Text-based vs. vowel-based automatic evaluation of tracheoesophageal substitute voice

The Hoarseness Diagram, a program for voice quality analysis using recordings of sustained vowels, was compared to an automatic speech recognition system with a module for prosodic analysis. The latter computed prosodic features on a text recording. We examined whether the voice analysis of sustained vowel and text analysis correlate on a group of 24 male laryngectomees (average age: 60.6 +/- 8.9 years) using tracheoesophageal substitute speech. Each person read the German version of the text "The North Wind and the Sun" which consists of 108 words. Additionally, 5 sustained vowels were recorded from each patient. The correlation between the measures obtained by the Hoarseness Diagram and the prosodic features from the prosody module was determined. Parameters like jitter, shimmer, F0 and irregularity computed by the Hoarseness Diagram on vowel recordings show correlations of about -0.8 to prosodic features obtained from the text recordings. Hence, voice properties can reliably be evaluated both on a vowel and a text recording. The text analysis, however, offers also possibilities for automatic speech evaluation since it represents a real communication situation better.