Introducing MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2 in Microsoft Foundry

2026-04-02 · Source: Microsoft Foundry Blog articles · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, medium

Summary

Microsoft Foundry has announced the public preview of three new Microsoft AI models: MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2. MAI-Transcribe-1 is a speech recognition model offering enterprise-grade accuracy across 25 languages at approximately 50% lower GPU cost than alternatives, ranking 1st on overall WER on the FLEURS benchmark for 11 core languages. MAI-Voice-1 is a high-fidelity speech generation model capable of producing 60 seconds of expressive audio in under one second on a single GPU. MAI-Image-2 is a text-to-image model that debuted at #3 on the Arena.ai leaderboard, excelling in photorealistic generation, in-image text rendering, and complex layouts. These models currently power Microsoft products like Copilot, Bing, and PowerPoint, and are now available to developers through Foundry and Azure Speech.

Key takeaway

For engineering leaders evaluating AI model integration, these new Microsoft AI models offer a compelling balance of performance and cost efficiency. MAI-Transcribe-1's lower GPU cost and MAI-Voice-1's rapid generation can significantly reduce operational expenses for voice-driven applications, while MAI-Image-2 provides advanced creative capabilities. Consider piloting these models in your next project to assess their impact on both performance benchmarks and infrastructure costs, especially for high-volume multimedia processing.

Key insights

Microsoft Foundry introduces three new AI models for efficient, high-quality speech recognition, speech generation, and text-to-image capabilities.

Principles

• Efficiency drives scalability and predictable enterprise pricing.
• First-party audio AI stacks enhance developer control.
• Collaboration improves creative workflow integration.

Method

Developers can access MAI-Transcribe-1 and MAI-Voice-1 via Azure Speech, with MAI-Transcribe-1 priced at $0.36 USD/hour and MAI-Voice-1 at $22 USD/1M characters. MAI-Image-2 is available via API, costing $5 USD/1M tokens for text input and $33 USD/1M tokens for image output.

In practice

• Use MAI-Transcribe-1 for real-time IVR transcription.
• Generate custom voices with MAI-Voice-1's Personal Voice feature.
• Integrate MAI-Image-2 for enterprise branding visuals.

Topics

• MAI-Transcribe-1
• MAI-Voice-1
• MAI-Image-2
• Microsoft Foundry
• Azure Speech

Best for: CTO, VP of Engineering/Data, Director of AI/ML, AI Engineer, Machine Learning Engineer, Software Engineer

Related on AIssential

• 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.
• Mistral drops new speech-to-text AI models — Mistral's new speech-to-text models prioritize speed, privacy, and affordability for diverse enterprise applications.
• New MAI models in Microsoft Foundry across text, image, voice, and speech — Microsoft's new MAI models in Foundry offer advanced, cost-efficient AI capabilities across text, image, and audio modalities.
• Building real-time voice assistants with Amazon Nova Sonic compared to cascading architectures — Amazon Nova Sonic unifies speech processing for real-time, human-like voice AI, simplifying architecture and reducing latency.
• 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.
• OpenBMB / VoxCPM — VoxCPM2's tokenizer-free, diffusion autoregressive architecture delivers highly natural, multilingual, and controllable TTS with advanced cloning.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Microsoft Foundry Blog articles.