A well-planned automation road map can guide key decision makers to make fiscally responsible funding decisions. This presentation will provide a proven methodology for creating a self-service automation road map by identifying a list of candidate applications, quantifying the benefits using ROI models and prioritizing these applications into small manageable releases. You will also learn how to create a customer-centric adoption strategy that will get your callers ready, excited, and motivated to use your self-service applications.

Enterprises turn to speech self-service in a recession to increase customer service and reduce costs. Be wary of speech crimes being committed by vendors that lock you into high-risk projects, inflexible applications, and charges for application changes that will bleed your budget. Crime scene investigation will reveal: 1) top 10 speech crimes and the clues to prevent them; 2) most-requested speech applications in a recession; and 3) customer service survival strategies.

Traditional customer contact methods treat most customers the same under most circumstances. Some enterprises have sophisticated data warehouses and customer segmentation schemes, yet this information is not leveraged effectively in the customer contact environment. In the future, enterprises that fail to leverage what they know about customers stand to lose market share to competitors who take a smarter, data-driven, personalized approach to speech applications and other automated customer contact methods. This session will discuss key components of advanced customer personalization for speech and other applications.

Technology is becoming increasingly more targeted toward users, incorporating ancillary technology and data, and using word spotting and data mining techniques to specifically target users. This trend continues to grow within speech applications, to more customer-direct applications. This presentation will examine adaptive VUI based on frequency of use and adjusting user experience. This is based on research and will look at personalization and supporting technologies, VUI techniques for becoming more personal, and user responses to increased familiarity and when to maintain distance.

We talk of usable solutions, but usable to who? How usable should these solutions be based on constrained budgets and crashed project timelines? Balancing such deployment pressures against usability best practices in a time of economic downturn create tension between delivering an application that serves the business's self-interest while still making it inherently usable to the caller. Providers and stakeholders should remember that these solutions must be designed to solve the problems of everyday users to meet corporate objectives and put the correct perspective on the who in usability.

Companies are spending more time talking to customers they don't want to talk to (low-value, frequent-contact, costly customers) and not enough time talking with those they do (affluent, time-constrained customers). This presentation will explore how your speech-enabled automated phone system can be instrumented in supporting value-based customer segmentation and contact center resource alignment to extract costs, drive revenue, and deepen valuable customer relationships.

In an effort to increase customer service, many business initiatives are emerging toward proactive engagement of customers by reaching out to them. A feature-rich, proactive solution must first build a trusting relationship with the customer, and then customers will participate in the interaction. Technologies may assist-with speaker verification leading the way-but simple, well-designed user interfaces are also important. This session will explore ways to leverage outbound self-service in a manner that customers will trust and use.

In today's economic climate, arrears and collections management can be an ongoing battle for businesses. Offering a proactive over-the-phone system where the customer can be reminded of a missed payment, and then offered the opportunity to make a payment, broadens the options for the customer and increases the chances of receiving payments. In addition, it offers the caller an alternative to engaging with an agent on what can sometimes be a tricky personal topic. But will customers willingly engage with the outbound call? Vicorp conducted a survey to address business and consumer attitudes toward outbound calls with the aim of providing guidance to those considering this technology.

Speech community: This is a call to action! There is no apparent standard for designing, testing, and developing voice user interfaces, nor any strong consensus as to what would constitute a complete methodology. Non-VUI standards are incomplete and in many ways irrelevant to VUI design. This presentation will address the problems of nonstandardized VUI development methodologies and encourage movement toward standardization. Audience participation and feedback will be specifically encouraged.

Over the life of an IVR project, team members often change and vital history is soon forgotten. The solution? Style guides. For designers, this document details the style of that IVR to keep the “feel” consistent, even if the designers change. This presentation will explore when style guides should be created, what's important to put in them, and explain the guide's benefits. We'll also explore having similar documents for development and speech science as well.

