Recognition accuracy no longer the gating factor

In the beginning of commercial speech applications, a key characteristic of an application was the accuracy of the speech recognizer that it used.  Mis-recognition frequency, or substitution error rates, varied substantially between speech recognizer implementations and configurations so it made sense.
 
Today, however, speech recognizers have matured to the point that recognition error rates are no longer the driving factor.  Typically errors are introduced by other parts of the application.  Our research shows that the most common errors experienced by callers are due to maintenance and tuning issues.  For example, a call that looks for a name that is not in the directory may be perceived to be a recognition error.  However, it is not and no improvements to the speech recognizer could fix this error.  Instead, it is the tools and service processes used to update and tune the application that can reduce these types of errors.  Parlance uses a connection rate, along with call duration, as key application performance metrics so that we can ensure we address the full scope of the problem.

Connection rates characterize complete system.
Recognition accuracy is the measure of a speech recognizer’s ability to correctly recognize intelligible, spoken requests for items listed in the speech recognizer.  Connection rate is the measure of a system’s ability to connect callers to their desired destination. 

Whereas ‘recognition accuracy’ characterizes the effectiveness of a sub-component of the system, ‘connection rate’ characterizes the effectiveness of the complete system.  Missing names, missing alternate names, and incorrect pronunciations do not impact the recognition accuracy of a system.  On the contrary, the connection rate is significantly impacted as the caller simply hangs-up when encountering such a system.  In simplest terms, recognizer accuracy provides insight into a system’s performance in the lab.  Connection rate describes the system’s performance in your environment.

These observations lead Parlance to develop processes to solve common business problems such as incomplete enterprise databases, in-frequent database updates, and the like.  The key to our success includes not only accurate recognition against in-grammar requests, but the inclusiveness and accuracy of the NameConnector registry.

Automation cures common connection issue - data
Parlance achieves the goal of 90+% connection rates by using unique tools to implement and maintain the registry of names and telephone numbers for its customers with great efficiency.  Some key features of our automation tools are : 

• Web-based names management – Manage names and telephone numbers through a Web-based databases that can be securely accessed simultaneously by multiple team members anytime, anywhere there is Internet access.
• Automatic data imports – Import names and numbers desired for use by NameConnector from multiple enterprise databases and other names lists, via numerous methods and protocols, and in virtually any format.  Furthermore, associate additional fields (such as alternate names) to the imported data whose contents persist through multiple imports.
• Automatic names processing – Use complex filters to merge imports in the manner most appropriate for the target application.  Automatically add common nicknames and view automatically generated database analysis reports useful in cleaning up your source database.

Parlance Managed Voice Service- we do it, so you don’t have to.
Parlance provides NameConnector as a managed voice service whose goal is to maximize the caller connection value for the end-user while minimizing the maintenance complexity for the telecommunications organization.  The result is simple -- a solution that is implemented within days of a purchasing decision and associates a 90+% connection accuracy.