Automatic Assessment of the Spasmodic Voice

The present work proposes automatic methods for determining the extent of dysphonia in patients with spasmodic dysphonia. Spasmodic dysphonia is a disorder of the larynx which involves involuntary contractions of the vocal folds during speech production. Numerous acoustical aspects of a voiced spasmodic speech signal which indicate the presence of these vocal spasms have been identified, including voicing breaks, shifts in fundamental frequency, and signal aperiodicities. We have developed algorithms for assessing the severity of dysphonia which are based on periodicity analysis, fundamental frequency estimation, and the processing of signals in the time-frequency and time-lag domains. These methods are tested on thirty-seven recordings of patients with spasmodic dysphonia reading a phonetically balanced passage of text. The algorithm’s assessments are evaluated by comparing them to ratings generated by trained listeners on the same data set. A baseline algorithm achieves a mean rank correlation of 0.76 when compared to the five judges, as compared with a mean rank correlation of 0.83 among the judges.