Cohere Launches Open-Source AI Transcription Model for Enterprise Use
Enterprise AI company Cohere has released an open-source version of Cohere Transcribe, a speech-to-text model designed to generate text from audio in real time. The model targets enterprise needs, addressing challenges such as processing massive volumes of unstructured audio data in noisy, multi-speaker environments with diverse accents. According to Cassie Cao, senior staff product manager at Cohere, the model was built from scratch with a focus on minimizing word error rate and maximizing speed, measured by high RTFx throughput. It currently leads Hugging Face’s leaderboard for speech recognition models based on latency, accuracy, and multilingual performance. Unlike model-agnostic platforms like Granola, Cohere develops its own underlying technology, though collaboration remains possible. The model utilizes a 2B parameter conformer-based encoder-decoder architecture. Future plans include integrating the transcription capabilities into North, Cohere’s flagship enterprise workplace AI agent platform. This release highlights the growing importance of voice as a primary interface for enterprise workflows, aiming to provide robust, production-ready solutions for real-world business applications.
Wire timeline
Cohere Launches Open-Source AI Transcription Model for Enterprise Use
Enterprise AI company Cohere has released an open-source version of Cohere Transcribe, a speech-to-text model designed to generate text from audio in real time. The model targets enterprise needs, addressing challenges such as processing massive volumes of unstructured audio data in noisy, multi-speaker environments with diverse accents. According to Cassie Cao, senior staff product manager at Cohere, the model was built from scratch with a focus on minimizing word error rate and maximizing speed, measured by high RTFx throughput. It currently leads Hugging Face’s leaderboard for speech recognition models based on latency, accuracy, and multilingual performance. Unlike model-agnostic platforms like Granola, Cohere develops its own underlying technology, though collaboration remains possible. The model utilizes a 2B parameter conformer-based encoder-decoder architecture. Future plans include integrating the transcription capabilities into North, Cohere’s flagship enterprise workplace AI agent platform. This release highlights the growing importance of voice as a primary interface for enterprise workflows, aiming to provide robust, production-ready solutions for real-world business applications.
BetaKit