Acoustic analysis and digital signal processing for the assessment of voice quality

Individual Voice Dimensions' Prediction of Overall Dysphonia Severity on Two Auditory-Perceptual Scales

BACKGROUND

Auditory-perceptual evaluation of dysphonic voice is an essential clinical activity that characterizes the nature of dysphonia and aids in planning its clinical management. Although there are multidimensional acoustic measures that correlate well with overall severity ratings, they tend to include measures that have only small or moderate correlations with individual voice characteristics frequently perceptually measured (e.g., breathiness or roughness). Given this difference between perceptual and acoustic measures, it is unclear how much individual voice characteristics contribute to a listener's perception of overall severity of dysphonia.

PURPOSE

The purpose of this study was to explore individual voice characteristics' relative contribution to the rating of overall dysphonia severity and to explore sex-related differences.

METHOD

Two hundred ninety-six voice samples were accessed from the Perceptual Voice Qualities Database. Roughness, breathiness, asthenia, strain, pitch, and loudness ratings from the Grade, Roughness, Breathiness, Asthenia, Strain and Consensus Auditory-Perceptual Evaluation of Voice scales were used to predict overall voice quality severity in linear regression with bootstrapped coefficients.

RESULTS

Roughness, breathiness, and strain were the strongest predictors of overall severity. Asthenia and, to a lesser extent, pitch were also significant predictors of overall severity. Loudness was not a significant predictor. There were several sex-related differences noted, as well as differences related to the scale used.

CONCLUSIONS

Breathiness, roughness, and strain were all important predictors of overall severity for all regressions. Clinicians should be aware of scale-related differences if they are using auditory-perceptual measures to choose voice therapy targets. Analyses accounting for perceptual strategy differences were recommended for future studies.

Lightweight Deep Learning Model for Assessment of Substitution Voicing and Speech after Laryngeal Carcinoma Surgery

Laryngeal carcinoma is the most common malignant tumor of the upper respiratory tract. Total laryngectomy provides complete and permanent detachment of the upper and lower airways that causes the loss of voice, leading to a patient's inability to verbally communicate in the postoperative period. This paper aims to exploit modern areas of deep learning research to objectively classify, extract and measure the substitution voicing after laryngeal oncosurgery from the audio signal. We propose using well-known convolutional neural networks (CNNs) applied for image classification for the analysis of voice audio signal. Our approach takes an input of Mel-frequency spectrogram (MFCC) as an input of deep neural network architecture. A database of digital speech recordings of 367 male subjects (279 normal speech samples and 88 pathological speech samples) was used. Our approach has shown the best true-positive rate of any of the compared state-of-the-art approaches, achieving an overall accuracy of 89.47%.