BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//cfp.devcon.org//devcon8//talk//HWKQHS
BEGIN:VTIMEZONE
TZID:IST
BEGIN:STANDARD
DTSTART:20000101T000000
RRULE:FREQ=YEARLY;BYMONTH=1
TZNAME:IST
TZOFFSETFROM:+0530
TZOFFSETTO:+0530
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
UID:pretalx-devcon8-HWKQHS@cfp.devcon.org
DTSTART;TZID=IST:20261105T100000
DTEND;TZID=IST:20261105T101500
DESCRIPTION:**When 2 Seconds Matter: Building Real-Time AI Captions for Liv
 e Events**\n\nLive events need captions and translations that feel instant
 aneous\, but building real-time AI streaming is very different from proces
 sing uploaded audio.\n\nIn this talk\, I’ll share what I learned buildin
 g **VoxBento**\, an open source real-time multilingual streaming infrastru
 cture using FastAPI\, WebRTC\, MediaMTX\, and pluggable transcription prov
 iders.\n\nI’ll walk through the architecture from live audio ingestion t
 o streaming transcription and caption delivery\, and the engineering chall
 enges we encountered along the way. In particular\, I’ll cover why we mo
 ved from HLS to WHEP to reduce playback handoff latency\, how we integrate
 d both cloud and local transcription services\, and why naïve chunk-based
  Whisper inference can cause context and caption consistency problems.\n\n
 I’ll also discuss practical approaches to handling overloaded transcript
 ion pipelines\, including bounded queues\, dropping stale audio\, voice ac
 tivity detection\, and caption aggregation.\n\nThe goal is to give attende
 es a practical look at the tradeoffs involved in building low-latency\, ex
 tensible AI infrastructure in open source\, along with lessons that can be
  applied to other real-time AI applications.
DTSTAMP:20261101T032009Z
LOCATION:Community Led Stage
SUMMARY:Voxbento: When 2 Seconds Matter — Building Real-Time AI Captions 
 and Live Interpretations - Arnav Angarkar
URL:https://cfp.devcon.org/devcon8/talk/HWKQHS/
END:VEVENT
END:VCALENDAR