How to build voice interfaces in a professional way
2 likes673 views
The document outlines a professional approach to building voice user interfaces (VUIs), emphasizing the growing importance of conversational interfaces and the potential for custom solutions beyond platforms like Alexa and Google Assistant. It covers key aspects such as when to consider VUIs, how to design effective user experiences, and the development process using best practices, including multi-language support and automated testing. The author seeks input and collaboration for continuous improvement in VUI development.
How to build voice interfaces in a professional way
- 1. How to build voice interfaces professionally Fabian Dill, DieProduktMacher Grand Event about Chatbots and Voice Interfaces June 23rd
- 2. Agenda When to consider voice user interfaces (VUIs) How to design great experiences for VUIs How to develop VUIs 2 1 2 3
- 3. When to consider VUIs? 3
- 4. Conversational is the new normal MESSAGING APPS HAVE SURPASSED SOCIAL NETWORKS (Monthly active users for top 4 social networks and messaging apps) Big 4 Social Network Apps Big 4 Messaging Apps 4
- 5. Why now? Because we can 0 5 10 15 20 25 30 2010 2011 2012 2013 2014 2015 Error rate 0 2 4 6 8 10 12 14 16 Error rate 1. Deep Learning Algorithm 5 ERROR RATE IN IMAGE RECOGNITION (ILSVRC RESULTS) ERROR RATE IN SPEECH RECOGNITION
- 6. Two types of conversational interfaces are on the rise. 6 What kind of conversational interfaces? TEXT | Chatbots VOICE | existing platforms custom build
- 7. 7 • Siri (Apple) • Google Assistent (Google) • Google Home (Google) • Cortana (Microsoft) • Alexa (Amazon) • Bragi Voice interfaces are not restricted to Alexa skills
- 8. In addition to these voice interface platforms custom build solutions are possible: • Matrix Labs • Google Speech API • Alexa Services / Amazon Lex 8 You can also built your custom voice interfaces
- 9. 9 Half of the Germans already used a voice interface
- 10. 10 Before Alexa VUIs were tied to Smartphones
- 11. 11 As you can see in the typical applications
- 12. 12 In the following I will focus on Alexa skills
- 13. OPERATORS & EXECUTORS • Commerce (fashion, flowers, food delivery) • Booking (Tickets, Travel, …) • Banking DOMAIN KNOWLEDGE EXPERTS • Customer Support • Advice and Recommendations • Domain-specific knowledge: • Sports • News • Travel • Insurance 13 Three types of Conversational UIs (Chatbots) PERSONAL ASSISTANT • Google Assistant • Facebook M • Apple Siri • Microsoft Cortana • Amazon Alexa
- 14. 14 (SHORT) INFORMATION Types of Voice Interfaces TV program, weather, news, knowledge, status information ENTERTAINMENT Jokes, Games, … ADVICE AND RECOMMENDATION Fashion, Travel, ... TRANSACTIONAL Shopping, Booking CUSTOMER CARE?
- 15. 15 Smart Home Physical Device Game Complex Interaction Handsfree environment (e.g. car, climbing, assembly, …) Your user experience can significantly improve When to consider Voice Interfaces?
- 16. 16 How to design great experiences for VUIs
- 17. 17 PRODUCT ATTRIBUTES • Goal • Target Group • Core Value PERSONALITY • Age and gender • Adjective • Name EXAMPLE DIALOG • Text • Talk • Record Our current best practices
- 18. 18 ASK TELL REMEMBER … keeps the dialog going, listens to further user input … use it for questions, reprompts, error messages Endpoint of the dialog, shuts down the session Use it for results, stop and goodbye Creating a DynamoDB storage is just one line of code! Create dialogs
- 19. 19 How to develop VUIs
- 20. 20 The usual Alexa skill environment
- 21. 21 • Use your favorite IDE (code completions, shortcuts, etc.) • Realise parallel development (>1 developer) • Extend live product with new features • Be able to rollback (in the unlikely case someone messed it up) Why professional development?
- 22. 22 Serverless to the rescue
- 24. Our road to a professional development enviroment for VUIs (Alexa) AUTOM. TESTING How to create unit and integration tests MULTI- LANGUAGE SUPPORT Separate language from functionality STAGING Create stages for interaction model and backend lambdas LOCAL DEVELOPMENT Set up a local development environment
- 25. 25 • Develop a skill in several languages (English, German) • Separate utterances from functionality • Require the correct language file at runtime Multi-language Support Preparation
- 26. 26 Multi-language Support Input / Output language based on user settings in Alexa App locale: 'de-DE' handler.js i18n/de-DE.js i18n/en-US.js i18n/en-UK.js locale: 'de-DE' // require specific language file const languages = require(`./locales/${lang}.js`) // get text message according to locale const getText = (alexa, message, attributes, index) => { return Language.getText( alexa.event.request.locale, message, attributes, index) } ALEXA AWS lambda
- 27. 27 Separate language from functionality (DEMO)
- 28. SERVERLESS INDIVIDUAL DEVELOPMENT INTEGRATION PRODUCTION INTERACTION MODEL / SKILL BACKEND / LAMBDA GIT REPO / GITFLOW static arn 28 Parallel Development, Staging and releases with Alexa manual copy manual copy manual copy FEATURE DEVELOP MASTER RELEASE
- 29. 29 Copying interaction models = manual 5-step process
- 30. 30
- 31. • Error prone developer has to be on the correct branch • Next step: automate or use continuous integration system 31 We use Serverless to connect correct branch with referring stage
- 32. • Install as a serverless plugin • Start a local server • Share localhost with the internet • Configure Alexa skill to use your HTTPS endpoint 32 We are working on a serverless plugin to allow for local development
- 33. 33 Develop and test locally until commit
- 34. • Several use cases applicable for voice interfaces • Design dialogs by prototyping them extensively • Alexa infrastructure is not yet ready for a professional setup • There are several ways to optimize your development process (we shared one of them) • We are on our journey to continuously improve this process • We are looking for input, comments, other experiences and new team members 34 Conclusion
- 35. Contact: 35 Fabian Dill CEO Email: fabian.dill@produktmacher.com Telefon: 089 / 189 46 54