Cohere's ASR model challenges speech API dominance

Cohere has released Transcribe, an open-weight automatic speech recognition (ASR) model designed for enterprise use, boasting a 5.42% word error rate. This performance aims to rival and surpass existing closed-source APIs and open models, offering improved accuracy and data control. Unlike research-focused models, Transcribe is immediately available for commercial applications and can be deployed on-premises, addressing data residency and latency concerns for businesses. Its architecture allows for efficient inference on local GPUs, making it a viable alternative for production pipelines, voice-enabled workflows, and audio search applications. The model has demonstrated superior performance against competitors like OpenAI's Whisper and ElevenLabs on various benchmarks.