ElevenLabs Scribe

Scribe is ElevenLabs' speech-to-text model, available through the same account as its well-known voices, with a real-time variant that responds in about 150 milliseconds. It supports Arabic as one language, names Gulf among the dialects it handles, and offers speaker diarization for up to 32 speakers plus word-level timestamps, which is useful for long, crowded recordings. ElevenLabs places Arabic in its "good" tier, between 10% and 20% word error rate, and its own table shows roughly 11% on a standard read-speech test. An independent test of dialect-heavy Arabic by a Gulf competitor put Scribe at about 40%, so the gap between clean and real audio is large here, as it is everywhere. Pricing as of October 2026 is about 22 cents per hour beyond plan allowances, with EU and Indian data residency on higher tiers but nothing in the Gulf.