Cohere Unveils Transcribe: An Open Source Voice Model Tailored for Enterprise Transcription

Cohere, a leading enterprise AI company, has launched Transcribe, its first open source voice model designed for automatic speech recognition and transcription tasks. With 2 billion parameters, Transcribe is optimized for consumer-grade GPUs, enabling enterprises to self-host the model efficiently. It supports 14 languages, including widely spoken ones like English, Spanish, and Chinese, making it suitable for global applications.

On performance benchmarks, Transcribe achieved a word error rate of 5.42 on the Hugging Face Open ASR leaderboard, outperforming several established models. Human evaluations also favored Transcribe’s transcriptions for accuracy and coherence, with an average win rate of 61% compared to competitors. However, it showed some challenges with Portuguese, German, and Spanish transcriptions.

The model processes audio at a rate of 525 minutes per minute, which is notably fast for its class. Cohere plans to integrate Transcribe into its enterprise orchestration platform, North, and is offering the model freely through its API and managed inference platform, Model Vault. This enables enterprises to adopt advanced speech recognition technology with ease.

The launch comes amid growing demand for AI-powered transcription and dictation tools in the enterprise sector. Cohere’s open source approach and multilingual support position Transcribe as a competitive solution in this expanding market.

With strong revenue projections and potential plans for an IPO, Cohere continues to strengthen its footprint in enterprise AI innovation. Transcribe represents a significant step forward in making high-quality, scalable speech recognition accessible to businesses worldwide.

Explore Cohere’s Transcribe to enhance your enterprise transcription capabilities today.