UNDERSTANDING SPEECH: APPROACHES TO VOICE INPUT IN IOS APPS

Presented at NSNorth 2019 in Montreal

Transcript

1. UNDERSTANDING SPEECH: APPROACHES TO VOICE INPUT IN IOS APPS C H R I S T I N A M O U LTO N TEAK MOBILE INC. TEAKMOBILE.COM GROKSWIFT.COM @ C H R I S T I N A M LT N
2. WHY NOT?           Freeform input Mixed modes Background noise Accents Distraction WHY?         Hands/Eyes Free “Natural” RSI Accessibility
3. APPROACHES TO iOS VOICE INPUT Siri Shortcuts Voice Recognition and Local Parsing 3r d Party Chatbot SDK
4. SIRI SHORTCUTS •  Trigger actions in app from Siri •  Repeat previous action •  User defined command •  No actual voice recognition in app
6. SIRI SHORTCUTS PROS & CONS •  Simple implementation •  User gets to say what they want •  Suggestions •  Watch Support •  Inflexible •  Difficult to encourage use
7. VOICE RECOGNITION & LOCAL PARSING •  User takes action •  App starts listening •  Convert voice-to-text •  Work with text
8. SPEECH RECOGNITION
9. PARSING OPTIONS
10. VOICE REC & LOCAL PARSING PROS & CONS •  Uses Siri •  Apple’s servers: Privacy, limitations •  Simple dialogs •  Not intelligent •  Small bits of data
11. 3rd PARTY CHATBOT SDK •  Send audio/text to web API •  NLP/ML → voice-to-intent •  Higher level abstraction •  E.g., Amazon Lex, Microsoft Bot Framework, IBM Watson, Google Dialogflow
14. 3rd PARTY CHATBOT SDK PROS & CONS •  ML/NLP •  Conversation flows, intents, elicit parameters •  Maintain context •  User frustration •  Untrained voice recognition •  Cost •  Privacy
15. Seamlessness Relevance Trust Costs CONSIDERATIONS
16. UNDERSTANDING SPEECH: APPROACHES TO VOICE INPUT IN IOS APPS C H R I S T I N A M O U LTO N TEAK MOBILE INC. TEAKMOBILE.COM GROKSWIFT.COM @ C H R I S T I N A M LT N