United StatesMedia and entertainment2025 – ongoing

Pacers Sports & Entertainment

Live arena captioning with a custom speech model

Indiana Pacers' live captions reach a 1.14% error rate with a fine-tuned Azure speech model

WorkingVendor's case studyVendor ↗

Problem

The team wanted live captions for deaf, hard-of-hearing and non-English-speaking fans. Standard speech tools struggled with fast announcers, game terms, player names and arena noise.

Approach

The team fine-tuned a custom model in Azure AI Speech using hundreds of hours of Pacers and Fever broadcasts. Captions reach arena screens and the mobile app through Azure Functions and SignalR. English launched first, Spanish followed, and 12 more languages are planned.

Results

1.14%Caption error rate, against an original goal of 10%Vendor's case study[1]

87%Reduction in speech-to-text error rateVendor's case study[1]

Takeaway

Caption error rate was 1.14%, against an original goal of 10%. Audio preparation dropped from up to five days to about two minutes.

Sources

Indiana Pacers use Azure AI Foundry to create first live arena captioning service
Microsoft19 May 2025Vendor publication

Source summaries

[1]Microsoft19 May 2025

Microsoft describes how the Pacers' parent company trained a speech model on its own broadcasts for live captions. It reports a 1.14% error rate and far shorter training prep.

Full details are in the original sources.

Added 25 Oct 2026 · Updated 25 Oct 2026