Domain

The target domain can be characterized as medical interviewing. Users of the system are doctors and patients in a clinic situation, with patients coming to the clinic with routine complaints such as toothaches and stomachaches, as well as more urgent problems like broken bones and wounds. The domain does not extend to full medical triage, due to the ASR-side difficulties with recognition of excited speech. In principle, though, the synthesizer we describe could be used under such conditions as well.

Our system supports both Arabic-speaking doctor / English-speaking patient and English-speaking doctor / Arabic-speaking patient situations, although realistically the latter is of greater concern.

This scenario presents an interesting challenge to our sociolinguistic model. In some Arabic-speaking communities, male doctors would not be able to treat female patients, and female doctors would not treat male patients. We have for the time being ignored this problem, and the system always speaks with a male voice, although the patient can be male or female if gender differences are supported by the translation model. In real-world use, however, we would want the user to have the option of providing a female synthesized voice when appropriate.