Microsoft Ships First Streaming Transcription Model, Completing Voice Agent Pipeline on Foundry

Microsoft launched its first real-time streaming transcription model, MAI-Transcribe-2-Streaming, alongside two voice-generation models on October 1, 2026, enabling voice agents to complete conversational turns in under one second via its Azure AI Speech platform. The streaming model supports 60 languages with automatic detection and drastically reduces end-of-speech latency to 130 milliseconds, outperforming competitors like Grok Voice Transcribe 2.0, but still awaits independent validation on accuracy benchmarks.

This Wire brief sits within Fusion42's coverage of AI & ML.