Cohere Transcribe Arabic: Frontier Speech Recognition for Arabic Speakers | Cohere - Cohere

2026-07-07 · Source: cohere.com via Google News · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, long

Summary

Released on July 07, 2026, Cohere Transcribe Arabic is an open-source Automatic Speech Recognition (ASR) model designed for Arabic speakers, built upon Cohere's 2B frontier ASR technology. It is presented as the most accurate open-source Arabic speech-to-text model, specifically engineered to handle the language's dialectical richness, bilingual Arabic-English speech, code-switching, and domain-specific vocabulary prevalent in business and developer settings. The model achieves an average Word Error Rate (WER) of 25.87 on the Hugging Face Arabic ASR Leaderboard, surpassing Meta's OmniASR-LLM-7B by 2.45 points and OpenAI's Whisper Large V3 by 11 points. Human evaluations showed a 95.8% preference for Cohere Transcribe Arabic over Whisper. Available under the Apache 2.0 license, it offers high-throughput serving, optimized with vLLM to achieve an RTFx score of 525, and runs efficiently on consumer hardware.

Key takeaway

For AI Engineers and developers building Arabic speech recognition applications, you should evaluate Cohere Transcribe Arabic. This open-source model provides significantly higher accuracy for diverse Arabic dialects and code-switching, outperforming alternatives like Whisper. Its enterprise-ready throughput and efficient operation on consumer hardware make it a strong candidate for production deployments. Consider integrating its weights from Hugging Face or utilizing the Cohere API to enhance your Arabic ASR capabilities.

Key insights

Cohere Transcribe Arabic offers the most accurate open-source ASR for diverse Arabic speech, including dialects and code-switching.

Principles

• ASR must preserve dialectal nuance and code-switching.
• Open-source models can achieve frontier performance.
• Sovereign AI requires control over models and infrastructure.

Method

The model was developed by extensively training Cohere Transcribe on diverse Arabic data, encompassing dialects, professional language, code-switching, and varied acoustic conditions.

In practice

• Download model weights from Hugging Face.
• Utilize Cohere API for hosted access.
• Provision Model Vault for managed deployment.

Topics

• Automatic Speech Recognition
• Arabic Language Processing
• Open-source AI Models
• Dialectal Speech Recognition
• Code-switching
• Enterprise ASR

Code references

• vllm-project/vllm

Best for: NLP Engineer, Research Scientist, AI Engineer, Machine Learning Engineer, AI Scientist

Related on AIssential

• A Comparative Study of Pretrained Transformer Models for Quranic ASR: Speech Representations, Label Formats, and Dataset Composition — Fine-tuning pretrained Transformers with specific speech representations and label formats significantly improves Quranic ASR accuracy and efficiency.
• New Open Audio Models 🤗 | Recap with Jeff — New open audio models from Mistral and Cohear offer efficient text-to-speech and speech-to-text capabilities.
• Code Switching in Real-Time | Universal-Streaming Speech-to-Text — Assembly AI's universal model enables real-time, zero-latency code-switched speech-to-text across six languages.
• Software for Automatic Speech Recognition via Whisper models applied to Oral History interviews in the Portuguese language — Ethos AT provides private, accessible automatic speech transcription for Portuguese using Whisper models.
• Now in Foundry: Cohere Transcribe, Nanbeige 4.1-3B, and Octen Embedding — Specialized, smaller open-source models can outperform larger general-purpose models across various AI tasks.
• WAXAL: A large-scale open resource for African language speech technology — WAXAL provides a large, open-access dataset for 27 African languages, fostering inclusive speech technology development.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cohere.com via Google News.