Skip to content
Zenteck
Latest
Mercor Study: AI Outpaces Accountants in SpeedSuno Speech: how to generate voiceovers with AI musicMicrosoft MAI Models: Voice Agents with 100ms LatencyFlux 3 Image: Targeted Editing and 4K OutputChatGPT Virtual Try-On: What Changes for RetailGoogle Guided Vision: real-time AI fine print readingTavus Griffin: AI Video Avatars Fool 48% in TestsIdeogram 4.5: Localized Image Editing at 2K Resolution
Automation

Microsoft MAI Models: Voice Agents with 100ms Latency

Por Alex da Cruz1 min read0 comments
models-may-microsoft-voice-natural-latency-100ms
A

Alex da Cruz

Alex da Cruz is a full-stack developer based in São Paulo, Brazil. He works with React, TypeScript and automation, and uses AI daily to solve real problems in code and operations — not as a demo. He has run an e-commerce operation end to end, and now builds and maintains the automation pipeline behind this blog. He writes about what he actually tests.

Ver perfil →

According to Microsoft, the new MAI-Transcribe-2-Streaming model delivers partial results in just over 100 milliseconds across 60 languages. This speed allows voice agents to process user speech and respond while the user is still mid-sentence, removing the awkward pauses common in automated phone trees.

How much do the new Microsoft voice models cost?

Through the end of the year, transcription costs $0.54 per hour of audio under an introductory price. For text-to-speech, the MAI-Voice-2.1-Flash variant achieves a latency of 150 milliseconds at a rate of $15 per million characters, down from the standard $22 tier.

What changes for customer service automation?

Both voice models can clone a target voice using only a few seconds of reference audio, while integrated safeguards attempt to prevent unauthorized use. In tests cited by Microsoft, roughly half of 4,000 participants could not distinguish the generated voice from a real human speaker. Developers can access these models immediately through Microsoft Foundry, the MAI Playground, and OpenRouter for integration into production workflows.

Sources

  1. Microsoft AI releases new transcription and text-to-speech models for voice agents — the-decoder.com

Frequently asked questions

What is the latency of MAI-Transcribe-2-Streaming?
The model delivers its first partial results in just over 100 milliseconds, allowing voice agents to respond while users are still speaking.
How much does the new Microsoft transcription cost?
Through the end of the year, transcription costs $0.54 per hour of audio under the introductory pricing.
Where can developers access MAI-Voice models?
The models are available through Microsoft Foundry, the MAI Playground, and OpenRouter.

Comments

0 comments

Be the first to comment.

RECOMENDADOS

Continue Lendo