Natural language understanding (along with speech recognition) is one of the foundational technologies underlying the Voice-First revolution. When it works well, the user experience is natural, frictionless, and efficient. When it doesn’t work well, the results are frustrating and irritating. This session brings attendees up-to-date on current natural language understanding technology, explaining how it works and what’s going wrong when it doesn’t. We cover current technologies, including both traditional rule-based approaches as well as machine learning technologies such as deep learning. We also review current proprietary natural language application tools such as the Amazon Alexa Skills Kit, Google Dialogflow, and Microsoft LUIS and discuss open source alternatives. Attendees come away from the session with an understanding of current natural language technology, its capabilities, and its future directions.

With the explosion of smart-speakers, chatbots, and high-quality cloud speech technologies, conversational interfaces are burgeoning. In spite of the progress of AI, most conversational technologies still require a high degree of design skill to make them usable and effective. This class takes a very practical deep dive into the discipline of conversational design. Attendees get a brief primer into the current technologies, including an understanding of why good design is still so necessary in spite of recent advances in AI. Attendees learn how to select the target tasks for a conversational user interface, and then design a small conversational interface. Topics covered include task selection, intent design, scripting, and structuring your natural conversation to make it discoverable and stable. The course is open and appropriate for those considering or working with conversational solutions for call centers, smart speakers, kiosks, or chatbots. It is not technical, and no prior experience is necessary.

This practical, hands-on workshop provides an in-depth introduction to the design and development of conversational chatbots using Google’s Dialogflow platform. We start with a brief overview of chatbots and conversational interfaces, looking at relevant technologies and tools. The main part of the workshop is devoted to hands-on design and development of several sample conversational chatbots. Covered topics include basic one-shot queries; slot-filling dialogs; dealing with context and follow-up queries; testing your chatbot on the Google Assistant Simulator; training the chatbot to improve its performance; enhancing system output using SSML (Speech Synthesis Markup Language); using fulfillment to interact with external web services; and how to display on surfaces that have visual and well as speech interfaces. Attendees will be provided with instructions for downloading software to their PC laptops prior to this workshop.

Wizard of Oz  (WOZ) testing is the number-one VUI test method. The name comes from the concept of "the man behind the curtain" in The Wizard of Oz stories: what appears to be a machine is actually controlled by a person. The tester (the Wizard) responds to voice requests; participants interact as if speaking with a functional voice system. With WOZ testing, you get valid user data long before any development work has been done. UX/UI designers and product owners at all levels of expertise will benefit from this fun, hands-on workshop. Attendees learn all about WOZ testing, including what it is, why to do it, when to do it, and what can be learned from it. Attendees also learn how to choose test cases, set up, run, test, interpret, and present results.

You’ve heard it before. Over and over, in fact. “Improve your customer service.” “Optimize the customer experience.” But what does that even mean? Today, when your intersection points with your customers may number in the dozens (or even hundreds), tackling customer service or customer experience holistically is impossible. You can’t magically get better at every customer touchpoint. But you can get better at the touchpoints that matter. Think Small is a fast-paced, dynamic, inspirational, and hilarious presentation that shows you how to gain and keep more customers by improving on only the three things that really matter to your customers.

Chack back for the latest details.

Digital assistants that communicate with customers using human language—text or speech—are increasingly necessary. They are evolving beyond classical customer service to include marketing and sales. Companies that recognize this opportunity will benefit from both publicity and new customers, leaving those that move more slowly scrambling to catch up. This talk by the author of the new book, Computer Intelligence, describes what to expect in advanced deployments and approaches to creating your own.

The goal: to incrementally and economically add AI and machine learning capabilities to legacy IVR applications. This session provides the road maps, steps, options, and examples of successful AI migrations and deployments. We also examine how to use these approaches to add conversational and omnichannel components to the migration architecture. Finally, we address AI integration targets, and options in relation to an organization’s culture, vertical market, size, appetite for early technologies, and degree of investment in legacy applications.

Join us as we seek to demystify the complex topic of voice biometrics, one of the strongest ways available for you to secure your voice applications and telephone agents against fraudsters and identity phishing. We simplify the common terms, discuss ideal use cases for the different approaches to the technology, and share insights around best practices. With no extra cost for hardware, voice biometrics can be the smartest way to introduce a highly secure and convenient authentication mechanism.

