Improving Voice Command Recognition Accuracy Using DialogFlow NLP
summary
Android uses the on-device SpeechRecognizer to obtain transcripts and sends them to DialogFlow (via SDK or a server-authenticated v2 proxy) using console intents, entities, required params and @sys.any. The workflow includes annotating real user utterances and handling SpeechRecognizer mis-hears; a prototype covering five commands (launch app, navigate, play music, send message, call) reduced no-match rate from 32.6% to 17.3%, demonstrating DialogFlow’s NLP pipeline improves voice-command recognition accuracy.
AI-written summary · about
chapters (19)
- 0:00Introduction
- 1:09Voice command methods
- 3:27Simple commands
- 4:11Complex commands
- 4:35Dialogflow NLP
- 6:55Dialogflow setup
- 8:53Intents and parameters
- 10:50Testing responses
- 11:39Android integration
- 13:52SDK and auth
- 14:13Speech-to-Text options
- 14:35Dialogflow entities
- 17:26Custom/system entities
- 21:55Accuracy improvements
- 22:43UI/UX for voice
- 24:14Training annotations
- 26:01Annotation challenges
- 27:26Measuring accuracy
- 28:36Summary
related talks
47mBuilding Voice-First Android Apps- 47mBuilding Voice-First Android Apps
35mHaving a Voice on Android: How hard can it be?
42mThe State of Voice Interactions on Android
55mConversational power for Android apps
29mDialogflow (API.AI) for Android — Introduction to Dialogflow (API.AI) App Development
43mThe State of Voice Interactions on Android
40mConversational power for Android apps