Tuning is the process by which speech recognition application developers evaluate the
success of their applications and make changes. It is an iterative process, meaning
that after one round of changes is made to an application, the results are reviewed and
more changes are made.
Industry best practices usually recommend that approximately 40% of the overall time
spent building a speech application should be spent tuning it. What separates tuning
from the normal process of testing an application is that tuning should rely heavily
on actual production data.
Unlike traditional application development, speech applications are hard to test in
a development environment. Many real-world variables affect application performance,
such as: unexpected input from users, accented callers, difficult line conditions,
and noisy backgrounds.
Simulating all of these conditions is very difficult, which is why the speech
recognition industry relies on tuning. The tuning process starts by putting a speech
application live (often you will start with a small subset of users who act as
real-world beta testers).
By listening to calls, transcribing them, and then running tests, you can
see the accuracy of the application and quickly pinpoint any problems in your application.
Tuning often means adjusting grammars to include new utterances, or rewriting prompts
to be more clear, or redesigning an application's logic to help better guide callers through
the flow of the call.
If you are brand new to Tuning, we recommend you look at some of the following resources
available elsewhere on the LumenVox Web site:
You should also read our Getting Started page for information
about how to start using the Tuner quickly.