Acoustic assessment of neurogenic voice disorders in a clinical setting

A new tool for the computerized clinical assessment of neurogenic voice disorders is presented. It is part of a more comprehensive PC-based system for the diagnosis of speech impairments in patients with neurologic disorders, including modules for the assessment of speech rate, fluency, and articulation, in addition to the voice module. The design of this instrument is optimized with regard to stability and precision rather than computational speed. The voice analysis parameters used to-date are fundamental frequency and its variation over time, intensity and its variation over time, spectral tilt, and the cepstrum. The system controls all diagnostic steps, from the presentation of stimulus materials to the evaluation of results. Although the procedure is highly automatized, it allows for interactive control of recording quality, handling of algorithm instabilities, and correction of implausible results. INTRODUCTION Virtually all neurologic diseases interfering with general motor control can also involve laryngeal motor functions and, as a consequence, lead to voice disorders. Usually, neurogenic dysphonia is part of a more general dysarthric disorder which also includes motor problems of the respiratory and the supralaryngeal muscles during speaking. The voice problems encountered in these patients may either reflect direct involvement of the laryngeal motor system or a compensatory reaction to respiratory or articulatory dysfunctions [1]. Clinical assessment of neurogenic voice disorders is in the first instance based on perceptual evaluations of the patient's voice [2]. This method, however, is inherently subjective and, in most cases, not sufficiently reliable for the purpose of assessing spontaneous or therapeutically induced change. More objective methods of evaluating vocal performance are based on speech signal analysis techniques. Although there is abundant literature on acoustic measures of vocal pitch, voice quality, and voice stability, and although several measurement systems are commercially available, many clinicians are still hesitant in actually using these measures in their routine work. One of the reasons for their reluctance may be the fact that most systems are too unconstrained and fail to provide a stable framework, from suitable assessment protocols over recording tools and interactive analysis methods to comprehensive tabular and graphical presentations of the relevant data. The system presented here tries to provide a user-friendly assessment tool for use in the diagnosis of neurogenic speech disorders. The assessment protocol encompasses sentence repetition, sustained vowel production, word repetition, and articulatory diadochokinesis. Acoustic variables are focussed on speech rate, fluency, consonant articulation, vowel articulation and voice. Several design principles were adopted to increase the acceptance of the system in everyday clinical applications: Signal analysis methods were optimized on precision rather than computational speed. No manual segmentation is required. Results can be checked for their plausibility and can be accepted, corrected, or rejected on different stages of the analysis process. Individual tasks can easily be repeated in cases of questionable data. The results are presented graphically or in tabular form. MAVEBA 1999, Firenze, Italy 83 Models and Analysis of Vocal Emissions for Biomedical Applications