Voice Tech TO #1
3 likes442 views
The document provides an overview of voice technologies and conversational interfaces, detailing their definitions, use cases, and business value. It emphasizes the importance of natural language processing and machine learning in virtual personal assistants, highlighting various applications across industries such as customer service, education, and entertainment. Additionally, it discusses product design and development processes, illustrating these concepts through a case study on a voice tool for news updates.
Voice Tech TO #1
- 2. Connected Lab December 22, 2017 370 King St W #300 Toronto, ON M5V 1J9 / (647) 478-7493 An overview Voice Technologies Today
- 3. Yours truly Tim BETTRIDGE Product Designer at Connected Lab Guy TONYE Software engineer at Connected Lab Polina CHERKASHYNA Product manager at Connected Lab
- 4. Contents Understanding the key terms Use cases and business value Product and design process Case Study “Tailor News”
- 5. Understanding the key terms
- 6. Voice Technologies are a subset of conversational interface. Definition • A conversational interface describes a place where a user interact with a system using a conversation. • Those interface have been available for a little while. One of the most famous is the Interactive Voice Response (IVR) which is often used for automated customer service (for example when calling a bank and the automate voice guides the user by asking to use the keypad to provide information). Conversational interface
- 7. Chat bot interface A user interface which makes human interaction with computers possible through chat-like written conversation. Hybrid interfaces Interfaces which combine natural spoken conversation, text input, video etc. Voice interface Makes human interaction with computers possible through close to natural conversation by voice. Zoom on 3 types of conversational interfaces
- 8. What happens under the hood? A Virtual Personal Assistant at the core • The devices are the interface for the consumer. • When users converse in a conversational interface, the device forward the requests to the Virtual Private Assistant (VPA) behind the scene. The latter process the request and send an appropriate response to the device as a response. • The VPA is a web server. VPA NLP- NLU ML
- 9. VPA - zoom on two features Natural Language Two powerful features of the VPA are Natural Language Processing (NLP) and Natural Language Understanding (NLU) NLP and NLU are a tools and techniques useful to help the server convert human requests into action or set of actions that the machine can execute. Machine learning The VPA also leverages Machine Learning (ML) for two purposes. Understand the request • Because of accent and idiom, ML helps to adapt and improve the understanding of the user request Fulfill the request • ML is also used to tailor the actions and the response that is given to the user upon request
- 10. Some example of VPA Siri Assistant Alexa M Cortana
- 11. Use cases and business value
- 12. 15 mln+ Alexa devices 5 mln Google 50% of teens and 40% of adults $18.30 Billion USD by 2023 All major car manufacturers 55% of U.S. households by 2022
- 13. When is voice relevant for my business/ product? Fewer steps than via phone or PC Hands occupied Multi- tasking Assisting differently abled people Fun/ leisure Interactive learning Based on understanding these core use cases it’s possible to think of multiple voice specific use cases for different businesses and domains.
- 14. Kitchen is THE place where we have our hands occupied. Think of the various voice services which could be useful in this situation and brands which could own them. Step by step voice recipes Food delivery Voice search for music Kitchen timer (an embedded Alexa feature)
- 15. Cars are another “hands occupied” space. They hold ample opportunities for “aftermarket products” - devices which can help drivers operate navigation or car infotainment by voice if the vehicle doesn’t have an embedded voice assistant. At the same time most car manufacturers are building in voice capabilities into the new models. Those will cover multiple use cases: voice search, navigation input, adjusting car AC and other features and even coaching for new drivers.
- 16. Requesting for music is much faster by voice than using the phone and key music service providers are already offering voice-first music discovery and playback capabilities. Amazon Music launched a feature where users can search for music by voice using any combinations like “babymaking country music” or “slow Italian dinner music”. Play some Italian dinner music
- 17. Feeling sick is not a “hands occupied” situation, but it’s a time when ordering something by voice is much faster and and easier than doing so via a phone or lap top. Multiple brands could leverage this opportunity and create an additional entry point for their customers.
- 18. Products for people with special needs could benefit a lot from voice features. Bionic Laboratories are already using a voice interface to help patients, who are unable to use their legs to operate an eco skeleton.
- 19. Leisure and entertainment are one of the most demanded voice-first experiences. Providers of news, shows, radio and games could leverage are already actively leveraging voice for new product and services development and as an additional digital touch point with their users.
- 20. Situations where a voice interface is more suitable than touch aren't limited by our planet.
- 21. Smart voice assistants offer ample opportunities for education providers to create interactive “tutors”, who can help learn facts, listen to lectures and test your knowledge.
- 22. Customer services is a known area for voice capabilities. Here the experience will be improved as voice assistants are becoming smarter and offering more human-like communication patterns.
- 23. Voice platforms offer an opportunity for automation and thus savings for businesses. Think about a voice capability, which could help visitors seamlessly make an order in a restaurant, without waiting for a server.
- 24. There is space for product improvement, new product development, marketing and investment.
