1. Voice Computing and Reaching the 3B People at the Bottom of the Pyramid Raj Reddy Carnegie Mellon University Pittsburgh, PA 15213 Sep 20, 2016 Heidelberg Laureate Foundation Talk

2. Bottom of the Pyramid Represents 3 Billion People with Incomes of less than $2.5 a day  The Bottom of The Pyramid is The Largest, But Poorest Socio-economic Group.  Globally that is the 3 billion people who live on less than say $2.50 per day.  Most of Them Are Also Semi-literate, i.e., Cannot Read, Write and/or Understand Any Language  Cannot use Keyboard or Touch based Computing Apps  If You Are a Semi-literate Person on The Planet that Only Acceptable Mode of Communication is Speech  Voice Computing a la Amazon Echo and Enhancements is the Key  Personal Assistants that Require Only Speech based Interaction are Essential for Such Populations

3. Voice Computing to The Rescue  Voice Computing (No Keyboard or Touch) Can Help The Semiliterate to Read Newspapers, Watch Foreign Language Movies, Listen to Khan Academy Lectures, Order Groceries Online, Banking Online, and Vote Online  A Mobile App for Entertainment and Education  Dynamic Real-time Translation of a Video Dialog from English to Telugu  Text to Speech App for Newspaper Reading Assistant  Ecommerce and eBanking  Voice Authentication, Authorization and Audit  Learning Without a Teacher  Tutor for Listening and Speaking English  Enabling Digital Democracy  Vote Online (Authentication, Authorization and Audit)  Illiterate Populations Will be the Biggest Source of Customers for Speech Based Apps in the Future

4. Existing Technology Can Create Compelling Apps to Empower the Semi-literate Population of the World  Speech to Speech Exists (Microsoft, Facebook and Others)  BTW, Current Implementations are Based on Incorrect Business Assumptions  Available only for Commercial Languages  English to Chinese Speech to Speech Translation Demonstrated in 2012  Text Based “Translate” App of 2016 has to become Speech Based  Languages Supported Based on Commercial Viability  Not Need based  Unlikely to result in Killer Apps  Apps Tailored to Semi-literate Populations Will Become Killer Apps  1 Minute Learning Time; Two clicks; and Spoken Dialog   No Keyboard or Touch   All Such Apps Will Require Speech Recognition, Speech Synthesis, Spoken Dialog, and Speech to Speech Translation  Speech to Speech Translation  Entertainment (Movies) and Education (Khan Academy)  Translate Live Dialog  QA Dialog (Siri and Cortana)  eCommerce and eBanking  English Language Learning - Detect Pronunciation Errors

5. Typical App for the Semi-literate: Asha Asha is an Intelligent Agent That Anticipate What You Want To Do And Helps You To Do It Using Local Language and Clarification Dialog Entertainment and Education: Streaming Video Translation Entertainment and Education: Streaming Video Translation Asha play Hamlet (BBC Shakespeare) Asha play Hamlet (BBC Shakespeare) Reading Newspapers: Text to Speech Translation and/or Synthesis Reading Newspapers: Text to Speech Translation and/or Synthesis Asha read Eenadu Asha read Eenadu Buying and Selling: Voice Dialog Management Buying and Selling: Voice Dialog Management Asha order milk and bread Asha order milk and bread Banking: Monitor Bank account, Pay Bills Banking: Monitor Bank account, Pay Bills Asha charge my mobile device with 1000 rupees Asha charge my mobile device with 1000 rupees Communication: Voice and/or Video Email, Chat Communication: Voice and/or Video Email, Chat Asha call my Grandson in Seattle Asha call my Grandson in Seattle Online Voting Online Voting Voice Dialog to enable the Voting Process Voice Dialog to enable the Voting Process

6. 6 Architecture of Asha Voice Computing App Asha is a Mobile App for Customized for Each Person Asha is a Mobile App for Customized for Each Person Designed to be Non-intrusive, Autonomic, and Device Independent Designed to be Non-intrusive, Autonomic, and Device Independent Always On, Always Present and Always Working Always Learning Enduring (Life-Long) Asha Monitors, Analyzes and Learns From Experience; Asha Monitors, Analyzes and Learns From Experience; Learn From Own Experience And Experience of Others Learn From Own Experience And Experience of Others And s hare knowledge with a community of Ashas And s hare knowledge with a community of Ashas Automated Discovery of Data and Information Sources Automated Discovery of Data and Information Sources Sharing Data Among Asha Apps: Data, suitably anonymized, can be used to learn appropriate responses for every situation Learning preferences by observing user choices, Learning by task similarity and user similarity, Learning by error correction and Simply learning thru clarification dialog ( does that mean yes? Would you care to define it?).

7. PPP Business Model A Public-Private Partnership Model of App Development  No Single Company Can Afford To Make Significant Investments in Orphan Languages  Partner with Local Governments  Industry Provides Technology, Develops and Maintains Apps  Local Government Pays for Data Collection and Deployment  Initially Worth Considering for Populations of 20 million or More In The First Phase  India: Bengali, Telugu, Marathi, Tamil, Gujarathi, Kannada, etc  Africa: Arabic, Swahili, Berber, Hausa  For Illiterate Populations It Will Become A Lifeline And Used Everyday  Once working even literate members of the region will use it because of Convenience

8. In Conclusion 3B Semi-literate Populations of The World Are A Major Untapped Market For IT Companies 3B Semi-literate Populations of The World Are A Major Untapped Market For IT Companies Effective Use Of Speech Technology and Voice Computing Is The Only Option To Support Their Needs Effective Use Of Speech Technology and Voice Computing Is The Only Option To Support Their Needs We Have All The Needed Technology And Tools We Have All The Needed Technology And Tools Partnering With Local Governments May Be One Way To Reduce The Cost of App Development Partnering With Local Governments May Be One Way To Reduce The Cost of App Development Sharing Sparse Language Data Among Leading Tech Companies May Be Desirable Sharing Sparse Language Data Among Leading Tech Companies May Be Desirable