Frequently Asked Questions About

Getting Started with TA^Ì Powered by SpeechMagic

How do we know which physicians within our hospital will be good candidates for speech recognition?

Successful candidates follow these practices. They:

1)    Speak clearly  

2)  Speak with good volume  

3)  Use consistent terminology  

4)  Dictate headings and punctuation  

5)    Dictate in locations without background noise

What steps will ensure success with speech recognition?

1)       Perform Context Tuning*, which means adding site-specific phrases and local terminology to the Context*, or medical specialty vocabulary, before the go-live date.

2)       Complete Context Adaptation*.  This process updates TA^Ì Powered by SpeechMagic with each dictator’s unique language model.  Context adaptation is accomplished by sending corrected documents, after speech recognition, back through the engine for learning and also by sending  previous transcripts through the engine that were created by a transcriptionist.  Five to six transcripts created by a transcriptionist per author is helpful for adaptation.

3)       Standardize the text and dictated formats for dates and measurements.  ASR-delivered text will be consistent to the department’s agreed-upon formats.  The greater the dictator consistency within a department relating to dates and dosages, the greater the accuracy in the delivered text.  For example, all dictators would say “Feb.12, 2008” and the delivered text would be either ‘2-12-2008’ or ‘Feb. 12, 2008’ or ‘February 12, 2008’, whichever method is selected by the client. 

4)       Dictate in a quiet location, without excessive background noise, so that ASR does not attempt to translate background sounds as text. 

How quickly after dictation will the TA^Ì Powered by SpeechMagic delivered text be available? 

Text, returned from the TA^Ì Powered by SpeechMagic engines, will be available within a few minutes after dictation.  This holds true for onsite license or ASP arrangements. The ‘raw’ text will be allocated to Correctionists for review and editing by the TA^Ì system.

How does the system improve accuracy during the first few months?

After jobs are edited by a Correctionist, TA^Ì Powered by SpeechMagic workflow sends the edited version back through the speech recognition servers to continue ConText adaptation.

How does the system learn the individual differences of each dictator’s voice patterns? 

At first the system uses the Initial Language Model (ILM)* to convert voice to text.  As the system learns, a User Language Model (ULM)* develops for each author.  In time, the system reaches a point where 50% of the conversion will be performed using the ILM while the other 50% will be performed using the ULM. 

Does Arrendale provide any voice system training materials?

Yes, we provide sample texts to be used for context adaptation before go-live.  These are not required, but suggested. 

Can Physicians enter their own Normals (pre-identified text) with TA^Ì Powered by SpeechMagic? 

TA^Ì Powered by SpeechMagic allows a dictator to enter his/her own Normals* if desired. In this case, the TA^Ì system inserts the Normal, which has been defined in the TA^Ì system and is already being used by medical transcriptionists. Dictators should alert the system by speaking “Insert Normal Chest” instead of speaking “Normal Chest”.  This will notify the system not to interpret those words from the dictation wav file.

Does TA^Ì Powered by SpeechMagic integrate with Arrendale’s e-Sign^Ì? 

Yes, for sites that use both TA^Ì Powered by SpeechMagic and e-Sign^Ì, the physicians will notice a new queue when e-signing.  This queue, ‘Preliminary Speech Recognition,’ contains those jobs that have returned from ASR, but not yet been edited by a Correctionist.  Physicians may select a job from this queue to edit and sign if the job is needed for patient care, bypassing the Correctionist.  Converted text jobs appear in this queue five to ten minutes after dictation, under normal conditions. 

*Words with an asterisk are defined in the TA^Ì Powered by SpeechMagic Glossary