Objective: This study aimed to develop an index of breathiness degree based on fuzzy logic.
Method: This is a methodological study of instrument validation, which was divided into seven stages: preparation of the database samples; extraction of acoustic measurements; development of a computer program for auditory-perceptual judgment; assessment of inter-judge agreement; implementation of a fuzzy visual analog scale based on the judges' judgments; development of a multiple linear regression model with acoustic measurements; and evaluation and validation of the model's performance. The samples consisted of 295 patients with vocal complaints. Two speech tasks were used: the sustained vowel [a] concatenated with a sequenced speech task (counting numbers from 1 to 10). Five judges performed auditory-perceptual judgments of the general degree of breathiness using a visual analog scale (VAS). Acoustic measurements were extracted from the voice tasks using the free Praat software. Statistical measures were extracted using the R software. To enable the auditory-perceptual judgment of the judges, a computer program called VoxMore: Auditory-Perceptual Judgment was created using the Python programming language. In evaluating the agreement between the judges, the intraclass correlation coefficient was used. Next, a fuzzy visual-analog scale was created, and then a multiple linear regression model was applied to determine the degree of breathiness.
Results: The acoustic index was composed of four measures, including the Smoothed Cepstral Peak Prominence (CPPS), jitter, Glottal-to-Noise Excitation (GNE) at 3,000 Hz, and high-frequency noise (Hfno). The coefficient of determination was 0.80. The difference between the values predicted by the judges in the VAS and the values of the vocal breathiness index was less than 12.88 points.
Conclusion: The final version of the index was named the Breathiness Degree Index (BDI), which explained 80% of the variability and demonstrated robustness across different seeds tested. The regression model showed good adherence and can provide information about the degree of breathiness.