In our work we examine how TE speech is recognized by an automatic speech recognition system and if the quality of the substitute voice can (partially) be evaluated automatically. For this goal the evaluation results of the automatic system and a human expert group have to correlate. The patients' self-evaluation (SF-36, V-RQOL, VHI, Trierer Skalen) will also be part of an automatically computed measure which describes the quality of the substitute voice.

The speech data collection from the patients does not only take into account spontaneous speech from a close-talking microphone but also recordings via telephone.

Up to now we found a high correlation between the raters' intelligibility score for the patients' continuous speech and the word accuracy computed by an automatic speech recognizer both for close-talk recordings (r=-0.88) and for telephone speech (r=-0.80). Similar correlations could be obtained between automatically computed prosodic features and the rating criteria "match of breath and sense units" and "speaking effort". The visualization of speech disorders was successfully done using the Sammon transform.