San Diego, CA (FEB 12, 2010) — Developers building speech recognition applications using the LumenVox Speech Engine will find their jobs significantly easier with the release of the new LumenVox Speech Tuner.
LumenVox, a worldwide leader and innovator in speech recognition software, announced today the general release of the new Tuner tool as part of the new version 9.2 suite of LumenVox speech recognition products.
One of the problems often faced by speech recognition developers is troubleshooting errors. When a speech recognizer returns an incorrect result, application developers are often stumped as to how they can improve the accuracy.
"Speech recognition is often a sort of black box where you do not get to see what is happening under the hood," LumenVox President Edward Miller said. "What the Tuner does is to help pop the hood and let you take a look at exactly what's going on with the Speech Engine."
The Tuner will display the most common errors an application is experiencing, and produce useful statistics and graphs about those errors.
"Speech recognition developers can now zero in on problems with incredible speed and efficiency," Miller said. "The 9.2 revamp of the Tuner improves dramatically on our award–winning tool, and represents one of the most sophisticated speech tuning utilities on the market."
Some of the new or drastically changed features are:
While previous versions of the Speech Tuner provided much useful information, LumenVox concluded that too often it was information that was more useful to speech scientists than application developers, who often have little or no formal training in the science of speech recognition.
The version 9.2 Speech Tuner alleviates this, with a redesigned interface that clearly spells out accuracy, errors, and suggested settings for optimal improvement.
"With the new Tuner, you can see at a glance exactly how to improve your speech application. It helps to make speech tuning go much faster," Miller said.
One particular new feature that users will find helpful are suggestions for confidence score settings. When the Speech Engine performs speech recognition, it returns a numeric value called a confidence score along with the recognition. This score represents how confident the Engine is that the returned result matches what the user said.
It is common for applications that use speech recognition software to use these scores to determine whether to confirm an answer. For example, on a low confidence score, an application may ask users to repeat themselves.
The version 9.2 Speech Tuner can analyze the confidence scores being returned for an application and suggest optimal values at which the application should confirm, accept, or reject answers.
"Giving developers optimal thresholds based on their actual results takes the guesswork out of setting confidence thresholds," Miller said. "It is a task that used to take a lot of time to get just right, but now can be done in a matter of seconds."
Other features of the new Speech Tuner include a new grammar editor to help build and test grammars, a new transcriber to make performing transcription faster, and a significantly faster tester so developers can test proposed changes to an application without affecting production systems.
The LumenVox Speech Tuner version 9.2 is available immediately to users with current LumenVox software maintenance. It is free for users with current Speech Engine licenses, but requires two valid LumenVox Speech Engine licenses to run.
More information about the Speech Tuner can be found in the online documentation.
© 2016 LumenVox, LLC. All rights reserved.