Alisa, Don’t Lag!
summary
Describes efforts to make Yandex’s voice assistant Alice respond faster by improving its response latency. Covers Alice’s system architecture and how that architecture evolved, plus practical measures and optimizations applied to reduce reply time.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 1:31Alice’s hidden history
- 7:37Architecture changes
- 9:02Speed optimization start
- 12:57End-of-request detector improvements
- 15:23Dialog engine optimization
- 20:53Response speed UX study
- 23:23Optimization results and impact
- 27:40Infrastructure and load issues
- 31:32Q&A part 1
- 38:57Q&A part 2
- 46:10Interacting with Alice
- 49:24New features and roadmap
- 51:59Q&A part 3
related talks
35mInside the New Generation of Alisa
19mReal-Time AI: Low Latency Solutions for Interactive Applications
31mMy AI is slow. Make it faster!
13mHow We Reduced Latency by 70% While Maintaining 99% Accuracy
25mEngineering voice agents: Latency, quality, and scale
32mHow Alexa keeps getting smarter
17mServing Voice AI at Scale
39mIt's All in the Data: The ML Behind Alexa's AI Systems