Skip to content

Voicemail as an inbound channel

Faxart treats a voicemail as an inbound item whose text happens to come from speech instead of OCR. Once you have the text, everything a fax can do, a voicemail can do: match a rule, wait in the review queue, earn auto-mode, file to a chart. This page explains the design.

The insight is that classification, routing, review, auto-mode, and audit do not care whether the text came from a scanned page or a spoken message. So the inbound logic is generalized to span both rather than forked. A fax is OCR then classify; a voicemail is transcribe then classify. From the classify step on, the path is the same.

flowchart LR
  FAX[Fax] --> OCR[OCR] --> CL[Classify]
  VM[Voicemail] --> STT[Transcribe] --> CL
  CL --> RULES[Rules] --> RQ[Review queue]
  RQ --> AUTO[Auto-mode eligible?]
  RQ --> FILE[Human-confirmed filing]

Voicemails live in their own table rather than being crammed into the fax table. That keeps each record honest (a voicemail has a duration and a transcript, not pages and a PDF). The cost of a separate table is that shared logic has to span both, and the load-bearing example is auto-mode trust: a rule’s eligibility is scored over a union of its fax, voicemail, and dictation decisions, so a shared rule keeps one trust score across every channel instead of fragmenting per channel.

That union is the whole point. If each channel’s trust were tracked separately, a rule that is clearly trustworthy from its combined history might never graduate on any one channel alone.

Transcription runs on your own hardware. Audio never leaves the building, the same rule that governs page images. Reading or streaming a stored recording is a PHI access and is audited like viewing a fax.

Auto-mode can route a voicemail without a person. It can never file one to a chart without a person confirming the patient, because filing needs an independent identity cross-check that a routing decision does not provide. A voicemail files to MEDITECH the same human-confirmed way a fax does: the transcript is rendered to a one-page “Telephone Message” document and rides the identical HL7 path.