White noise may improve accuracy of SR system

CHICAGO - A small level of ambient white noise in reading room environments may improve speech recognition (SR) accuracy, according to a presentation at the 2007 RSNA meeting.

In an experiment conducted at the University of Maryland Medical Center in Baltimore, white nose was added at four different levels to recordings of 20 random reports dictated by 10 different radiologists in a noise-free environment. The reports were representative of those dictated for CT, MR, and interventional radiology, according to presenter Dr. Jonah Zwemer.

The reports were processed by a Nuance Dragon NaturallySpeaking 8 speech recognition system. The noise-free recordings were also processed and all were analyzed for errors.

The baseline error rate for recordings made in a noise-free environment averaged 11.6%, with a range of 6.5% to 26.1%. At 41.6 dBA ambient noise level, the error rate dropped to an average of 10.3%. The three higher noise levels generated errors of 12.3% (at 49.3 dBA), 13.0% (at 51.4 dBA), and 13.5% (at 53.9 dBA).

"We think that the results of this initial evaluation merit investigation into evaluating the impact of other low-level background noise, such as that generated by people talking in the background and doors opening and shutting," Zwemer said.

The University of Maryland conducted this experiment as a part of assessing the environments of reading rooms, and evaluating or incorporating new technologies for use in designing an optimum "Reading Room of the Future," a long-standing project headed by Dr. Eliot Siegel, chairman of the Baltimore Veterans Affairs Medical Center and member of the University of Maryland Medical Center faculty.

By Cynthia Keen
AuntMinnie.com contributing writer
November 27, 2007

Related Reading

Implementing SR before a hospital mandates gives radiology practice upper hand, November 27, 2007

Keys to a successful voice recognition implementation, July 12, 2007

ARRS study: Gender differences affect speech recognition accuracy, May 8, 2007

Speech recognition cuts report turnaround times, May 4, 2007

RIS-SR-PACS combo speeds radiology reporting, April 17, 2007

Copyright © 2007 AuntMinnie.com

Page 1 of 603
Next Page