RESUMO
Digital recordings of thyroarytenoid (TA) myoelectric activity (via percutaneous hooked-wire electrodes) were obtained for 10 normal control subjects and 10 subjects with spasmodic dysphonia during performance of five tasks of varying complexity: quiet breathing, Valsalva maneuver, whispered /i/, voiced /i/, and "beep beep went the heap." Time-frequency power spectral density functions, PSD(f,t), were determined for a selected segment of the signals, and measures of median frequency, mean frequency, bandwidth, and center frequency of PSD(f,t) were derived for each sample point. Statistical median, mean, standard deviation, minimum, maximum, and mode of the power spectral density measures were computed to compose feature vectors for each TA myoelectric recording. Statistical pattern recognition procedures using maximum likelihood classification tests were applied to the feature space to discriminate disordered from normal speakers for each task. Findings indicate a high level of discriminability between subject groups for phonated speaking tasks in contrast to low levels of discriminability for whispered and nonspeech tasks. Graphical presentations of three-dimensional PSD(f,t) plots are given that illustrate changes in spectral characteristics of TA EMG at the onset of laryngospasm.