Cohere has released Cohere Transcribe Arabic, an open-source model designed specifically for Arabic speech recognition. This 2-billion-parameter system claims to be the most accurate open-source Arabic speech-to-text tool currently available. It targets specific difficulties within the language, including dialect variety, bilingual Arabic-English conversations, code-switching, and specialised vocabulary. The developers state the model outperforms Whisper Large V3, the standard Cohere Transcribe model, and other systems on benchmarks.
The release addresses a gap where general models often fail on regional dialects or mixed-language interactions common in the Middle East and North Africa. By offering an Apache 2.0 licensed model available on Hugging Face and via the Cohere API, the tool allows developers to process local audio without relying on proprietary black boxes. This specificity matters for applications requiring high fidelity in diverse linguistic environments.
* Available under the Apache 2.0 license
* Accessible via Hugging Face and the Cohere API
* Benchmarks available on the Cohere blog
