How Many Clocks? Overcoming the Technical Challenges in Engineering a Real-Time AI-Enabled Live Subtitling Solution

Track: Technical | T6 |
Friday, June 7, 2024, 11:30am – 12:00pm
Held in: Room 6
Presenters:
Tijmen Brommet - CaptionHub 
Colin Willis - AWS
Host: Gary Lefman

Live subtitling solutions have historically faced significant drawbacks: latency, inaccuracies, ability to scale for large audiences. This results in reverting to human translators for live events at a significant cost, and is entirely prohibitive for other live events. In 2023 CaptionHub’s live subtitling product was released leveraging a variety of AWS technologies – using several components such as transcription, translation, encryption, streaming, clock orchestration, to name a few – working perfectly together in real-time. This panel delves into the engineering and machine learning layers and learnings required in delivering live subtitles in real-time.

Takeaways:

• Live subtitling using AI is incredibly hard and complex;
• Live subtitling is in its infancy in the context of LLM;
• Live multilingual subtitling engineering requires focused and dedicated partnership.