Topic Update: 02-27-2023
  • 1 Minute to read
  • Dark
    Light

Topic Update: 02-27-2023

  • Dark
    Light

Article Summary

Accuracy - General

  • Acoustic data used to train all Topics has been improved -  Audio from a variety of speakers including some from each gender and sources in addition to TEDLIUM have been added. (B-17014)
  • New drugs/devices- Corpus and dictionary support for new drugs and medical devices (from FDA) for December 2022 and January 2023 has been added. (B-17003, B-17042)
  • Corrections added to All Topics - Transcripts that were corrected during May-September 2022, December 2022 and January 2023 have been added to their corresponding Topic corpus . New words found in these corrections have been added to the dictionary. (B-16984, B-17000, B-16969,B-17043) 
  • Dictionary Cleanup- Cleanup work of the dictionary continues.  This effort involves removing nonsensical words as well as improving phonetic representations of existing words. (B-17086)
  • Support for a "range" of dimensions has been added-
    A dimension is a two or three digit number that is separated by the word "by" for example "1.56 by 0.7 by 0.5" or "2.1 by 0.91" which return as  "1.56 x 0.7 x 0.5" or "2.1 x 0.91"
    Now a range of dimensions separated by the word "to" or "dash" will return properly. For example "1.5 by 0.7 by 0.5 cm to 2.5 by 2.1 by 0.9 cm" will return as "1.5 x 0.7 x 0.5 cm-2.5 x 2.1 x 0.9 cm" (D-13142)
  • Incorrect numbers- Fixed an issue where "a" was returning for "8" in numbers. For example "2 point a" was returning instead of "2.8" (D-13142)

    

Accuracy - Radiology

  • Additional Corpus- Added additional corpus to the Radiology topic. All unknown words in the new material were added to the dictionary. (B-17084)