the largest Lyapunov exponent and the first minimum of mutual information function), besides LPC coefficients obtained from linear predictive analysis. The results suggest the feasibility of the employed technique to discriminate between healthy and pathological voices in general, but also among specific laryngeal diseases as vocal fold edema, nodules and paralysis.