This talk discusses the latest methods for increasing security and why increased security and frictionless user experiences don’t have to be an “either/or” situation. It discusses how conversational AI leverages voice biometrics and NLU to authenticate individuals within seconds of users speaking across voice channels (smart speakers, IVRs, etc.). Attendees learn new ways for flagging fraudulent activity in real time based on a choice of words, natural utterances, and patterns of speech or writing during an interaction with a human or a virtual assistant. This talk also describes continual, multi-layered analysis that combines traditional voice biometrics with other methods like conversational biometrics and behavioral biometrics.

Encouraging customers to enroll in voice biometric applications is critical to the success of these applications. Persuading customers to enroll (for active/text-dependent biometrics) or talk long enough (for passive/text-independent biometrics) to create an enrollment is a frequent challenge. We discuss real-world best practices to increase enrollment, including enticing and encouraging customers, reducing agent friction, and ways that you might be able to leverage existing data and prior customer interactions in the enrollment process.

The quality of synthesized speech and the ease with which a person’s voice can be synthesized from a shrinking amount of speech data poses a threat to all systems that use a person’s voice for security. Global competitions, such as the Interspeech Conference, encourage companies to develop new strategies against a baseline set of threats. ID R&D’s Amein discusses insights and strategies submitted for publication in the industry leading Interspeech Conference of March 2019, as well as ID R&D’s perspective on the areas of synthetic speech that present the greatest threats.

At its best, spoken languages connect two minds. Conscious and sentient humans have minds. Machines do not. In this session, Balentine describes a continuum of three conversational idioms— task-speech, thought-speech and social-speech—that together define a newly emerging “alien mind,” which is useful and productive, unlike many human minds. Intelligent conversation with an alien mind bridges the gap from today’s early (but mindless) successes to tomorrow’s ubiquitous interactions with agents, robots, kiosks, and disembodied entities.

Defining the intents, dialogue responses, and variations can all increase the time to create a conversational assistant. Yet the content needed to power an assistant may already exists in places like public FAQ pages or knowledge management systems. This talk addresses how to leverage document understanding AI to extract valuable content and uses AI search to find and combine speech and visual interactions to quickly bootstrap an effective conversational interaction.

High-quality Actions see high engagement with Google Assistant users. What makes them high quality? Drawing from real case studies, learn how to reduce development errors, enhance discovery of your Action, grow your userbase, and avoid mistakes along the way when building quality and engaging Actions.

e’re ready to help you evaluate the right speech analytics solution for YOUR organization. Join us to learn all you need to know about AI Speech Analytics from the bottom up. In this educational, interactive session we’ll help you get a better understanding of: What you need to know as you research your options, including critical questions to ask to avoid speech analytics failure; How to establish a high-level business case and identify applicable use cases; Proof of value (POV) before you buy. Join Roger Lee, aka Professor CXi, Vice President, Customer Success with Gridspace, as he shares valuable insights that will help make you an informed buyer. 

This talk reviews techniques for gaining insights through text mining of recorded customer interactions, including the following: scoping and sampling to ensure that a text mining solution is quick to market, economic to scale, and conformable to other processes or speech solutions in the enterprise; champion/challenger testing of multiple text mining platforms; understanding the benefits of sentiment analysis; and using unsupervised clustering to illustrate regularities in elements such as products, features, events, and responses.

This talk discusses the benefits of performing a deep analysis of user turns in a conversation. As an example, we review a case where, through deep analysis of bot interactions, the U.S. Army learned that military personnel were frequently asking the bot about post-traumatic stress disorder (PTSD) symptoms and support. Consequently, the Defense Health Agency was informed that it needed to provide more anonymous self-service PTSD resources for soldiers.

A large manual effort is required to collect dialogue flows and intent training data needed to design an effective dialogue system. We extract intents, flows, slots, and training data from existing customer service call transcriptions and chat logs (corpora). This session surveys the state of art, direction of research, and review product offerings from companies including Facebook, Google, Nuance, Microsoft, IBM, and Discourse.ai.

