CallScribe for WhatsApp Business

Where WhatsApp voice notes actually originate

In the GCC, WhatsApp is the default informal-business channel. A potential customer asks a sales question via voice note; a property tenant complains via voice note to a real-estate agency; a clinic patient describes symptoms in dialect. The audio is short (15 to 90 seconds typically), conversational, and almost always dialectal. None of the voice notes get transcribed by default — they sit in WhatsApp threads that one person can read, slowing every downstream operational process.

How the integration works (Cloud API vs. On-Premises API)

Meta provides two flavours of the WhatsApp Business API: Cloud API (Meta-hosted, the standard for new deployments) and the older On-Premises Business API (deprecated, sunsetting in 2025). CallScribe integrates with the Cloud API. You configure your Business Phone Number's webhook to point at CallScribe's endpoint. Inbound messages — including audio media IDs — are delivered to us; we GET the audio file via the Media endpoint, transcribe with our Arabic dialect-aware pipeline, and post the transcript back into a CallScribe project tied to the Business Phone Number.

For organisations using a Business Solution Provider (BSP) — Twilio, MessageBird, 360dialog, Infobip, Vonage — CallScribe integrates via the BSP's webhook forwarding rather than direct WhatsApp Cloud API. The end result is the same.

Audio retention rules and the 30-day window

Meta retains media (audio voice notes included) for 30 days from message receipt. After 30 days, the media-ID returns 410 Gone. CallScribe fetches and transcribes within minutes of receipt, so the 30-day window matters only for retroactive bulk imports — if you want to import six months of WhatsApp voice notes that already happened, the audio is mostly gone and only metadata remains.

CallScribe stores the transcript per its own retention rule. By default we do not retain the original audio file; if you need original audio retention beyond Meta's 30 days, enable original-audio storage on the project.

PDPL, customer consent, and disclosure

Voice notes are personal data. Under UAE PDPL and KSA PDPL, lawful basis for processing must be established. The standard practice is to disclose recording/transcription in your business's WhatsApp opt-in flow and in the WhatsApp Business Profile. CallScribe does not handle disclosure for you — that is on the customer-facing business; we sit downstream as a data processor under your DPA.

Use cases that pay back fast

Property-management voice notes from tenants get parsed into a complaint ticket with category and severity. Sales-team voice-note enquiries get auto-extracted into a CRM lead with intent classification. Customer-support voice-note tickets get prioritised by sentiment severity. Each of these saves a person from listening to every voice note linearly. The volume varies wildly: small businesses handle 10-50 voice notes per day; large hotel groups or property platforms handle thousands.

Audio quality and Arabic accuracy on WhatsApp

WhatsApp voice notes are encoded in OGG/Opus at typically 16kHz, 32kbps. Audio quality is generally fine for ASR — better than telephony — but background noise (street, café, car) is more variable than recorded business calls. WER on clear single-speaker Khaleeji voice notes is typically 9-13%; noisy outdoor recordings can rise to 18-25% but the model degrades gracefully rather than producing nonsense.