Browse Prior Art Database

An Effective Tool to diagnose grammar issues and aid grammar tuning in Speech Applications

IP.com Disclosure Number: IPCOM000127328D
Original Publication Date: 2005-Aug-23
Included in the Prior Art Database: 2005-Aug-23
Document File: 2 page(s) / 112K

Publishing Venue

IBM

Abstract

Speech applications are prone to problems that may only be discovered by usability testing the application in production environment by importing the logs generated by the servers into a tool and going over turn by turn. Poor accuracy in speech applications is a major concern, especially with more and more applications being speech enabled that are being deployed. A unique setup is a must to diagnose such problems to save time and energy.

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 52% of the total text.

Page 1 of 2

An Effective Tool to diagnose grammar issues and aid grammar tuning in Speech Applications

A speech application is prone to problems that may only be discovered by usability testing or by observation of the application in production by importing the logs into a tool. Poor accuracy in speech applications is a major concern, especially with more and more applications being speech enabled that are being deployed. When a user mispronounces a phrase or utters something that is unexpected and not in the grammar, the speech engine cannot match these inputs against any grammars and eventually will result as misrecognized inputs, consequently reducing the application's accuracy. A tool with unique setup will help in grammar tuning, a process where the existing logs are analyzed in a User friendly manner, and grammars modified, updating the grammar with commonly spoken phrases, removing highly confusing words and adding multiple pronunciations and re-testing the same application and comparing the accuracy results.

A tool with a unique interface may be designed, which aids in diagnosing grammar related issues and tuning each grammar. In particular, a User Interface where each grammar is displayed in a row (either a URL, if located on an external web server, or if inline source is available, a unique reference is used) and displaying all the utterance results that matched each selected grammar. The matched utterance result also displays such information as duration, recognized phrase and confidence scores for the utterance. Further, once a grammar is tuned, the same utterance may be run against these grammars on a live server and the results captured and listed for comparison.

Known solutions inclu...