Using Dragon for Speech-to-Text Transcription in Support of Human-Autonomy Teaming Research
Abstract
Dragon speech-to-text is an off-the-shelf transcription software developed by Nuance. Dragon version 15 was used to transcribe four prerecorded audio files of voice communication between two team members during a laboratory study to determine if this software is a feasible alternative to manual transcription to support human-autonomy teaming (HAT) research. In the present work, Dragon was able to transcribe the four audio files with 90 or better accuracy, without prior subject-voice training and using out of the box settings. Errors consisted of both incorrect word translations (e.g., daring translated as during, teal translated as steel or chill) and words that were missed by Dragon (i.e., words spoken but not picked up by Dragon). Missed words were most often conjunctions or connecting words (e.g., be). However, these errors did not seem to change the overall content of the communication. Transcription time with Dragon was, as expected, significantly reduced compared to manual transcription. For the longest audio file (15 min in duration) Dragon transcribed the file in 1 min, whereas manual transcription took 32 min. Overall, these results suggest that Dragon provides an efficient method for transcribing audio files and is a suitable platform to support HAT research. Efforts are currently underway to conduct real-time transcription with Dragon, thereby saving additional time and effort in the transcription process.
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 12, 2019
- Accession Number
- AD1084561
Entities
People
- Andrea Krausman
- Kristin E. Schaefer
- Sean Fitzhugh
- Sean Mcghee
- Troy Kelley
Organizations
- United States Army Research Laboratory