9.2 Speech Tuner

New and Improved Speech Tuner

(FEB 2010) — LumenVox is excited to announce the release of its new and improved Speech Tuner application, as part of the 9.2 software release. Completely rebuilt from the ground up, this new version provides a more thorough and robust set of tools for speech application tuning than previously available. The Speech Tuner requires a separate annual subscription license.

The LumenVox Speech Tuner is a powerhouse combination of tools that helps you in fine–tuning your speech applications. Any speech–driven system needs the occasional tune–up to keep it running with maximum accuracy. Whether your grammars need tweaking or your Engine's configuration needs to be adjusted, the Speech Tuner is your go–to tool.

What Does it Do?

The Speech Tuner allows you to analyze your recorded calls for areas of improvement. One of its primary features is the Call Browser, which allows you to load, filter, and listen to calls logged by your LumenVox Speech Engine.

Transcription of your calls is an integral part of the tuning process. Without comparing what callers said with what the Speech Engine recognized, your tuning efforts will only be partially effective. Once you have used our Transcriber tool to complete this step, you can accurately determine which decodes were incorrect and focus on fixing problems.

One of the areas most commonly focused on in tuning projects are grammars. If your analysis points to words that are missing from your grammar, you want to be confident that when you add them into the grammar that you've done so correctly. The Speech Tuner's Grammar Editor provides you a tool to load, modify and then test your grammars for correct syntax and logic.

Another means of extracting accuracy from a speech application is changing how the Speech Engine behaves. However, you shouldn't have to change your production system on a hunch to see if those changes had a positive effect on recognition accuracy. The Tuner's interface with the Speech Engine allows you to make Engine changes and then run audio through the Engine to see, in real–time, whether your changes were for better or worse. This will save you a lot of time by helping you to cut down on the number of tuning cycles.

What's New?

New GUI: The first, and most obvious, area of improvement is the user interface. With ease–of–use always in mind, we rebuilt the GUI to be more efficient and logical. Gone are the separate popup windows for each application, replaced with a clean, tabbed, single screen. We've also added a Summary tab to the Tuner, which provides you with an at–a–glance view of your current set of calls and connected platforms.

Better Data, Statistics and Reporting: Tuning is all about statistics. In an effort to help you understand them, we have improved the Tuner's reporting capabilities, included some new statistics, and made the results more accessible than ever before. For example, in almost any pane of the Tuner, you can right–click and export the data to a .csv file for analysis in an application of your choice. Our new Confidence Histogram will even calculate where you should be setting your confirmation and rejection thresholds, figuring the optimal settings for your data.

The Call Indexer: A brand new feature to the Tuner is our ability to make the Tuner instantly aware of newly logged calls through indexing. When you start the Tuner up, it will know where your logged calls are, and provide you with automatic quicker access to them. No more need to browse to the .callsre files — we've already done it.

Color–Coded Results: As part of the GUI enhancements, we have used a customizable color scheme to indicate tested results. So, once you have completed your transcription and run the calls through our improved Tester, those calls are immediately visible and sortable by categories such as correct, incorrect, out–of–grammar, or several other results.

Improved Grammar Parsing: The new Tuner is most effectively used alongside a running LumenVox Speech Engine. When testing a grammar, the Tuner will actually use the Engine to parse the input, whereas the old version used an approximation of the Engine. This allows you to be confident that the results you get in your Tuner tests will be reflected in production.

Send to LumenVox: We have added a new button at the top of the Tuner's menu bar: Send to LumenVox. When clicked, a wizard walks you through sending the logged calls, grammars, etc. to LumenVox for assistance. When we receive the email, we will contact you to get more details on how we can help you. Since we are experts on tuning LumenVox applications, let us do the heavy lifting.

Stability and Speed: The new Tuner has a revamped database design which has resulted in a far more stable tool. It means the Tuner is much less likely to run into problems when building a database, and it makes loading and creating databases faster than ever before.

How do I get it, and what do I need?

The LumenVox Speech Tuner requires a separate annual subscription license. To purchase this subscription please contact a LumenVox sales representative or Contact Us.

The Speech Tuner is a Windows–only application, so if you're one of our many Linux users, you'll need a standalone machine running Windows XP, 2003 Server, Vista, or Windows 7. In order to make full use of the Tuner's Tester, you'll need to make sure that it's pointed to a running Speech Engine with at least two licenses available.

© 2018 LumenVox, LLC. All rights reserved.