The LumenVox Speech Tuner is a tool that allows you to analyze speech applications by reviewing response files — data about calls and interactions saved by a speech engine. "Tuning" an application means tweaking grammars and various settings in order to receive higher success rates and to pinpoint areas where a piece of speech recognition software may be failing users.
An important part of speech development is to evaluate how real users interact with your systems and then to make changes to the speech application. The LumenVox Speech Tuner aids this process by allowing you to transcribe call data, see statistics about recognition accuracy, and dynamically change grammars and test how the changes would affect the accuracy of your system.
If you are using the Speech Engine on Linux, you will need to run your Speech Tuner on a different machine as the Tuner is currently Windows–only.
You need to first configure the licensing. In the Tuner installation directory, by default C:\Program Files\LumenVox\Speech Tuner\bin, there is a file called license_client.conf. Open that file in a text editor and enter the IP address of the machine where your License Server is located. The Tuner requires a separate license from the Speech Engine license to use. For more information about licensing, Contact Us.
The other requirement is the Tuner needs an active connection to the Speech Engine in order to use the Grammar Tester functionality. By default, the Tuner looks on the local machine for an Engine. If you have the Engine on a remote machine, you will have to specify that IP address as follows:
The Speech Tuner only works with the LumenVox Speech Engine, though we will be adding support for other engines in the future. The current stable version of our Tuner does have support for other engines.
You will need to purchase a separate annual license to use the Tuner.
The response files are sound files saved by the Speech Engine and contain all the data used by the Tuner to evaluate your calls. These are saved in various folders depending on operating system and platform. These files all end in .callsre.
You may need to enable the "Save Sound Files" option for your platform in order for the Engine to generate them. If you are using our C or C++ API to interface with the Engine, see the Engine help files for more information on how to set that flag.
On Linux (including Asterisk), speech files are saved in the Engine installation directory. This is likely /opt/lumenvox/engine_7.0/Lang/Responses/ and then in folders by date. If you are using the Asterisk Dial Plan for your speech application, you will need to set save_sound_files=yes in the lumenvox.conf file located in the /etc/asterisk/ directory.
On Windows, the Engine by default saves files to the Engine installation directory, C:\Program Files\Lumenvox\Engine\Lang\Responses\ and then in folders by date.
If you are using the LumenVox Speech Platform, they will be saved to C:\Program Files\Lumenvox\Platform\Lang\Responses\ and then in folders by date.
The Speech Tuner does not load individual .callsre files. Rather, it takes all the response files in a directory and turns them into a single database file.
To create a database, you select New from the File menu. A window will open up prompting you to select a folder. This is the folder that has the response files you want to use for creating a database. You are just picking a folder and not specifying individual response files, so you will not see any .callsre files listed.
You will then see a new window that prompts you to save your database file. By default, it will be called calltrackermaster.bts. Again, you are saving a database and not any .callsre files, so the program will not display the response files.
Once you have loaded a database, simply click the Launch Call Browser button to view the calls in that database. From within the Call Browser, you can select specific calls from the top pane and view and listen to. The middle pane contains a breakdown of the call's interactions, and the very bottom pane has audio controls.
© 2016 LumenVox, LLC. All rights reserved.