Cohere Transcribe
lastmod 2026-04-04created 2026-04-042 edits
SOTA open-source speech recognition model
About
- A state-of-the-art automatic speech recognition (ASR) model with 2B (2 billion) parameters, supporting 14 languages including English, Korean, and Chinese
- Uses a Conformer-based encoder-decoder architecture and is distributed under the Apache 2.0 License
- Outperforms major competing models such as Whisper Large v3 with an average word error rate (WER) of 5.42% in English, ranking 1st on the Hugging Face Open ASR Leaderboard
- Showed high accuracy and consistency in both real-world evaluations and human evaluations, and maintains stable performance in multilingual transcription
- Achieves low latency and high processing efficiency at the same time, making it suitable for real-time products and workflows
See also
- Korean: 자동 음성 인식
- Korean: 음성-텍스트 변환
- Conformer
- Apache 2.0 License
- Hugging Face
- Whisper