Customer service AI chatbots (bots) are being unleashed into the world without sufficient testing. This talk covers the new approaches to AI bot validation and how to prevent your customers from being customer experience guinea pigs. The most thorough way to test bots is with bots. Attendees learn how to use technology to create virtual customers, approaches to generate additional training and testing data, and how humans fit in the loop.

The MRS “Adam Team” describes how we built an AI-enhanced call center. From an initial idea using a Raspberry Pi to a virtual agent which has collected more than $2 million in payments during the last 18 months, Adam, a multi-cloud solution based on IBM Watson, has radically changed our company and has the potential to disrupt the call center industry. By partnering with IBM Watson, MRS has significantly grown and improved Adam.

We cover some of the top tips and pitfalls learned by bringing hundreds of AI voice applications to production at Google Cloud. While we use Google products as an example, the concepts in this session are universal to all AI-based voice applications. We focus on realworld, production-level product management, engineering, and operational issues and developing, deploying, and managing AI-based voice apps from project inception to maintenance in production.

Open to all full-conference attendees and Networking Pass holders. Check SpeechTEK.com for details or follow #SpeechTEK on Twitter.

There are many ethical issues raised by modern speech processing technology, including using synthesized speech to mislead people, or trick biometrics systems. Should speech technologies be available individuals, organizations or governments that do not respect human rights, or who may use speech technology to commit crimes? How concerned should we be with the problem of human workers with speech technology? What might seem right to one person will seem wrong to another. What are the most important speech technology abuses and can we as speech technology professionals help prevent them?

Smart speakers are enjoying a high adoption rate. A demographic that potentially could benefit from this emerging technology is seniors. This talk provides a summary of 10 face-to-face interviews with residents of assisted-living homes who use smart speakers. Questions discussed in the interviews include the ease of adoption of the technology, how the technology has affected people’s daily routine, what features they find most useful, and which features they find frustrating to use.

This session includes demonstrations of two innovative applications of speech technology.

This demonstration shows the power of multimodal biometrics with a demonstration of frictionless authentication in a chatbot. Using five layers of biometrics, users are automatically and continuously authenticated as they interact normally with the app.

The Sentence Weaver addresses language learning for children with autism. It elicits phrases and sentences from users and then analyzes the users’ spoken responses, highlighting instances of improper word choice, word endings, or word order.

The Keynote will be announced soon. Check back for the latest details.

This is a three-part introduction to the Open Voice Network, the nonprofit industry initiative dedicated to a future of voice assistance that is open–standards-based, interoperable, accessible, and data-protected. We describe the why and the who, with core value propositions; explain the how and how fast; and explore an approach to the what (the proposed standards and architectures) in this fast-paced glimpse into the future.

Due to the lack of standardization in the domain of intelligent agents, there is no interoperability among them. This talk introduces some first thoughts toward standardization of voice-based intelligent personal assistants that are currently in discussion at the W3C Voice Interaction Community Group, including identification of areas for standardization, interoperability of IPAs, architectural drafts, and messaging formats like a JSON representation of semantic interpretation. We conclude with an outlook of next steps of our standardization efforts.

The dramatic improvements in speech technology that we’ve seen in the last few years have been based to a great extent on massive amounts of recorded speech data. This panel discusses what the limits should be with the use of this speech. Should callers be asked if each conversation can be recorded, or is it enough to ask once when they start using an app or device? Should callers be told how the recordings will be used? May the recordings be used to improve the caller’s speech recognizer, improve the caller’s dialogue, analyze the recordings to extract the caller’s interest in products and services, or synthesize the caller’s voice? May the recording be sold to others? Who owns the callers’ recordings? What legal protections are currently available, and are additional legal protections needed?

Learn all you need to know about speech analytics from the bottom up. In this educational session, we help you get a better understanding of the following: what you need to know as you research your options, including critical questions to ask to avoid speech analytics failure; how to establish a high-level business case; and proof of value (POV) before you buy. This session provides valuable insights to make you an informed buyer.

Voice analytics AI integrated into speech recognition engines is bringing a personalized customer experience to the voice age, allowing voice kiosks to instantly get a read about new customers based on voice characteristics, understand emotional state, and learn customer ordering habits over time. This talk includes a service application demo that combines wake words, NLU, voice analytics, biometric recognition, and a variety of other technologies to enable a truly personalized ordering experience.

