Document Processing
Microsoft needed to deliver accurate live captions and real-time meeting transcripts in many languages for Teams while keeping inference latency low and compute costs under control at massive user scale.
Reported impact
Microsoft delivers cost-efficient, accurate live captioning and transcription across 28 languages for Teams, using Triton to improve throughput and reduce the computational resources needed for real-time speech services.
Microsoft built Teams live captioning and transcription on Azure speech infrastructure that combines Azure Cognitive Services, NVIDIA GPUs on Azure, and NVIDIA Triton Inference Server. Meeting audio is sent from the Teams client to speech models tuned to recognize names, jargon, and meeting context across 28 languages. Triton manages low-latency inference and keeps GPU resources saturated, while Azure packages the service into production-ready captioning and transcript features for massive hybrid-work traffic. The architecture lets Microsoft serve real-time subtitles and transcripts at global scale without overprovisioning compute, so accessibility features remain responsive for daily meetings while keeping throughput high and infrastructure use efficient.
Teams sends meeting audio into Azure Cognitive Services, speech models run on NVIDIA GPUs with Triton managing inference, and the resulting captions and transcripts are returned to meeting participants in real time.
Microsoft Teams is a collaboration app with nearly 250 million active monthly users.
This enables them to leverage advanced language models that recognize jargon, names, and other meeting context, to deliver highly accurate, personalized speech-to-text results-in real time-with very low latency.
This reduces the cost to customers by delivering higher throughput using fewer computational resources.
Using Triton Inference Server in Azure Cognitive Services seamlessly enables live transcription and captions with state of the art speech models in 28 languages.
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
NVIDIA Sample Sep 22, 2026 Sluicebox Sluicebox needed to turn fragmented bills of materials, supplier records, PDFs, CAD files, and emissions inventories into timely product carbon assessments, but manual life-cycle-analysis work was slow, expensive, and… 51-80% lower inference costs Lucy · NVIDIA Nemotron 3 Ultra · carbon graph NVIDIA Sample Sep 1, 2026 Onetera Onetera needed to help more than 90 municipalities turn fragmented regulations, zoning maps, permit rules, and legacy documents into fast resident-facing services without imposing heavy IT overhead or expensive manual… 1.7x increase in data throughput GPT OSS 20B · Government Intelligence Agent · LangGraph NVIDIA Sample Jun 26, 2026 MISA MISA needed to move from expensive, externally hosted LLM APIs to a self-hosted Vietnamese AI stack that could protect sensitive data, reduce latency and operating cost, and support many concurrent public-service and… $200,000 in monthly cost savings MISA AVA · NVIDIA A40 GPUs · NVIDIA DGX H200 