Cohere launches high accuracy Arabic transcription model

Cohere, a Toronto-based AI company, has launched Transcribe Arabic, an open-source speech recognition model noted for its high accuracy. The 2-billion-parameter model achieved a word error rate of 25.87 on the Arabic ASR Leaderboard, outperforming models like OmniASR-LLM-7B and Whisper Large V3. This model's significance lies in the historical lag of Arabic AI compared to English, especially in speech recognition and transcription. Spoken by over 300 million people, Arabic has around 30 dialects, making voice recognition tools crucial for engaging audiences in their native dialects.