Conversational AI in Amazon Alexa
summary
Overview of how Alexa actually works: far‑field microphone arrays plus on‑device beamforming detect the wake word and send audio to cloud ASR and NLU that must answer in under a second; a hybrid of ML models and signal processing powers intent routing to 15,000 skills or domain experts, and device embedding via the Alexa Voice Service. Novel features include echo spatial perception to pick which device to address and a $1,000,000 Alexa Prize for 20‑minute coherent chat.
AI-written summary · about
chapters (13)
- 0:00Introduction to Alexa
- 2:58The Alexa device ecosystem
- 4:12Alexa capabilities and use
- 6:22User demographics and patterns
- 7:32Developer tools and integration
- 9:34AI challenges and solutions
- 12:43Conversational AI and context
- 15:12How Alexa processes requests
- 18:16Machine learning and synthesis
- 20:32Customer experience and voice
- 22:11Engagement programs and funds
- 24:30Alexa prize and the team
- 26:19Q&A session
related talks
32mHow Alexa keeps getting smarter
43mBuild Voice-Enabled Experiences
39mIt's All in the Data: The ML Behind Alexa's AI Systems
01h 08mUnderstanding Amazon Alexa
52mThe most natural UI. Building voice-enabled experiences with Amazon Alexa
18mAlexa's next AI trick: Disappearing
45mKeynote: Conversations with Machines
59mProgramming for Amazon Alexa