the problem with cloud transcription for clinical audio
clinical audio — dictated chart notes, recorded patient consultations, peer-review case discussions — is PHI under HIPAA. uploading it to a third-party transcription service requires a business-associate agreement and shifts a portion of HIPAA liability to the vendor; the practice's compliance posture still extends to the vendor's data handling, breach response, retention, and access controls. it's manageable, and many practices manage it. it's also the part that breaks down at scale, during audits, and after vendor policy changes.
the practical answer most physicians land on is that they don't transcribe clinical audio. they dictate into the EHR's voice-to-text (which uses cloud-based STT under the EHR's BAA), or they hire a scribe, or they write notes by hand from memory. none of those is what the audio could actually be — a structured, time-stamped record of the encounter that's referenceable later.
what changes with on-device transcription
when the speech-recognition model runs in your browser, the PHI never leaves the device that recorded it. there is no third party. there is no log, no cache, no access control that wasn't already in your practice's possession. HIPAA's BAA requirement applies to business associates; an in-browser tool that you run yourself isn't one.
this isn't a workaround. it's the structurally correct answer for a workflow where the audio is sensitive, the practice is small, and the EHR's built-in dictation isn't what the workflow needs. (we are not your compliance counsel; this is not legal advice. ask your compliance officer about the parts that aren't clear.)
workflow
- record the dictation or consultation. phone, laptop, dedicated dictation recorder, encrypted field recorder. nothing sent.
- open audiohighlight in private mode. drop the file in. transcription runs locally with no network request — chrome, edge, or arc on a current machine handles a 30-minute consultation in real-time.
- review and edit. per-account custom vocabulary catches medical terms, medications, abbreviations the model would otherwise mishear. fix proper nouns once; they propagate. the built-in medical-term dictionary covers ICD-10, CPT, common medications, and standard clinical jargon.
- export to your EHR or notes system. .docx for direct paste, plain text for any EHR's free-text field, json for systems that ingest structured input. the audio file and transcript both stay on your machine; we don't store either.
- close the tab. the in-browser session is gone. no cache, no recoverable history.
where this fits and where it doesn't
- fits: solo or small-practice dictation. physicians dictating chart notes, consultation summaries, referral letters, and clinical correspondence on their own machines, where the audio never needs to be shared or collaborated on.
- fits: medical scribes. scribes processing recorded encounters from their practitioner's recorded audio, on the scribe's machine, without sending the audio to a third-party processor.
- fits: peer-review and case-conference recordings where the practice already records but doesn't transcribe because of vendor handling concerns.
- doesn't fit: real-time AI scribing during the encounter. for live ambient-listening tools that produce notes during the visit, you want a different category of product. ours transcribes recordings after the fact.
- doesn't fit: large hospital systems with integrated EHR voice-to-text. if your EHR (epic, cerner, athena) provides built-in HIPAA-covered voice transcription that fits your workflow, keep using it. our wedge is for the cases that don't fit the EHR's built-in option.
what we don't promise
medical-grade accuracy from any AI transcription tool, ours included. clinical transcription requires careful review for medication names, dosages, and specialized terminology. the editor is built for that review — every word is an audio anchor; click it to hear the exact second.
we also don't replace the parts of HIPAA that exist beyond data-flow: physical safeguards on the device that holds the audio, access controls if the device is shared, retention policies that fit your state and specialty, the practice-management policies you maintain. on-device transcription removes one specific component (the third-party processor). the rest is yours.
BAA for practices that need cloud mode
for practices where on-device isn't workable (older hardware, shared workstations, multi-clinician collaboration on the same recording), cloud mode under a signed BAA is also available. write hello@audiohighlight.com and we'll send the BAA along with the technical safeguards documentation. the price is the same as private mode — $0.25/minute, no subscription.
pricing
$0.25 per minute. a 30-minute consultation is $6. an hour-long peer-review case discussion is $15. private mode and cloud mode are the same price. no subscription, no minimum. for clinics with steady weekly volume, group pricing arrives after launch.