Introducing Cohere Transcribe: a new state-of-the-art in open-source speech recognition
Summary
Cohere Transcribe is an open-source, 2B-parameter speech recognition model that emphasizes production readiness, low word error rate, and enterprise workflows. It claims top accuracy on HuggingFace Open ASR Leaderboard (WER 5.42%), supports 14 languages, and offers deployment options including edge runtimes and Cohere Model Vault for private inference. The article outlines architecture, performance, and integration paths with Cohere North for enterprise speech intelligence.