MT Speech Recognition & Editing 3 — Questions and Answers
Question 1: An SR system transcribes 'the patient was given Lasix' but the dictation actually said 'the patient was given Lozol.' What is the most likely cause?
- Acoustic similarity between the two drug names
- The language model defaulting to a more common drug (Correct answer)
- A network latency error
- Incorrect microphone calibration
Correct answer: The language model defaulting to a more common drug
Language models favor statistically common words; Lasix is far more frequently dictated than Lozol, causing the model to substitute the more probable term.
Question 2: Which practice best prevents SR-introduced medication errors from reaching the final report?
- Relying solely on spell-check
- Comparing SR output against the original audio for all drug names (Correct answer)
- Using only abbreviations for medications
- Routing all reports through a pharmacist
Correct answer: Comparing SR output against the original audio for all drug names
Listening to the original audio while reviewing SR-generated drug names is the most direct method to catch substitution errors before finalization.
Question 3: What is 'normalization' in the context of medical SR output?
- Converting spoken numbers and abbreviations into standard written forms (Correct answer)
- Reducing audio file size before processing
- Standardizing physician voice profiles
- Removing duplicate sentences from output
Correct answer: Converting spoken numbers and abbreviations into standard written forms
Normalization converts spoken forms such as 'four milligrams' or 'BP' into written standard formats like '4 mg' or expanded abbreviations per style guidelines.
Question 4: A dictating physician says 'peri-operative' but SR outputs 'pre-operative.' The MT should:
- Accept 'pre-operative' as close enough
- Correct to 'perioperative' based on audio review (Correct answer)
- Flag the report and send to QA without correcting
- Query the physician via telephone
Correct answer: Correct to 'perioperative' based on audio review
Perioperative and preoperative have distinct clinical meanings; the MT must correct based on what was dictated, verified against the audio.
Question 5: Front-end SR differs from back-end SR primarily in that front-end SR:
- Requires manual transcription as a fallback
- Displays text to the dictating physician in real time (Correct answer)
- Processes audio in overnight batch runs
- Is only used for radiology and pathology
Correct answer: Displays text to the dictating physician in real time
Front-end SR presents the transcribed text immediately to the physician during dictation so they can make corrections before the report leaves their hands.
Question 6: When SR output shows blank placeholders (blanks) for unrecognized segments, the MT's responsibility is to:
- Leave the blanks for the physician to fill in later
- Listen to the audio for that segment and transcribe it manually (Correct answer)
- Insert 'inaudible' for every blank without review
- Delete the surrounding sentence
Correct answer: Listen to the audio for that segment and transcribe it manually
Blanks indicate SR failure; the MT must review the audio and manually transcribe what was said, or mark as inaudible only if truly unintelligible.
Question 7: Which metric is commonly used to measure SR system performance in medical transcription?
- Words per minute (WPM)
- Word error rate (WER) (Correct answer)
- Character recognition score (CRS)
- Transcription turnaround time (TAT) only
Correct answer: Word error rate (WER)
Word error rate (WER) quantifies the percentage of words that differ between SR output and the reference transcript, making it the standard SR accuracy metric.
MT Speech Recognition & Editing 3
An SR system transcribes 'the patient was given Lasix' but the dictation actually said 'the patient was given Lozol.' What is the most likely cause?