Transcript de-identification is the removal of all 18 HIPAA Safe Harbor IDs (45 CFR §164.514(b)) from a recorded telemedicine call. anonym.plus handles it on your own device. The spoken text stays readable, yet it no longer names the patient.
When this applies
Spoken visits get typed into long transcripts full of names said aloud. Before you reuse one for review or model training, those spoken IDs must go.
How anonym.plus handles it
- Open the typed transcript in anonym.plus on your device.
- Local OCR reads pasted screenshots of the dialogue too.
- The tool marks names, dates, places, and contact details.
- Skim each flag and fix any clinical term caught wrongly.
- Swap each ID for a safe label, or black it out.
- Store the clean text. The source never leaves your machine.
What you need to provide
- The transcript (TXT, DOCX, PDF, or pasted dialogue).
- An operator: Replace, Redact, or Mask.
- Optional speaker map for [PATIENT] and [CLINICIAN] roles.
PHI entity types detected
| Category | anonym.plus entity type | Example |
|---|---|---|
| Names | PERSON | “Hi, this is Dana” → [PATIENT] |
| Dates | DATE_TIME | “since last Tuesday” → [DATE] |
| Contact | PHONE_NUMBER | callback 503-555-0182 → [PHONE] |
| Contact | EMAIL_ADDRESS | dana@example.com → [EMAIL] |
| Location | LOCATION | “I live in Reno” → [PLACE] |
| Record IDs | MEDICAL_RECORD_NUMBER | chart 77120 → [MRN] |
Compliance achieved
- Strips all 18 ID classes for HIPAA Safe Harbor (45 CFR §164.514(b)).
- Runs offline, so the tool itself needs no BAA.
- Working files are kept safe with AES-256-GCM.
- Covers GDPR Art. 9 health data for EU callers too.
Anonymize telemedicine transcripts offline — see plans & start free →
Limitations & cautions
Spoken words get typed loosely, so misheard names and odd spellings can slip past once in a while. Always check the flags before export. Auto-typed audio is the riskiest source, since the speech-to-text step adds its own errors.
Frequently asked questions
Are spoken names harder to catch than typed ones?
A bit. Speech-to-text can mangle a name, which then reads oddly in the text. The tool flags clear names well; review the rest, above all on auto-typed audio.
Does this need a BAA?
No. anonym.plus runs on your own device with no cloud step. No outside party touches the PHI, so the tool itself needs no BAA.
Can I keep speaker turns readable?
Yes. Replace puts a steady role label in each turn, so the back-and-forth still flows without naming a real person.