“The best testing plan for speech applications will combine the methods above or will be a variation of one or more of them. When collecting user feedback on a speech application, it’s usually a good idea to capture response files at the same time in order to perform more in-depth speech tuning. Full recordings should be enabled when doing Wizard of Oz testing, and so on. These methods will allow the designer to understand how real-world users interact with a speech system, and provide instructive input for improving and enhancing the quality of the dialog design. More generally, the same testing methodologies can also be adapted to other types of user interfaces outside of speech recognition. This includes the UX for web transactions, web chat, call center scripting, kiosk interfaces, and other systems where user input may be open ended or require semantic interpretation. The more real world testing that can be performed prior to building a system, the closer the launched product will serve its intended purpose right out of the gate, and the less rework will be required.”

(Stephen Keller ~ UX magazine)