As discussed in the introduction, the purpose of the LumenVox Speech Tuner is to make tuning fast and easy. Briefly, a normal use of the LumenVox Speech Tuner to tune an application would encompass the following steps:
- Download and install the Speech Tuner.
- If necessary, configure licensing.
- Collect the data from an application using the LumenVox Speech Engine using the Call Indexer.
- Alternatively, you may manually load the data into the Tuner.
- Listen to the audio and look at the recognition results to get an idea of how callers are using the system.
- Transcribe the dataset (or a representative sample) to get detailed accuracy statistics.
- Identify problem areas, such as the most misrecognized words or various utterances that are not contained in the grammar.
- Make changes to the grammar files to attempt to fix the problems identified.
- Run a test to see how the changes affect accuracy.
- Evaluate your results, and either deploy the changes to production or continue tuning.
The Tuner Screens
The Speech Tuner is divided into the following sections. For more information on any given section, please see the full reference page.
The first screen you will see when using the Tuner is the Summary screen. Once a dataset has been loaded, this screen will allow you to view statistics about the data. You can see information about the grammars used in the dataset, a list of computers where the datais coming from, call volumes over time, and accuracy information (data must be transcribed for accuracy details).
The call browser allows you to quickly listen to calls and the individual utterances within a call. It is a good way to get a quick overview of how your users are actually interacting with your speech application.
The grammar editor is one of the most useful portions of the Tuner, as it can be used even before you have any call data. It can be used to build new grammars or troubleshoot existing grammars. In addition to providing syntax highlighting for grammar files, you can also type a word or phrase into the input box, press Parse, and immediately see how the LumenVox Speech Engine interprets that phrase through the grammar.
This is a vital part of troubleshooting grammar development. You can also use the grammar editor in conjunction with the tester, for tuning purposes.
The transcriber is used to transcribe data. It makes it easy to listen to utterances and quickly transcribe large sets. The transcriptions are used by the tester, to generate accuracy statistics, and to pinpoint any problems a speech application may be experiencing.
Text To Speech Editor
The TTS editor can be used with or without existing data. It can be used to build or test SSML documents as would be used within applications when performing Text To Speech synthesis operations. This can be particularly useful to check the syntax of SSML documents as well as hearing how they will sound without having to run your application.
When SSML documents are perfected, these can simply be exported as needed and incorporated in your applications, whether this is using VXML to connect to LumenVox or connecting directly to one of the LumenVox TTS APIs.
The tester displays information detailed about the loaded dataset. If the data has been transcribed, the tester can provide accuracy information. The tester's main purpose is to test changes, such as those made in the grammar editor. Once you have made changes to a grammar, you can go to the tester, clear the old results, and run the audio through the Speech Engine again, using the modified grammar. You can then compare the new results to the old to see if your changes were harmful or beneficial.
The even log view displays a high-level list of events encountered by the Speech Tuner, such as when it loads a database and so on.