Application Tuning

The SASDK offers several options for debugging and tuning your speech application. When testing and deploying an application used by the general public, it is vital to maintain a log of all recognitions. These log files can be used to later identify problems with the grammar files.

Speech Debugging Console Logging

Speech Debugging Console allows you to enable logging which is helpful during the development and debugging process.

To enable logging, execute the following steps:

1.	Go to Start and Microsoft Speech Application SDK 1.0.
2.	Click Debugging Tools and Speech Debugging Console Configuration.
3.	Select Record Log Files and specify an existing log files base path. This is the location where log files will be stored. It might be helpful to specify an application folder within your Inetpub directory (i.e., C:\Inetpub\wwwroot\SpeechLogs\).
4.	Click OK to exit the Speech Debugging Console Configuration application. From here on, every time a recognition event occurs, whether it results in success or failure, a new XML flog file will be created.

Refer to Figure 4.7 to see what a log file resulting in a non-recognition exception looks like. In this case, the user entered the phrase "list all financial accounting 1" and would have received an error stating that the phrase was not recognized. Non-recognition exceptions are those resulting in a Type value of "NoReco."

Figure 4.7. Portion of an application log file in which the recognition was unsuccessful. In this case the user entered the phrase "list all financial accounting 1." This phrase was not recognized prior to a change in the grammar and would have resulted in an error. The non-recognition exception is indicated with a Type value of "NoReco."

Tip

You do not want to leave logging enabled indefinitely because it adds overhead to the speech process. However, the log files can be invaluable during a period in which you are identifying application grammar errors.

Examining Log Files from MSS

Trace log files, produced on the Speech Server, can also be used to debug a speech application. The SASDK provides three methods for extracting data from trace log files. In order to retrieve this data you must first enable trace logging on the Microsoft Speech Server by creating a log trace session. This is typically done by an Administrator using the logman utility provided with Windows XP. The session is defined from a command prompt and an output path is specified. Starting a trace session results in the creation of a file with an .etl extension

It is not necessary for developers to install MSS in order to access logging functionality. Instead, developers can ask the Administrator of the Microsoft Speech Server to send them the resulting .etl file. This text-based file can then be analyzed using any of three tools available with the SASDK installation:

1. MSSLogToText creates a log file containing event categories you may be interested in. These categories include such events as operational messages, audio streams, detailed analysis messages, and debug trace messages. The resulting text file contains a row for each event captured.

2. The MSSUsageReport analysis tool is used to create a text file containing events within a certain date range or from a certain URL. The resulting file contains different sections with detailed analysis about the restricted events. Sections included are a summary section, tasks report, latency report, and speech recognition report. This is useful for getting the big picture of how your application is performing.

3. MSSContentExtract is a tool used in combination with the MSSLogToText tool to identify certain suspect recognition events. The log tool is then used to recreate a wave file containing the incoming audio content for these events.

The three logging tools available with the SDSDK can be accessed by browsing to Microsoft Speech Application SDK 1.0, Debugging Tools, and Microsoft Speech Application SDK Command Prompt. This will open a command-prompt window that points to the location of the logging tools. Instructions on which command parameters to use can be accessed by typing the tool name followed by "/?".

Call Viewer

Chapter 2 introduced the Call Viewer Utility as a method for analyzing MSS .etl files. This is a great visual tool for developers to use while debugging an application. The utility is not installed by default, but is available with the SASDK. To install the tools, execute the following steps:

1.	From the setup directory used to install the SASDK, browse to the ..\Redistributable Installers\Microsoft Speech Application Log Analysis Tools\directory.
2.	Execute Setup.exe and follow the wizard to install the tools.
3.	Access the Call Viewer application by browsing to Microsoft Speech Application SDK 1.0 and Log Analysis Tools.

Once Call Viewer is installed, you will need to import the log data to a Microsoft SQL Server Database. The import can be done with a DTS (Data Transformation Services) package or with the MSSLogToDatabase command line tool. To import the log files with a DTS package, refer to the help file for the Log Analysis Tools. The help file, named LogAnalysis.chm is available in the ..\Redistributable Installers\Microsoft Speech Application Log Analysis Tools\Program Files\Microsoft Speech Application SDK 1.0\SDKTools\LogAnalysis\directory.