Cohere Transcribe: Open-Source Speech Recognition | Cohere

Cohere has released Transcribe, an open-weights speech recognition model.

The 2B-parameter model was trained from scratch and is released under the Apache 2.0 license.

It supports 14 languages and, according to Cohere, leads the Hugging Face Open ASR leaderboard with a 5.42% average word error rate.

Enterprises can run it with full infrastructure control or use Cohere’s managed Model Vault offering.