Wayne Kaniewski, MD

This article is intended to refresh some key concepts since you completed your Dragon Medical Training.

For clinicians who were accustomed to dictating to a human transcriptionist in the past, you discovered quickly that you cannot dictate at “warp speed” when using medical speech recognition technologies (MSRT).

Q:  Must I dictate very slowly when using Dragon voice recognition software?

A:  No.  But you must e-nun-see-ate more than you may be used to.  Dragon™ Medical “listens” for patterns of sound (phonemes).  So, you must pronounce some words more distinctly that you may be used to.  When you speak to another person, they can usually “fill in the blanks” if you are not speaking clearly; with Dragon, enunciation is more important.  Examples:

  • Sound-alike words
    • he vs. she
    • an vs. and
    • is vs. it
  • Terminal syllables
    • discuss vs. discussed
    • depress vs. depressed
    • patient vs patients

Generally, you are able to dictate as fast as you are able to enunciate clearly.  When you start “tripping over your tongue”, Dragon Medical will start making misrecognitions.

The more “sound data” that you provide, the more that Dragon will be able to decide, using complicated mathematical algorithms, between several words that sound alike.  These algorithms also look for word pairs that commonly occur together, and these pairs may vary greatly depending on the vocabulary (e.g. casual speech vs. medical terms) and the user.

For instance, if you dictate “an EKG” and the software repeatedly transcribes “in EKG”, it won’t accomplish much to correct or train “an" vs. "in” since these words each consist of very little sound data to allow distinguishing between them.  You will be more successful by training the phrase “an EKG”.

Just as Dragon will have difficulty distinguishing between monosyllabic words, it also has no way of knowing what homonym forms you intend if those words are used in isolation.  For instance, how can you distinguish between these words, when they are all pronounced the same?

  • their, there, they’re
  • two, too, to, 2
  • past, passed

The solution is to dictate in complete sentences, with the period (or ! or ?)  The software engine “looks” at the context of a sentence, and is thus able to be more accurate in determining the correct word form to use.  The speech engine for medical specialties actually looks at the medical context of a sentence, and that is one reason why the medical versions of speech recognition software are going to work better for medical documentation than available non-medical products.  It’s not just the difference in the software vocabularies!

We’re not used to dictating in complete sentences.  It’s a bit easier if you first think about what you are going to say in the next sentence, then dictate it.

When a misrecognition occurs, try to avoid the habit of correcting it using the mouse & keyboard.  Instead, follow these steps:

  • Use the “Select” voice command to highlight the misrecognized word(s)
  • Speak the correct word(s), or "Choose" from the Corrections Dialog Menu
  • Use the"Train That" command

The software will “learn” and be less likely to make that misrecognition in the future.  It’s a difficult habit to break, but it will pay off in the long run!  This process, once mastered, will be faster than always reaching for your mouse & keyboard, and will also decrease the chances of repetitive stress injuries.   Also, selecting then correcting 2-3 words together may be more effective than just one word, particularly if that one word is a monosyllable.

Despite your best efforts, you will sometimes have unexpected, nonsensical, and even hilarious things transcribed into your notes.  So, the final and most important point is to always read your notes before you sign them.  Putting a disclaimer in your electronic notes that “Speech recognition software was used to create this note.” will not protect you in a court of law.  Not reading your notes can also, more importantly, potentially impact patient care.  How about this one?

Dictated:  “Questions were invited and addressed.”

Transcribed as:  “Russians were invited and undressed.”

In summary:

  • Enunciate
  • Dictate smoothly at an even, moderate pace
  • Speak in complete sentences with punctuation
  • Make corrections by voice, not keyboard