The Speech Overkill: Possible Solutions

I had a very interesting discussion with Jahanzeb Sherwani, a PhD student at Carnegie Mellon University who was visiting Pakistan on vacations in July. He pointed out the ineffectiveness of many Telephony Speech applications in the US that simply were too difficult for callers to use, and often irritating because of the way they would prompt users for input, without clearly understanding the spoken input provided to them, signifying a very weak VUI design. Prompting callers without specifically telling them what the system expects and how it expects the spoken input can elicit responses that can't always be defined using Grammar files. A hypothetical example could be

1. System: Welcome to the Apparel Directory. How many I help you?


2. Caller: I am just trying to check out the weaknesses in VUI design in your Speech application.


3. System: What designs are you interested in?


4. Caller: I am not interested in any Apparel Designs!


5. System: Good choice. Should I tell you the names of all the vendors or would you like to connect to a particular one?


6. Caller: Aarghhhh! Whats wrong with you motherfucker??!!!


7. System: Connecting to Mothercare now. Thank you for calling.

While Microsoft and IBM have come out with their (somewhat ambigiuous) guidelines for designing VUI's, a much better option would be to provide prompt controls, just like the Speech input controls that come with the Speech Application SDK for common answers as Yes/No/Cancel etc. as guidelines are not easy to enforce, and prompt controls may be able to provide boilerplate text for creating effective prompt messages depending on the type of input required. For unknown reasons, the speech developer still has to hand-code GRXML code and databind it to a table's column to create dynamic grammars. Is someone listening ... errr reading?