How Alexa keeps getting smarter
summary
Alexa applies machine learning across five areas—competence, context awareness, knowledge, natural interaction and self‑learning—running at cloud scale on AWS and processing billions of utterances. ASR reaches ~95 percent accuracy since the 2010 deep‑learning inflection; on‑device wake‑word and acoustic models detect glass break and smoke alarms via long short-term memory networks. Neural pipelines (Mel spectrograms → neural audio) produce neural TTS voices and styles; features like Alexa Guard, Alexa Hunches and Alexa Conversations chain skills together and are already deployed and working.
AI-written summary · about
chapters (13)
- 0:00Start
- 1:42Rise of the Machine Learning Technologies
- 2:23Computer Technologies
- 3:03The Evolution of User Interfaces
- 4:21Echo Show
- 5:25Machine Learning Technologies behind Alexa
- 6:00Tts
- 6:59Asr Automated Speech Recognition
- 9:55Context Awareness
- 16:08More Natural Scale Interaction
- 19:36Text-to-Speech Tts
- 26:02Laden Row of Detection
- 29:46Content Context Awareness
related talks
39mIt's All in the Data: The ML Behind Alexa's AI Systems
42mConversational AI in Amazon Alexa
18mAlexa's next AI trick: Disappearing
52mSelf-learning Alexa: ML model updates with no human in the loop
30mHuman-like reasoning for an AI
06mAI and machine learning at Amazon
55mMachine Learning and AI in Java on AWS
15mAI at Amazon