Troubleshooting Dictation Recognition Issues (Administrator Process)
  • 7 Minutes to read
  • Dark

Troubleshooting Dictation Recognition Issues (Administrator Process)

  • Dark

Article summary

If a user is having issues with recognition of dictation transcripts, an administrator should follow the steps below:  

  1. Listen to the misrecognized dictations. 
    • On the Review & Correct page in nVoq Administrator you can find and review dictations that include misrecognized words. Be sure to set the date range on the page to include the dates that the dictations occurred, and use the Look for Word filter to find a specific term in dictation transcripts.
      • If the dictation audio is poor due to microphone issues (not background noise), there may be a problem with the user's microphone. If so, send feedback to the user regarding what they can do to fix the issue.
        • If the user is dictating on a computer with nVoq Voice a partner client:
          • The user may need to disable extraneous microphones. See Enable Microphone
          • Make sure the recording was made using a recording device that is approved by nVoq for use with the voice client. You can see which microphone was used for the dictation on the Review & Correct Dictation Event Timeline. Using a device that has NOT been approved for use with the voice client may result in poor audio quality, which results in poor recognition accuracy. See the List of Recommended Microphones.
        • If the user is dictating with nVoq Mobile Voice:
          • The user could be covering the microphone with their hand. The microphone usually looks like a pinhole, and depending on the device it could be on the side, bottom, or top of the device. While dictating, the user should make sure that their hand is not covering or brushing up against the microphone pinhole.
          • The user could be in an area with poor connectivity. If they are in a location with a weak WiFi or cellular data connection, a notification should appear at the top of their screen to let them know (unless they have this setting disabled). Dictating in an area with a weak connection can affect dictation recognition.
    • If the misrecognition is not caused by poor audio quality, go to Step 2.

  2. If an uncommon term is not being recognized, add it to Vocabulary. 
    • Add uncommon names of people, places, products, or even uncommon drug names that do not return at all to vocabulary on the Vocabulary page in nVoq AdministratorIf the name is recognized but is just needs to be formatted differently (like it needs capitalization), add a substitution instead.
      While nVoq makes every effort to protect PHI, we recommend that users of any speech recognition technology follow the HIPAA minimum necessary rule, which states, “Protected health information (PHI) should not be used or disclosed when it is not necessary to satisfy a particular purpose or carry out a function”.
      Not dictating PHI will not only meet these federal requirements under HIPAA, but will also reduce concerns regarding poor accuracy that result from unique name spellings. Since most clinicians dictate into an EHR, they have already identified the patient and do not need to dictate PHI.
      • Common medical terms that are not returning correctly should be reported to nVoq Support.
      • Be sure to TEST IN DICTATION before adding to vocabulary to make sure the word or phrase isn't already in the dictionary or topic. Don't add words that are already in the dictionary or topic.
      • If this is a term that only one person will be using, add it to their account vocabulary.
      • If it's a term that will be dictated by others, it's fine to add the term to organization vocabulary for the organizations that are likely to dictate it.
    • If the misrecognition is not caused by an uncommon term, go to Step 3.

  3. If the misrecognition seems odd or misspelled, check existing vocabulary, substitutions, sentence modeling, and the dictation topic.
    • Use the Word Finder in nVoq Administrator to perform an overall search for a misspelled word.
      • If the misspelled word appears in the group topic, submit a ticket to nVoq Support.
      • If the misspelled word appears in a sentence model, download the sentence model file, delete the sentence model, remove the word from the file, and upload the corrected sentence model to the appropriate account or organization. Keep in mind however that UNKNOWN WORDS found in a sentence model will NOT appear in dictations UNLESS the word has been added as a vocabulary term. 
      • If the misspelled word appears a substitution, edit the substitution.
      • If the misspelled word appears account or organization vocabulary, edit the vocabulary term
    • Check for RegEx substitutions for any that could be causing the issue.
      • If the account or organization has RegEx or RegEx Plus substitutions, review them to make sure they are not the cause of the misrecognition.
      • Delete substitutions that may be causing the misrecognition.
    • If the misrecognition is not found in a vocabulary term, substitution, sentence model or the topic, go to Step 4.