- 25. But there’s always a “BUT” … Users don’t invest any effort into new feature/ skill discovery Users learn by talking to the assistant - not by using the companion app If a feature or capability didn’t work the first time they will not try it again
- 26. Product and design processes
- 27. Fundamentally lean Hypothesis/ product vision Build Learn Measure Scale (Rapid prototyping and user validation) When dealing with new products and technologies it’s important to keep in mind that risks of failure are inherently high and “gut instinct” is not enough to create a successful product/experience. Start user validation as soon as you have a clear hypothesis and ensure that user validation is a frequent and continuous during the discovery, definition and build phases of your project.
- 28. Invest more when risks are lower Time Investments Risk Continuous user validation will allow you to iterate faster and eliminate risks while making only small investments of time and energy.
- 29. Steps towards effective CUI design Not everything is a good fit Pick the right use cases Develop user stories Experiences that make things faster & simpler 1 How do you want your persona to feel and sound Create the Personality 2 Write Dialogs Figure out the ‘happy path’ and then think about the other paths and branches. Conversation repair is very important. 3 1:1 operator allows for high experiential fidelity prototypes Interview before and after Become the puppet master. Test with real users 4 Record your test sessions Capture user utterances Use analytics to illicit insights Measure & Learn 5 Iterate, test, measure, and repeat Iterate & test 6 Iterate, test, measure, and repeat
- 30. Prototyping tools and methods
- 31. Spreadsheet and OSX TTS Simple and effective for early validation. • Have an external speaker and mic to create higher experiential fidelity. • Be ready to improvise. • Record your sessions with an audio recorder.
- 32. Interactive Keynote Keynote is a familiar way to prototype quickly and effectively. • Easier to see the dialog flow and follow along. • Be ready to improvise. • Conversational repair is essential. • Record your sessions with an audio recorder.
- 33. CUI Prototyping Tools Tools for CUI design and testing are in development. • Highest experiential fidelity • Transcripts of tests allow for later review • Utterance capture and analytics • Eg. Simili and others
- 34. Case Study: Tailor News A work in progress
- 35. The hypothesis Idea: we have daily company stand ups where we share tech news. Why don’t we create a voice capability which would provide latest news on a this or category (tech, business, economy). Hypothesis: it's faster and easier for users to request a news update by voice than via a mobile app.
- 36. Initial user testing • Users want to request news proactively e.g. ask for the news on a certain topic (Facebook), or category (tech news). • Users want to get only the top 3-5 headlines first and then have an ability to ask for more on the headline they found interesting or send the full article to their phone.
- 37. Iterating on the flow: v1.0 We designed the first flow as an assumption and started to iterate from there taking each new iteration to users, collecting feedback and improving the flow before doing any engineering work other than research of platform capabilities.
- 38. Iterating on the flow: v1.0 feedback from users • The intro is too long, especially when you listen to it a second time.
- 39. • Users are not interested in the source as long as they hear top 3 news items. • Users want distinct discourse markers “Number 1… Number 2… Number 3…”. Iterating on the flow: v1.0 feedback from users
- 40. Iterating on the flow: v2.0
- 41. • Its hard to remember 3 options of what you can do with the skill. Users prefer a shorter conversation like flow. Iterating on the flow: v2.0 feedback from users
- 42. • The added intro to the article wasn’t creating enough value for users. We should look at new ways to summarize articles Iterating on the flow: v2.0 feedback from users
- 43. Iterating on the flow: v3.0 - prototype
- 44. Ready for next test. Iterating on the flow: v3.0 - prototype
- 45. An alpha prototype of Tailor news for Google Home was quickly built using: Google Assistant • When registering as a google action (action.developer.google.com) a Voice capability can leverage the Google Assistant to process request. Dialogflow • Interface where the model for user utterances and the mapping with the actions to do upon receiving those request can be defined. AWS Lambda • Fulfilment of the action is done using a webhook on an AWS Lambda. Quick prototype: Tailor news on Google Home NLP-NLU VPA Assistant https://newsapi.org
- 46. Three key bullet points Use SMMRY the API behind the famous TLDR Redditbot- article summary (3 bullet points) rather than plus one sentence. Beta Testing the skill with a larger set of users Refining the dialogues We’ve established the flow now we need to finalize the actual wording. Next steps
- 47. Summary
- 48. Lean Fully execute the lean methodology to reduce risks. Remember that gut instinct is not enough for new product development. 1 Design process Pick good use cases, create a persona, test with real users, measure and learn. 2 Grab the opportunity Build your expertise now and be the first to make your products better and win over competition. 3 Summary
- 49. 1) Using voice technologies to build better products (existing or new). 2) Designing for Voice (dialogue building, guidelines, prototyping). 3) A detailed client case study. 4) Engineering: rapid prototyping, cross-platform capability delivery. 5) Testing voice products (QA). 6) Results of Tailor News launch, available analytics. Upcoming events: your input is welcome
- 50. Thank you!