With all of the hype around smart speakers, it’s easy to get the impression that there’s nothing else interesting happening in speech technology today. In this presentation, you hear about some innovative applications of speech technology which don’t involve smart speakers.

Smart speakers have the fastest adoption of new technology in human history—even faster than the smart phone. However, most brands have had very limited success with their Alexa Skills and similar voice experiences. Why is that? Join Tobias Dengel as he explores how some of the largest brands in the world, such as HBO, Fox, Regal Cinemas and Synchrony Bank, are shifting their thinking from voice to multimodal to truly take advantage of the platform.

We analyzed the emotional speech outputs of films produced by eight renowned directors, characterizing their work on an emotional spectrum. We then developed a rich recommendation system, which is more objective because it takes into account the script’s and director’s intentions. The emotional charge of a movie can influence our preferences as it gives us one more piece of information to decide what we want to see next.

We describe the process of designing, implementing, and using an intelligent agent for collecting field data in the food supply chain. We present our key learnings about the process, methods, and acceptance of speech technology. We describe the advantages and limitations of speech interaction methods, as well as the value of a human-in-the loop at key stages. Lastly, we make some predictions about the future of using speech in commercial and industrial use cases.

Remote patient monitoring and screening overcome the shortage in mental health professionals and long waiting periods while significantly reducing medical costs. We use remote, objective, continuous patient monitoring and risk group screening based on an evidence-based vocal biomarker. The technology tracks changes in medical states by automatically monitoring patient speech patterns captured in everyday mobile interactions. Voice analysis extracts unique prosodic (non-content) speech features such as intonation, rhythm, pace, and emphasis, which glean universal voice parameters to address the physiological aspects of speech.

This talk details how Discover Financial Services achieved its implementation goals for a conversational assistant for customer service. Discover’s requirements were as follows: Provide a low-effort customer experience with a conversational UI; reduce Average Handle Time (AHT); reduce transfer rate; increase IVR containment; and achieve a No. 1 customer satisfaction ranking with J.D. Power. This talk describes the Discover project implementation and explains how these goals were met.

Customers today expect personalization and are increasingly comfortable interacting with virtual assistants like Alexa and Siri. Customers prefer self-service to the often-frustrating experiences of navigating traditional IVR systems and/or repeating themselves to strained live agents who must search for needed information. We’re designing a solution to improve call center operations by creating a more user-friendly customer experience that increases self-service across platforms, expedites call resolution times, and derives customer insights through call analysis.

All commercial conversational platforms provide an NLU engine that can be trained based on sample sentences for each intent in the application, but they typically do not provide good guidelines on how to make sure that NLU engine produces the best possible accuracy. This results in a poor conversational user experience. This presentation describes techniques that have been successfully used to improve NLU accuracy and presents experimental results obtained with different NLU engines and task domains.

Planning to create a voice application and you need to support both Alexa and Google Actions? What are the similarities and differences in their stacks, security models, and hosting architecture? What are the most important technical considerations before designing a voice-first, cross-platform experience? Learn from Microsoft Alums and AWS Certified Skill Builders how to overcome the challenges in a multi-platform environment, from design to implementation. Take home tips and tricks you can use for your next project.

This talk highlights how the introduction of natural language not only benefits the bottom line, but dramatically improves the customer and employee experience. We discuss how a firm focused on worldclass, award-winning care can actually improve the experience while delivering deeper omnichannel analytics and innovative routing and treatment strategies with voice. This talk includes tips on how others can raise their utilization in a customer-friendly way and how we made the case to the C-suite.

Vendors, designers, developers, and users of intelligent agents: Express what you need in standardization efforts for interoperable agents. What are your priorities for user interface consistency, intelligent agent independence from hardware, switching among agents, agents invoking one another, agents sharing data, agents sharing context, agent marketplace, discovery of agents and their capabilities, common design tools, templates, wizards, guidelines and best practices? What is your advice for standardization efforts?

Speech analysis is being used to assess recordings of human voices to diagnose conditions such as depression, Parkinson’s, Alzheimer’s, heart disease, concussions, migraines, PTSD, and even suicide risk, among others. That list is expected to grow. This may sound like a dream come true, but there are always concerns.

