Going Full Duplex - Building Apps for Speech
summary
Claim: modern speech APIs hit near-human transcription and let apps actually act on voice. Demos use the Microsoft Speech SDK with one-line recognize calls, a Custom Speech model that learns domain words (it correctly recognizes “Stegosaurus”), LUIS intent parsing to extract actions and entities, and a Azure Functions webhook triggered from Google Assistant to deploy resources — the deploy flow works live. Also shows an Alexa bridge sending raw utterances to a bot, plus ~80 synthetic voices in 48 languages and translation/recognition improvements.
AI-written summary · about
chapters (15)
- 0:00Introduction
- 1:50Agenda
- 2:45Communication channels
- 6:18Recognition
- 11:50Custom Speech
- 16:50Synthesis
- 18:31Machine Translation
- 18:58Demos
- 21:58Intent Detection
- 22:42NLP Demo
- 25:35Batch Transcription
- 33:13Demo App
- 36:08Test App
- 43:03Azure Functions
- 46:03Alexa Bridge
related talks
50mSpeak To Me: Voice App Conversation Practices
19mVoice AI: when is the "Her" moment?
17mBuilding Effective Voice Agents
09mConversation with Collin Borns (Head Biz Dev at Speechly) and @ianutile
47mBuilding Voice-First Android Apps
21mTalking to others for you: Leveraging machines that speak the “human conversation” protocol
46mWELCOME TO THE AGE OF CONVERSATIONAL INTERFACES!
20mDemo Session