User Is Stuck in ASR Training for an Extended Period of Time


On average the ASR system needs to process 90 minutes of audio and 50 reports to build a speech model and produce
drafts. * Some users may qualify sooner while others can take longer. If it seems to be taking an exceptionally long time
to get drafts, this article will help identify some areas to trouble shoot.


* This is per Nuance mapped document type, which is different than Emdat document types. Emdat document types could be linked to multiple
Nuance document types. Refer to this Enabling or Disabling a Document Type for ASR for more information.


Verify the user is enabled for ASR

A quick way to check this is to look in outstanding jobs. Pick a job and look at the details. Specifically look at the Activity
Log section.  If you see an activity log that looks like the one directly below, then ASR has not been enabled for the user.

Verify User 1.png

Compare the above log with the one below. The one below shows the user is going through ASR.

Verifty User2.png


If the activity log is like the first one, the user is not enabled for ASR.

Solution: Enabling or Disabling a User for ASR

If the activity log looks like the second one, verify the Disable Nuance Draft Generation box is NOT checked.


Is the Disable Nuance Draft Generation box checked?

If the user is enabled, make sure the Disable Draft Generation box is NOT checked. This option will prevent the Emdat
system from accepting drafts and make it look like the user is stuck in training.

Go to Client Maintenance > Users, edit a user, expand the User Information section to see the Disable Draft Generation
option.

User 1.png

If the box is checked, drafts may by getting generated but not being accepted by the Emdat system.

Solution: Uncheck the “Disable Nuance Draft Generation” box.

If you’ve verified the “Disable Nuance Draft Generation” box is unchecked, verify the user has met the 90 mins of audio
and 50 reports criteria. This is the minimum requirement to qualify for drafts.


Verify 90 mins and 50 reports

Run the ShadowScribe Author and MT Details report. This report will not tell you exactly if 90 mins of audio and 50
reports were submitted, but you can use the data from this report to determine this.

Run the report as a CSV. Start with the date the user was enabled for ASR and end with the current date. With the report
data, create a pivot table in excel. Summarize the data by delivered document type, user, audio in minutes, and number
of reports.

Report.png

In the example above, we selected 2 users. The user in red only has 2 documents and has submitted only 4 min and 11
seconds of audio. This does not meet the criteria to produce drafts. The other user in green, has submitted 263
documents and has over 4 hours of audio. This user should likely be getting drafts.

If the audio is less than or equal to 90 and/or the reports are less than 50, then the user has not met the criteria for
building the speech model.

Solution: Wait for user to submit 90 mins of audio and 50 documents. How long? This depends on speaker volume. If the
volume is high, give it a week or so and Re-run the report. If the volume is low you will need to wait longer.

If the user has 90 mins and at least 50 documents, then it’s possible it’s taking the speech engine longer to build the
model.

Solution: Wait longer to give the speech engine more time to build the model. How long? Try waiting another 90 mins
and 50 reports. If drafts are still not being produced, open up a support ticket for further investigation.