In addition to the LumenVox text logs, the LumenVox Media Server (and any client application) can also save more detailed files used primarily for speech tuning. Speech tuning is a complicated topic outside the scope of basic troubleshooting, but know that these response files — often called ".callsre files" since their file extension is .callsre — can be useful in troubleshooting all manner of problems.
Anytime there is a request for a recognition or a synthesis, the Media Server can save binary .callsre files that include a lot of data about the request. Included in .callsre files are:
- The audio to be recognized
- The grammars to be used in the recognition
- The text/SSML to be synthesized
- The result of the synthesis
- The result of the recognition
- All relevant ASR/TTS settings that were in use for the event
In order to examine these files, you must open them in the LumenVox Speech Tuner
utility. Once they are opened in the Speech Tuner, you may listen to the audio, look at grammars/SSML, and more.
These .callsre files are particularly useful when trying to pinpoint speech problems such as bad barge-in (e.g. barge-in is occurring too early or too late/not at all), poor recognition performance, strange TTS pronunciations, etc. They are not generally useful for troubleshooting lower-level problems (e.g. bad installations, misconfigurations), but are generally the starting point for helping users with more nebulous problems ("This thing isn't recognizing me well!").
Enabling Response Files
By default, the Media Server will not generate .callsre files. Unlike the text log files, there is no automatic cleanup/rolling of .callsre files. Because these are binary files that contain audio, it is possible that a single .callsre file can take up 100KB or more in disk space. On a busy server it is possible for .callsre files to be generated at a very rapid pace, so if you are running with them enabled in production it may be worthwhile to build some kind of cron job/scheduled task to clean them up from time to time.
In order to enable .callsre file generation, you must edit media_server.conf and change enable_sre_logging from the default 0 to a value between 1-3. The value range represents the amount of audio that will be saved:
- Saves basic information in .callsre files. For ASR events, this includes only the audio from when BARGE_IN occurred up to the END_OF_SPEECH.
- In addition to the information stored with the SAVE_SOUND_FILES_BASIC setting, when speech is streamed in to the Voice Activity Detection module, all of the collected data prior to a StreamCancel command is saved. This is usually when a NO_INPUT or TIMEOUT event occurs. This option collects data only when there is something to debug, which offers a balance between disk usage and retaining important information for debugging and is particularly useful when diagnosing NO-INPUT problems.
- Used to collect all streamed data in all cases. This option collects all streamed data, whether there was a NO_INPUT or TIMEOUT event or not, including untrimmed audio from 'good' decodes. This option can be useful in diagnosing barge-in problems as well as other potential issues.
Response File Location
The .callsre files are stored in the following locations on disk:
- C:\Program Files\LumenVox\Engine\Lang\Responses and then in directories by date (using the MM.DD.YYYY format).
- /var/lumenvox/responses and then in directories by date (using the MM.DD.YYYY format).