Voice diseases have been increasing dramatically in recent times due mainly to unhealthy social habits and voice abuse. These diseases must be diagnosed and treated at an early stage, especially in the case of larynx cancer. It is widely recognized that vocal and voice diseases do not necessarily cause changes in voice quality as perceived by a listener. Acoustic analysis could be a useful tool to diagnose this type of disease. Preliminary research has shown that the detection of voice alterations can be carried out by means of Gaussian mixture models and short-term mel cepstral parameters complemented by frame energy together with first and second derivatives. This paper, using the F-Ratio and Fisher's discriminant ratio, will demonstrate that the detection of voice impairments can be performed using both mel cepstral vectors and their first derivative, ignoring the second derivative

Dimensionality Reduction of a Pathological Voice Quality Assessment System Based on Gaussian Mixture Models and Short-Term Cepstral Parameters

It is well known that vocal and voice diseases do not necessarily cause perceptible changes in the acoustic voice signal. Acoustic analysis is a useful tool to diagnose voice diseases being a complementary technique to other methods based on direct observation of the vocal folds by laryngoscopy. Through the present paper two neural-network based classification approaches applied to the automatic detection of voice disorders will be studied. Structures studied are multilayer perceptron and learning vector quantization fed using short-term vectors calculated accordingly to the well-known Mel Frequency Coefficient cepstral parameterization. The paper shows that these architectures allow the detection of voice disorders-including glottic cancer-under highly reliable conditions. Within this context, the Learning Vector quantization methodology demonstrated to be more reliable than the multilayer perceptron architecture yielding 96% frame accuracy under similar working conditions.

Automatic detection of voice impairments by means of short-term cepstral parameters and neural network based detectors

This paper is a tutorial on linear, state space, model-based methods for certain nonlinear estimation problems commonly encountered in signal and data analysis. A prototypical problem that is studied is that of estimating the frequencies of multiple, superimposed sinusoids from a short record of noise-corrupted data. The approach expounded however, is applicable to a vast range of nonlinear signal analysis problems and applications in direction finding and damped sinusoid retrieval are dealt with in some detail. The benefits that result from using a state space description of the signal are highlighted in this paper. It is shown that state space models provide an elegant tool for exposing the structure present in the problem. The approach also allows for robust parameterization of the model with respect to finite precision errors. The robustness of the parameter set is complemented by the availability of numerically robust tools to estimate the parameters. The resulting algorithms are compatible with multiprocessor implementations. >

Model based processing of signals : a state space approach. Prolog

A computer system for acoustic analysis of pathological voices and laryngeal diseases screening.

Developmental aspects of infant’s cry melody and formants

For pathological voices, spectral noise is closely related to the degree of perceived hoarseness. In this paper, noise variations are tracked during an utterance by means of an adaptive version of the normalized noise energy method (H. Kasuya et al., J. Acoust. Soc. Amer., vol. 80, no. 5, p. 1329-34, 1986). A first step is devoted to pitch estimation, which allows defining the varying optimal time window length for noise retrieval, avoiding empty "dip" regions. The approach Is tested on synthetic data and applied to real data coming from cordectomised adult male patients.

Adaptive noise energy estimation in pathological speech signals

This paper aims at providing new objective parameters and plots, easily understandable and usable by clinicians and logopaedicians, in order to assess voice quality recovering after vocal fold surgery. The proposed software tool performs presurgical and postsurgical comparison of main voice characteristics (fundamental frequency, noise, formants) by means of robust analysis tools, specifically devoted to deal with highly degraded speech signals as those under study. Specifically, we address the problem of quantifying voice quality, before and after medialization thyroplasty, for patients affected by glottis incompetence. Functional evaluation after thyroplastic medialization is commonly based on several approaches: videolaryngostroboscopy (VLS), for morphological aspects evaluation, GRBAS scale and Voice Handicap Index (VHI), relative to perceptive and subjective voice analysis respectively, and Multi-Dimensional Voice Program (MDVP), that provides objective acoustic parameters. While GRBAS has the drawback to entirely rely on perceptive evaluation of trained professionals, MDVP often fails in performing analysis of highly degraded signals, thus preventing from presurgical/postsurgical comparison in such cases. On the contrary, the new tool, being capable to deal with severely corrupted signals, always allows a complete objective analysis. The new parameters are compared to scores obtained with the GRBAS scale and to some MDVP parameters, suitably modified, showing good correlation with them. Hence, the new tool could successfully replace or integrate existing ones. With the proposed approach, deeper insight into voice recovering and its possible changes after surgery can thus be obtained and easily evaluated by the clinician.

C. Manfredi

Papers

Adaptive noise energy estimation in pathological speech signals

A new insight into postsurgical objective voice quality evaluation: application to thyroplastic medialization