Selection and combination of acoustic features for the description of pathologic voices

Abstract

The glottal to noise excitation ratio (GNE) is an acoustic measure designed to assess the amount of noise in a pulse train generated by the oscillation of the vocal folds. So far its properties have only been studied for synthesized signals, where it was found to be independent of variations of fundamental frequency (jitter) and amplitude (shimmer). On the other hand, other features designed for the same purpose like NNE (normalized noise energy) or CHNR (cepstrum based harmonics-to-noise ratio) did not show this independence. This advantage of the GNE over NNE and CHNR, as well as its general applicability in voice quality assessment, is now tested for real speech using a large group of pathologic voices (n = 447). A set of four acoustic features is extracted from a total of 22 mostly well-known acoustic voice quality measures by correlation analysis, mutual information analysis, and principal components analysis. Three of these measures are chosen to assess primarily different aspects of signal aperiodicity, while the fourth one indicates the noise content of the signal. All analysis methods lead to the same feature set that consists of a measure of period correlation, jitter, shimmer, and GNE. The two-dimensional projection of this set named "hoarseness diagram" allows a graphical illustration of voice quality that can be easily interpreted.

It Sounds Like You Have a Cold! Testing Voice Features for the Interspeech 2017 Computational Paralinguistics Cold Challenge

“…or voice quality (jitter, shimmer, HNR, etc.) have been widely reported as changing due to brain and mental health disorders such as depression, schizophrenia, and Parkinson's [8,9,10,11,12].…”

Pathological Voice Detection Using Efficient Combination of Heterogeneous Features

“…A voiced speech signal, x(k), can be expressed as (1). Table 1 shows the average equal error rates (EERs) and the 95% confidence intervals (CIs) of the MFCC-based GMM algorithm according to the number of the Gaussian mixtures and the number of the mel frequency-based filter bank energies.…”

“…These parameters are based on the fundamental frequency. Correlations between these parameters and pathological voice detection have been demonstrated [1]. However, it is not easy to estimate the fundamental frequency in a pathological voice correctly.…”

A Hybrid of Genetic Algorithm and Support Vector Machine for Feature Reduction and Detection of Vocal Fold Pathology

“…Feature set Feature Reduction M ethod Classifier [10] 25 Acoustic parameters given by MDVP PCA Support Vector M achine [11] Spectral perturbation PCA K-M eans clustering [12] Acoustic feature, noise PCA Threshold [13] Linear prediction coefficients PCA K-Nearest Neighbours [14] M el-frequency-cepstral-coefficients LDA Gaussian M ixture M odel [15] Spectral -Artificial Neural Network…”

“…Also different approaches for feature reduction are used such as Principal Co mponent Analysis (PCA) [10][11][12][13] and Linear Discriminant Analysis (LDA) [14]. In the proposed method, the GA-based feature reduction has been used and the results of experiments show its better performance in co mparison with the PCA.…”