An ABA clinic introduced voice dictation for end-of-session notes. Technicians had been spending 20 to 25 minutes a day typing notes on tablets, often in their cars between home visits. Dictation cut that to about five.
Two weeks in, a supervisor reviewing notes stopped on one line: “Client engaged in self-injurious behavior and was prompted to use a break card.” She called the technician, confused, because nothing in the session data suggested it. The technician had actually said the client did not engage in self-injurious behavior. The transcription had dropped one word.
That note was headed for the clinical record and, eventually, an insurer’s review.
Advertisment
General dictation tools are tuned for everyday speech, where a missed word rarely changes much. Clinical notes are different. “Not,” “no,” “un-,” and numbers flip meaning entirely. “Prompted” and “unprompted” describe opposite levels of independence. “Fifteen minutes” and “fifty minutes” can change how a supervisor reads a behavior episode.
ABA notes also carry heavy jargon. Manding, echoic, tacting, functional communication training, differential reinforcement. Children’s names, many of them unusual, appear in nearly every sentence. A model that handles a business meeting well may stumble badly on a technician’s notes from a living room.
Vendors love to claim they offer the most accurate speech-to-text model, usually based on public benchmarks of clean, everyday speech. Those results say little about how a model will handle “mand training with an AAC device” spoken quickly in a parked car with the air conditioning running.
Test on your own material. With consent and appropriate privacy protections, collect a hundred real dictated notes, transcribe them correctly by hand, and run each candidate model against them. Then score more than the overall error rate. Count errors on negations, numbers, clinical terms, and client names separately. A model with a slightly higher overall error rate but near-perfect handling of negations is the safer choice.
Most services let you add a custom vocabulary list. Load your clinic’s terminology and current client first names, and retest. The most accurate speech-to-text model for your clinic is the one that performs best on that test, whatever the marketing says.
No model will be perfect, so the screen where technicians review transcripts decides whether errors reach the record.
Show the transcript immediately after dictation ends, while the session is still fresh in the technician’s memory. Highlight words the model was unsure about. Put negations and numbers in a slightly bolder color, or underline them, so they catch the eye during a quick scan. Offer a play button beside each sentence that replays the original audio for that line.
Structure helps too. The clinic’s redesigned flow extracts key items from the dictation, such as target behaviors, durations, and prompting levels, into separate fields the technician confirms one by one. Free narrative stays, but the fields that drive clinical decisions get a second look.
Never let a note submit automatically. The technician reads, corrects, and signs.
Dictation in a car should start hands-free. Review should happen only once parked. The app can simply hold the review screen until the vehicle stops moving.
Advertisment
The narrative note and the trial-by-trial data describe the same session, and they should agree.
Good ABA data collection software can attach the dictated note directly to the session record and compare them. If a note says the client met the target on 80 percent of trials but the recorded data shows 55, the system flags it before signing. The same goes for behaviors: a note mentioning an incident that doesn’t appear in the behavior data, or the reverse, should prompt a check.
Supervisors reviewing records in their ABA data collection software then see the note and the data side by side, with any mismatches marked. The self-injurious behavior error would have been flagged in seconds.
Dictating inside a family’s home risks capturing siblings, parents, or a television in the background. The clinic’s policy is to dictate only after leaving. Audio should be deleted once the transcript is confirmed, and every speech service involved needs a signed business associate agreement.
Advertisment
The clinic kept dictation, along with its time savings. Technicians now review every note against flagged negations and session data before signing. The supervisor who found that first error still spot-checks a few notes a week, and so far, she hasn’t found another like it.
Advertisment
Pin it for later!
If you found this post useful you might like to read these post about Graphic Design Inspiration.
Advertisment
If you like this post share it on your social media!
You Might Be Interested On These Articles
Advertisment
Latest Post
Want to make your Business Grow with Creative design?
Advertisment
Advertisment
Best Resources For Graphic Designers