Cohere releases open-source voice model for transcription

Cohere has launched Transcribe, an open-source automatic speech recognition model designed for enterprise use and self-hosting on consumer GPUs. This model, with 2 billion parameters, supports 14 languages and reportedly outperforms leading competitors like Zoom Scribe and IBM Granite on the Hugging Face Open ASR leaderboard, achieving a lower word error rate. Human evaluations also indicated superior accuracy and coherence compared to other models, though it faced challenges with Portuguese, German, and Spanish. Transcribe boasts high processing speeds and will be integrated into Cohere's North platform and offered via API for free, with availability on Model Valut.