Arabic that survives
the phone line.
Most speech models hear Gulf dialect and write Modern Standard Arabic — or mangle it entirely once the audio narrows to a phone call. Ours keeps what was actually said.
$5 of credit to start, about 33 hours of transcription. No card.
Highlighted words are dialect the model kept rather than flattening to Modern Standard Arabic. Audio: omnilingual ASR corpus, CC BY 4.0.
We tested it on a real call, against the incumbent.
Character error rate against the leading Arabic vendor, on a genuine narrow-band call-in.
Broadcast, spontaneous interview and telephony, all against human transcripts.
A one-hour recording transcribes in under three minutes.
No SDK, no console, no sales call.
Post a file, get text back with exactly what it cost. Long recordings are chunked for you. Failed requests are never billed.
curl -X POST https://nutq.dev/v1/transcribe \ -H "Authorization: Bearer $NUTQ_API_KEY" \ -F "file=@call.mp3"
{
"text": "الاجتماع يبدأ الساعة العاشرة",
"usage": { "audio_seconds": 13.2, "cost_usd": 0.00056 }
}Or run it inside your own network, with no per-minute licence.
We deploy the same models inside your network, on your hardware, where the recordings already live. Nothing is sent to us and nothing is metered. For banks and government work in the UAE, that is usually the whole decision.
Talk to us about deployment- Runs on
- Your hardware, your network
- Egress
- None — audio never leaves
- Per-minute fee
- None
- Hardware
- One 24 GB GPU