Cohere Launches Transcribe Arabic for More Accurate Speech Recognition
Listen to this article
Read by Anchor
What happened: Cohere announced Cohere Transcribe Arabic on July 7, 2026, an open-source model that converts Arabic speech into text. The company says it is built on a 2B ASR model and supports most Arabic dialects, Arabic-English code-switching and specialist workplace vocabulary. Open weights are available through Hugging Face under the Apache 2.0 licence.
Who is affected: Contact centres, internal documentation teams, government services and sector-specific applications that need more accurate Arabic transcription which can be deployed inside the enterprise.
The lens: Digital sovereignty
What it means for the region: This is more than a language addition. It is an infrastructure layer for control over Arabic speech itself. When models are designed for colloquial speech, dialects and code-switching instead of forcing users into a flattened standard form, local audio data becomes less dependent on a single supplier and easier for institutions to operate on their own terms. This gives the move practical importance across the Gulf, Levant and North Africa. Contact centres, government services, internal documentation and sector-specific applications need Arabic transcription that does not compromise accuracy or impose an English default. Cohere says human reviewers preferred its model to Whisper in most tests, but the more important point for the region is that the Arabic speech layer itself can now be built on with greater flexibility.
The practical takeaway: If you have internal Arabic recordings or customer service cases, test Transcribe Arabic on a small sample before deciding to buy or integrate it, and review dialect and code-switching errors early.