We are thrilled to announce the launch of
Cohere Transcribe, our first speech-to-text model. As speech recognition becomes increasingly important for AI-powered workflows, we wanted to push the boundaries of automatic speech recognition in real-world scenarios. Our new model was built from scratch with an emphasis on model accuracy for practical enterprise use: from meeting intelligence to speech-powered automations.
Its key benefits include:
- High degree of accuracy: Today, the model outperforms all open and closed-source ASR models on HuggingFace with an average word error rate of just 5.42%.
- Real-world performance: It has been built around challenging environments, such as multi-speaker meetings, boardroom acoustics, and diverse accents.
- Enterprise-ready, flexible deployment: It is open-source, allowing for full infrastructure control, with a GPU footprint optimized for local deployment or even edge environments.
If you're interested in learning more about Cohere Transcribe, including detailed model performance benchmarks and usage documentation, please
read our new blog post
.
Comments
Post a Comment