The topic you requested is included in another documentation set. For convenience, it's displayed below. Choose Switch to see the topic in its original location.

How to: Enable Event Tracing of Call Activity

Speech Server 2007

This content is no longer actively maintained. It is provided as is, for anyone who may still be using these technologies, with no warranties or claims of accuracy with regard to the most recent product version or service release.

Use the Speech Server Administrator console to enable event tracing of call activity, including call audio. When event tracing is activated, you can specify where the Windows Event Trace Log (ETL) files are saved as well as the type of call activity information that is captured in them.

ETL files can quickly become very large, especially when including sampled audio. Ensure that the drive partition that stores the ETL files has enough space to accommodate the logging requirements. Event tracing is activated by default when you install Speech Server. Check the logging options to make sure only the information you need is being captured in the logs.

Tracing Events

ETL files collect tracing events from Speech Server and the applications running on it. You can specify the following types of events:

  • Analytics and tuning events - Detailed application data (including grammar logging) needed for Analytics and Tuning Studio. For more information about application tuning, see Speech Application Analysis and Tuning.
  • Platform events - Events originating from applications that are useful in debugging problems with the Speech Server services and processes. You can choose to have these events logged only when errors occur or always.
  • Application events - Events originating from the Speech Server services and processes that are useful in debugging application problems. You can choose to have these events logged only when errors occur or always.

You can choose all or none of these events to log in the ETL files. If you choose none and have event tracing activated, Speech Server still logs a minimum set of basic call reporting events to the ETL files.

Grammar Logging

You can refine grammars using Analytics and Tuning Studio with grammar-related data logged in the ETL files. With the grammar data, you can use Grammar Tuning Advisor to find out-of-grammar words and Re-recognizer to validate grammar changes. To activate grammar logging, you must select the Analytics and tuning events option.

Audio Logging

You can also capture call audio in the ETL files. Speech Server saves audio in chunks to the ETL files. Analytics and Tuning Studio provides a mechanism for playing the audio saved in the ETL files. For more information about how to play logged audio, see How to: Listen to Prompts and User Speech.

You can also use the command-line tool MssContentExtract.exe to process ETL files and extract the audio into .wav files, which can be played using Microsoft Windows Media Player. The tool extracts each turn in the call session (that is, each interaction between the caller and the application) into a separate .wav file. For more information about MssContentExtract.exe, see Extract Audio and Grammar Streams to File.

Speech Server provides the following options for logging audio:

  • Recognizer audio - Spoken input audio to which the recognition engine actively listens. This audio is logged over a specified percentage of concurrent calls. Recognizer audio does not include dual tone multi-frequency (DTMF) input.
  • Prompt audio - When you select to log Recognizer audio, you can also choose to log the audio of the prompts played to the caller. This audio is logged for the same percentage of calls as the Recognizer audio. Prompt audio includes prompts played from the prompt database and the text-to-speech (TTS) engine.
  • All audio - All audio from the call, which includes Recognizer audio, prompt audio, and any other spoken utterances from the caller to which the recognition engine is not actively listening. This audio is logged over a specified percentage of concurrent calls. All audio logging is useful for understanding the overall caller experience with the application and determining, for example, whether callers cannot barge in when they expect to or become frustrated. However, logging all audio requires significant storage space and should be sampled at a lower rate than Recognizer audio.

Setting Servers to Local Time

If you have a deployment with servers in multiple locations in different time zones, and reporting on log data across time zones is important, you need to configure each server to use the local time in its time zone. This enables you to apply time period filters in Speech Server Reports and in Analytics and Tuning Studio to normalize time differences across time zones.

  1. Open the Speech Server Administrator console.

    For more information, see How to: Start the Speech Server Administrator Console.

  2. In the console tree, expand the applicable group, and then click Servers.

  3. In the details pane, double-click the applicable server.

  4. Click the Trace Logging tab, and then select Enable trace logging.

  5. In the Location box, type the directory in which log files should be stored or click the browse (...) button to browse to the directory.

  6. In the Start new file every (min) box, type the number of minutes to wait before starting a new log file.

    The range of time that you can enter in this box is 0 to 1440 minutes (24 hours). A value of 0 indicates that the current log file captures session data indefinitely (in other words, never start a new file), unless a new file is started manually.

    All files are saved until explicitly removed; this setting provides the ability to control log file size.

  7. Under In addition to standard call reporting, select one or more of the following types of events to capture:

    • Analytics and tuning events.
    • Platform events. Select this option to record platform events always or clear it to record events only when error occur.
    • Application events. Select this option to record application events always or clear it to record events only when error occur.
  8. Select one or more of the following types of audio to capture:

    • Recognizer audio. Type the percentage of concurrent calls to sample.
    • Prompt audio. You must select Recognizer audio to activate this option.
    • All audio. Type the percentage of concurrent calls to sample.
      If you specify a greater percentage for Recognizer audio than for All audio, the net effect is to log audio for the All audio percentage first and then log the remaining percentage as Recognizer audio. For example, if you specify Recognizer audio at 25% and All audio at 10%, 10% of the calls will have all audio logged and 15% of the calls will have Recognizer audio logged, because the first 10% of the calls for Recognizer audio are captured in the calls being logged for all audio. If the percentage specified for All audio is greater than or equal to the Recognizer audio percentage, only the All audio percentage of calls are logged.

  9. Click OK.

You can also configure trace logging options using a Windows Management Instrumentation (WMI) script. For more information, see LogSettings Class.

Community Additions

© 2015 Microsoft