  4. Correct the dictation transcript to match what the user said.
    • If the recording has good audio quality, correct dictation transcripts on the Detail Review and Correct page. A nightly job will collect the corrections you make and apply it to the user’s profile overnight as sentence modeling.
      • The sentence model tells the dictation engine to expect these words in this order. There is no need to send issues like this to nVoq Support.
      • Only correct transcripts that have good audio quality. This should be performed by someone familiar with clinical terminology.
      • Before you correct a dictation, make sure that any uncommon words that you correct are already in Vocabulary (Step 1). Unknown words in a sentence model that are not in vocabulary or the dictation topic will be skipped. If the word is uncommon, be sure to check both the user's Account Vocabulary and Organization Vocabulary for the term.
      • Corrections made in Review & Correct create a sentence model that is applied to the user's profile over night, so be sure to test the issue the day after correcting problem dictations.
    • If the misrecognition still occurs, go to Step 5.  

  5. Add an Account Substitution for common words or phrases that are returning incorrectly if they consistently return the same (incorrect) way.  
    • If the misrecognition of words occurs the same one or two ways every time it is dictated, add a substitution on the Substitutions page in nVoq.Administraor. If the word or phrase is returning two different ways, it's OK to add two different substitutions to cover both incorrect returns,  but DO NOT substitute something the user might actually dictate. 
      • Add substitutions at the account level to avoid negatively impacting other users.
      • Substitutions happen last in the dictation process, so they over-write what the dictation engine thinks it heard with what you configure as the substitution EVERY TIME. 
        • For example, if every time the user dictates, "lying supine" the transcript says, "lying spine," do NOT create a substitution to replace the word "spine" with "supine" because the user would never be able to get the word "spine" to return in a transcript ever again. Instead, make a substitution that replaces the whole phrase "lying spine" with "lying supine" since the user would not likely dictate "lying spine".
        • An example of something that you may not want to make as a substitution is if every time the user dictates, "I am" the transcript says "AMH," you would not want to make a substitution for "AMH" to be replaced with "I am" unless the user would NEVER need to dictate "AMH".
    • If the word or phrase cannot be added as a substitution or if it returns differently in the transcript each time it is dictated, go to Step 6.

  6. If a word or phrases cannot be added as a substitution because it returns differently each time, add it as Account Vocabulary
    • Add the word or phrase first without any phonetic representation ("Sounds Like") on the Vocabulary page in nVoq Administrator.
      • Small connecting words (like "a," "then," "the") should NEVER be added as vocabulary words. If these are a problem, address this using Sentence Modeling instead.
      • Adding common words or phrases to Vocabulary may make them more likely to return even when they are NOT dictated.
    • If the misrecognition still occurs, go to Step 7.

  7. Add phonetic support (Sounds Like) to the term in the Account Vocabulary. 
    • Add Sounds Like to the word or phrase on the Edit Vocabulary page in nVoq Administrator
      • This can be especially helpful if the person pronounces the word or phrase in a unique way. 
      • Keep one copy of the word without a Sounds Like and create a second copy of the term as another Account Vocabulary term and add the Sounds Like to that one. 
    • If the misrecognition still occurs, go to Step 8.

  8. Add more sentence modeling. 
    • Upload a sentence modeling document to the user’s Account Sentence Modeling that includes 6-10 instances of the word or phrase in sentencesthat are similar to what will be dictated.
      • It's best to add sentence models at the account level to start with to prevent it from negatively impacting other users.
    • If the misrecognition still occurs, go to Step 9.

  9. Contact nVoq Support.
    • You can use the email icon on the Detail Review and Correct page to send dictation transaction information to Tier 2 nVoq Support.
      • Include in the email all of the steps you have completed so far to resolve the misrecognition, and whether or not the misrecognition is affecting others or just one user.