Validação do protocolo de análise espectrográfica da voz baseada na relação com outras variáveis

Abstract

INTRODUCTION: The acoustic analysis of voice involves the use of digital processing, which can be quantitative or qualitative. Among the qualitative methods of acoustic analysis, visual inspection of spectrography is the main technique used. However, there is no validated assessment instrument that allows diagnosing voices as dysphonic and non-dysphonic. The Voice Spectrographic Analysis Protocol (VSAP), which is in the validation process, was used to classify dysphonic and non-dysphonic voices. OBJECTIVES: To verify the validity based on the relationship with other variables of the Spectrographic Analysis Protocol to discriminate between individuals with and without dysphonia. METHODOLOGY: This is a validation, cross-sectional, observational, documentary and quantitative approach study. Eighty-two voices were collected to generate spectrograms, and these spectrograms were saved as images and placed in a Google Forms questionnaire so that the VSAP could be applied. A voice expert judge with experience with the use of the VSAP was recruited to answer the questionnaire. Correlation calculations were performed to verify the convergent validity of the VSAP with the Cepstral Peak Prominence (CPP), Smoothed Cepstral Peak Prominence (CPPS), Dysphonia Acoustic Index (DAI) and General Degree of Vocal Deviation (EDV-G). In addition, the discrimination power of the protocol was sought to verify the validity of the predictive criterion, being a model based on the VSAP Score through the ROC Curve and another model of Logistic Regression. Values of accuracy, sensitivity, specificity, positive and negative predictive values and likelihood ratio were obtained to assess the performance of the VSAP. RESULTS: The VSAP showed positive and moderate correlations with DAI and EDV-GG, in addition to negative and weak correlations with CPP and CPPS. The ROC Curve sensitivity and specificity values based on the Score were 51.7% for sensitivity and 91.6% for specificity. The Logistic Regression model presented items Q2, Q12, Q13 and Q14 as benchmarks for the diagnosis of dysphonia, with values of 84.1% for accuracy, 84.2% for sensitivity and 84.0% for specificity. CONCLUSION: The VSAP showed convergent validity with the variables CPP, CPPS, DAI and EDV-G. In addition, the predictive criterion validity of the VSAP presented satisfactory values for the detection of dysphonia.