Objective: This study aimed to develop and validate an index of the General Degree of Vocal Deviation (GG) using fuzzy logic.
Method: We analyzed 295 vocal samples, including sustained vowels ([a]) and connected speech, from dysphonic and non-dysphonic adults of both genders. The process involved concatenating each sample into a single file for perceptual-auditory judgment (JPA), which five speech-language pathologists conducted using a 100 mm visual-analog scale. Additionally, we extracted 50 of the most frequently reported acoustic measurements using a script in the free Praat software. We aggregated the JPA results for GG into a single vector and formed clusters using the k-means algorithm, resulting in four categories: healthy, mild, moderate, and severe. Each category included a calculated centroid and dispersion. We then developed a fuzzy EAV (Escala Analógica Visual), modeling pertinence functions for each category based on centroids and dispersions.
Results: We weighted the results from the speech-language pathologists using the intraclass correlation coefficient (ICC) and implemented the fuzzy system with Mamdani inference and defuzzification by the centroid method. A linear regression model identified five acoustic measures—glottal-to-noise excitation (GNE 1000 Hz), dispersion of smoothed cepstral peak prominence (CPPSD), harmonic-to-noise ratio (HNR), the amplitude difference between the first harmonic and the third formant (H1A3), and the third quartile of the fundamental frequency (foq3)—as capable of explaining 84% of the variability in GG.
Conclusion: The mathematical representation of the General Degree of Vocal Deviation Index (GDI) is GDI = 106.16 – 63.36 _ ______ + 4.04 _ ________D – 2.43 _ ______ + 0.46 _ ________ + 0.10 _ ________. The GDI estimated the GG with an absolute difference of less than 11 points compared to the SLP assessment. Thus, the final GDI score is presented in the same range as the EAV (0 to 100 mm) and presents values compatible with expert assessment.