Acoustic Feature Analysis for Hypernasality Detection in Children
Genaro Daza (Universidad Nacional de Colombia Sede Manizales, Colombia), Luis Gonzalo Sánchez (Universidad Nacional de Colombia Sede Manizales, Colombia), Franklin A. Sepúlveda (Universidad Nacional de Colombia Sede Manizales, Colombia) and Castellanos D. Germán (Universidad Nacional de Colombia Sede Manizales, Colombia)
Copyright: © 2008
The present work analyzes the statistical effectiveness of different acoustic features in the automatic identification of hypernasality. Acoustic features reflect part of information contained in perceptual analysis; in part, due to their estimation is derived directly or indirectly from the vocal cords behavior. Consequently, it is convenient to apply multivariate analysis techniques in determining the effectiveness of voice features. The effectiveness is studied by using multivariate analysis techniques that are meant for feature extraction and feature selection, as well (latent variable models, heuristic search algorithms).
Key Terms in this Chapter
Nasalize: To make nasal or produce nasal sounds.
Latent Variable: A variable in the model that is not measured.
Heuristic Search: Any search strategy which makes use of heuristics to suggest the best nodes to consider at each stage of the search.
Multivariate Analysis: An analysis of the relationships between more than two variables.
Acoustic Features: Physical characteristics of speech sounds such as color, loudness, amplitude, frequency and so forth.
Bayesian Classifier: A simple probabilistic classifier based on applying Bayes’ theorem with strong (naive) independence assumptions.
Hypernasality: A quality of voice in which the emission of air through the nose is excessive due to velopharyngeal insufficiency; it causes deterioration of intelligibility of speech.