Voice user interaction and interface have struggled with definitions of appropriate approaches and starting points. However, other design practices have organized responses to the issues. We have few due to a lack of extensive formal education and continued struggles with internal debates and uncertainties. One practiced and valid approach is to create design patterns for commonly encountered situations to create a library of resources to help quickly form design foundations. More than recipes or magic spells, design pattern collections provide structures, approaches, and rationales so the designer can select a starting point for customized creativity.

Designers often rely on accumulated peer experiences and designs of projects of similar nature. But what about designs that need to be leveraged and reused by the same customer for other subsidiaries or that cater to a wide variety of customers? Can user requirements be captured the standard way? This session will explore the details of reusable design in practice and types of techniques available (such as patterns, modules, etc.), including their unique challenges and impact on the field.

The majority of IVR systems in production today use a female voice. As listeners age, however, they typically lose their high frequency hearing first. For those with presbycusia, this results in a relative difficulty in understanding female voices, as compared to male voices, which typically have lower formant and fundamental frequencies. The goal of this presentation is to make voice user interface designers aware of the impact of voice talent selection on intelligibility for some user populations.

As cell phones and other conversational technologies have become invisibly pervasive, privacy has become a serious issue. For example, if you need to give your account number or name in a public place using your cell phone or some other conversational technology your identity may be compromised by a nearby eavesdropper. This presentation will demonstrate an approach for anonymous spelling that has been implemented in an electronic voting system for candidate write-ins using voice.

IVR design can and should be based on well-established dramatic structures. Working from the top down, a VUI designer should first explicitly define the overall objective of the application, then each smaller and more granular structure in turn. As more detail is added to the design, the dictates of the parent structure are honored and amplified. An existing application will be rewritten according to these principles and analyzed by Jon Bloom in the next presentation.

Voice interaction designers often evaluate their work using high-level metrics such as automation rates and call duration. But what about deeper statistical analyses? We will present five design changes that were put into production for a high-speed internet troubleshooter, review the methods used to analyze the changes, and discuss their performance. One of the design changes we assess will be that presented by Peter Krogh in the previous presentation.

VUI designers are often under pressure to include bells and whistles in their opening prompts. These add-ons include the simple mention of an agent, branding, or basic instruction on using the system. What happens to caller experience when these are removed? This session will chronicle the transformation of an opening prompt believed to be an informative, branded, and reassuring greeting to a three-word introduction to the application. We will discuss the effects on caller experience through the presentation of data collected from production systems.

The work of Reeves and Nass showed that the reactions that people have to automated systems are similar to the reactions they have to real people exhibiting similar behavior. So why do we expect callers to want to interact with IVRs that demonstrate anti-social behavior? This talk will focus on some of the many ways that typical IVR behavior might be considered anti-social if coming from a real person-telling callers what to do, not following turn-taking rules, always talking about everything from its perspective rather than ours-and will provide field-tested suggestions on how to make the human-IVR interaction more closely fit human expectations of conversations.

The car is becoming an increasingly complex environment, with more connected and in-dash devices. New interactions are being designed, including multimodal voice and touch interactive user experiences, as well as voice-only experiences. This talk will cover ideas that help understand the car environment, evaluate some examples of where basic voice design principles are not being followed, and look at some of the new design principles emerging in this space.

The in-vehicle environment presents interesting challenges to easily and safely access and manipulate personal information. This presentation focuses on an innovative application of speech technology in a platform designed specifically for in-vehicle use. This voice-driven interactive platform enables the driver to focus on the road ahead while remaining informed and connected through email, text messages, telephony, and other common forms of communication. Relevant usability and safety issues in real-world scenarios are highlighted in this customer case study in addition to covering scalability considerations for interactive in-vehicle use of speech technologies.

Most commercial voice applications employ a form of finite state control mechanism for dialogue management. However, there are classes of applications where such an approach is impractical because of the complexity of the task or because of a caller’s limited understanding of the domain. Such is the case in our application involving diabetes management, where we are employing a rules-based approach for dialogue management derived from the domain itself.

