Tuning the System

Once the system is up and running, a few problems could conceivably arise. One of the largest issues when designing a Speech Application, is compensating for misrecognitions due to people speaking differently from the phonetic dictionary definitions included in the Engine\Platform. For instance, in a list of all 50 states, the Platform will understand the word "Mississippi". However, when spoken quickly, it often comes out "Mizippi". To allow for these sort of possibilities, tuning the system is recommended.

What is Tuning?

Tuning is using the Call Log Viewer to figure out exactly what the caller is saying and what the system is interpreting. While "Mizippi" has a low confirmation value for Mississippi, it has a higher value for Missouri. So often, the system will send a "Mizippi" response to the Missouri concept. To compensate for this, the programmer can listen to the callers speech and subsequently revise and add phrases based on real life pronunciation. For the Mississippi issue, the programmer would add the new phonetic spelling (Mizippi or {M IH Z IH P IY}) to the list of phrases in the Mississippi concept. Now when a caller says "Mizippi", the system will recognize it as belonging to the Mississippi concept and he or she will be sent to the correct state.