Blog Archive

Voice model

The Voice Model

Voice models process what the clinician says to produce a draft document. Here is how it

works:

  • The clinician dictates and the MT transcribes. AutoScript analyzes the dictation and compares it to the final written report.



  •  After approximately 1.5 hours of audio and 50 reports, AutoScript develops a voice model per clinician and work type.
  •  Based on each voice model, voice recognized drafts are produced.
  •  Voice recognized drafts are not perfect and always need some editing.

Voice models recognize sounds and strings them into words and phrases. Using complex formulas, they put these words into context and format the report based on the requirements of your institution. This is done by work type since each work type has its own terminology and possibly its own report format.

AutoScript does not produce a draft of a report if the quality is such that it would be faster to
transcribe the entire report. The goal is to make the job of the MT easier.

Examples of How the Voice Model Learns

  • You may work with a cardiologist who uses the phrase 'no rhythm' for 'normal rhythm'.

The MTs at your institution know this cardiologist and always change the phrase to 'normal rhythm'. AutoScript learns that 'no rhythm' should be recognized as 'normal rhythm' for that particular cardiologist. Soon the voice recognized drafts have the correct phrase, thereby eliminating the need to edit that phrase in the draft.

  •  Dr. Smith may use the word 'basically' repeatedly and inappropriately throughout his dictations. The MTs at your institution consistently omit 'basically' from Dr. Smith's reports. In time, AutoScript learns to eliminate the word 'basically' from Dr. Smith's reports even though he continues to use the word in his reports.

 Some clinicians may not be voice recognized.
Clinicians who are not organized speakers or who speak inconsistently may never be
recognized. As such, their reports will require standard transcription.
If a clinician is an organized speaker and has a thick accent or does not speak clearly, he

may be voice recognized.


Labels