Скачать или смотреть Voice AI & Voice Agents Course: Session 2 - Intro to Voice AI Infrastructure

Learn how to build and scale real-time voice agents with insights from Daily’s product team—covering WebRTC, latency, infrastructure, and the future of Voice AI.

In this session, Mark from Daily walks through how to build and host real-time voice agents using modern infrastructure tools. He explains the differences between REST backends and real-time audio systems, why WebRTC performs better than WebSockets for client communication, and how to manage latency, edge routing, and resource allocation per bot. The session also covers practical advice on handling cold starts, choosing platforms like Fly, Cerebrium, Modal, Baseten, Fireworks, or Vapi, and how telephony fits in with PSTN, SIP, and CPaaS providers like Twilio and Telnyx. The Q&A section dives into latency targets, container choices, and deployment strategies for developers building on Pipecat.

👉 Watch the full series:    • 2025 Voice AI and Voice Agents Course  
Join Daily's Discord:   / discord  
www.pipecat.ai
www.daily.co

Table of Contents:
0:00 Intro
1:22 Mark intro & session scope
2:48 REST vs real-time (voice infra is different)
8:47 WebSockets vs WebRTC (fit & tradeoffs)
13:05 Latency & edge routing
15:36 Per-bot CPU/memory rules of thumb
19:03 Getting started → scaling & cold starts
27:07 Infra Landscape
31:33 Telephony basics: PSTN/SIP & call flows
34:44 WebSocket example
36:38 SIP+WebRTC example
39:43 Q&A highlights

#voiceai #webrtc #aiinfrastructure #speechtotext #voiceagent #realtimecommunication #cpaas #pipecat #twilio #telnyx #vapi #latencyreduction