Every successful speech application is based on training data-representative examples of the utterances spoken by users when they interact with the system. Techniques for obtaining data include conversations with users, looking at call center logs, and generating examples from grammars. But example utterances aren’t enough. Data also must be annotated, or associated with meanings. This session addresses the full data collection and management cycle, including issues around handling sensitive data.

Siri, Alexa, and Assistant are used primarily for practicalities like playing music or setting timers. But the banality of the present obscures the promise of the near future, as voice interfaces allow computers to become personified as human-like beings. This prospect is both exciting and unsettling. Tomorrow’s synthesized agents will upend control of knowledge; befriend, advise, and surveil; and even preserve memories of lost loved ones, as with Vlahos' own Dadbot project.

Consumers have become accustomed to using speech to interact with devices in their everyday lives, but conversational AI is less common in the contact center. This session shows you how to leverage conversational AI technology in your contact center to give your customers the seamless, low-effort experience they deserve. We explore each of the critical steps—from reviewing your customer’s journey and using self-service to picking the right conversational IVR platform.

The age of AI-powered robots and devices that can have full-fledged intelligent conversations with people seemingly is now upon us. Or, is it? Not quite. While the last few years have seen an explosion in chatbot and virtual assistant (VA) development platforms, these chatbots can only handle simple tasks, like answering FAQs or routing to an agent. This talk describes innovations that will make it possible to have advanced, enterprise-grade, conversational experiences.

Current AI and machine learning (ML) technologies are beginning to change the way we build and innovate. However, the power of our current ML technologies is not fixed. This session explores where ML is at the moment and where it is heading in the next 10 years. How are its current use cases different from those we may see in the future? This session explores these themes and their impact on speech technology applications.

Open to all full-conference attendees and Networking Pass holders. Check SpeechTEK.com for details or follow #SpeechTEK on Twitter.

Learn more about the latest technological developments in speech technology from leading-industry companies. These sessions are open to all full-conference attendees and Networking Pass holders. Check SpeechTEK.com for details or follow #SpeechTEK on Twitter.

An essential focus of Nu Echo is to develop advanced conversational IVR applications offering sophisticated mixed initiative dialogues with a natural and effective user experience. To meet these objectives, we have explored dialogue management capabilities offered by the Rasa solution and its integration with VoiceXML. During this presentation, we describe the rationale behind the Rasa-VoiceXML solution, the use cases that were explored, and how we leveraged both deterministic dialogue strategies and machine learning capabilities.

Features needed for future conversational advancement—turn taking, prosody, multi-leveled confidence, and noise characterization— are limited or nonexistent in today’s most popular streaming-text black-box recognizers. We are at risk of losing experiential skill with basic speech technologies. Knowledge of speech matters. In this hands-on interactive session, we use newly developed tools from Intelligently Interactive to explore basic ASR technology—learning about terminology, error types, prompting methods, turn taking and dialogue design philosophies. The tools use open source Sphinx technology running on iPhone/iPad platforms to support under-thehood examination of speech recognition, voice-activity detection, basic dialogue design, and usability-testing methods. If you are an iOS user, keep the tools and continue your research after you leave. If not, we provide shared resources and you can apply the knowledge to other products and platforms. Attendees will be provided with instructions for downloading software to their iPhones or MacBooks prior to this workshop.

Testing and evaluation processes are crucial to the success of any NLP conversational system, but testing IVR and multimodal systems presents unique challenges. Focusing on multimodal applications that involve speech and other modalities, we describe the multiple layers of testing and QA: engine quality, functional application, VUI, interfaces and infrastructure, load balancing, backup, and recovery. Learn how to set testing goals, targets, and success factors; specify and measure metrics; test and measure “soft” and “immeasurable” targets; test documentation in all stages; manage a testing project; and identify who should be on the testing team.

How do you to achieve a shared understanding with the dozens of stakeholders. How do you predict the things that will be risky or cause scope creep? How do you ensure that the spirit of the project is not lost in an impersonal list of requirements? This session simulates a call center deployment kickoff workshop. At the end of this workshop, you should have the tools you need to design your own project kickoff workshop tailored to your own organization. This workshop draws on the delivery of this workshop dozens of times around the world, Google Ventures Design Sprint, Thought Works' Lean Inception, working in the Google Assistant team, bringing Google Contact Center AI to market, software engineering, product, and design resources that we've found invaluable.