The VoiceXML Tools Committee has been studying methods and architectures to support next-generation conversational systems-machines that appear intelligent to the user. Our objective is to find ways to make services of tomorrow scalable, maintainable, compatible with emerging standards, and able to be designed at a reasonable cost. This presentation will review proposed architectures and algorithms and will relate examples of current systems with humanlike reasoning abilities.

This presentation describes a research project known as Mobile Adaptive Assistance (MAA) that deals with human-machine interaction within a networked, intelligent environment. The system unifies the user interfaces of various devices and provides a VUI with the goal of allowing the user easy and natural communication. To process user requests accurately, the system uses knowledge of the current situation and the environment, linguistic analysis of user statements, and use of multiple levels of abstraction of the real world to link to the eventual action on real devices. Although this project is still in the development phase, it demonstrates that speech could be the ideal interface for ubiquitous computing!

Multimodal systems are now being used in a variety of situations ranging from quiet offices to walking down the street to driving in a car. These differing situations will bias users to rely more or less on each available input and output modality. This talk will address the key decisions around the design of systems to be used in these environments and will include results from Vlingo's in-field deployments.

This session will explain ASR (automatic speech recognizer) confidence rejection thresholds: what they are, where they come from, and their criticality to your ASR-enabled IVR. We describe the steps necessary to optimize this important threshold value throughout your application, covering transcription, the importance of grammar coverage, and an explanation of terms such as the Equal Error Rate. This session is ideal for those ready to take their ASR-enabled IVR tuning to the next level.

Statistical spoken language understanding grammars (SSLUs) are often used only at the top recognition contexts of modern large-scale spoken dialogue systems. We propose to use SSLUs at every recognition context and present a methodology of continuous improvement in which data are collected at every recognition context over an entire dialogue system. These data are used to automatically generate updated context-specific SSLUs at regular intervals and thus continually improve system performance over time. We have found that SSLUs significantly and consistently outperform even the most carefully designed rule-based grammars in a wide range of contexts in a corpus of more than 2 million utterances collected for a complex call-routing and troubleshooting dialogue system.

Speech-to-speech translation technology has advanced to an interesting level in recent years. It is now possible to monitor foreign media in real time without language barriers. We even see applications, customized for military use, which allows bidirectional speech-to-speech translation in real time in a little devise using limited vocabulary. This presentation will show recent advances in the technology with an emphasis on crowd-sourcing (learning from live feedback) and real-time issues.

Speakers' mental states are conveyed by acoustic/prosodic cues as well as words and gestures. Much research has been done to detect anger/frustration in IVR systems so upset customers can be redirected to a person. More recent research focuses on recognizing other speaker states, such as confidence/uncertainty in tutoring systems and emotions accompanying deception. This session will describe the current state of "speaker state" research and discuss how close it is to commercial application.

This session is an introduction to VoiceXML 3. Subjects will include an introduction to the V3 architecture, relationship to SCXML, new speech functionality (i.e., SIV, VCR controls), as well as profiles and modularity.

Leumi, the second largest bank in Israel, is the first deploy text-independent voice biometrics technology. This case study will focus on the recent rollout of a unique fraudsters' detection functionality- enabling the bank to identify known fraudsters as they interact with the bank's agents. The solution, which is based on PerSay FreeSpeech platform, utilizes unique techniques that minimize false-alarm rate and increase detection accuracy. Accuracy rates observed in large scale fraud attacks will be presented.

This session will cover the work being done in several standards bodies, including the results of the W3C's recent workshop on speaker identification and verification (SIV). It will also cover the collaboration between the VoiceXML Forum and the American National Standards Institute (ANSI) on a data interchange format, the data interchange format being developed by the International Standards Organization (ISO), and the SIV in the second version of the Media Resources Control Protocol V2 (MRCP V2) by the Internet Enterprise Technology Forum (IETF).

The proliferation of mobile devices and advances in speech processing are enabling a new and exciting set of voice applications that provide users the ability to adopt natural voice interfaces to quickly and easily find information, whether searching for videos, music, and business listings, or surfing the web. this presentation will review the technologies and challenges behind mobile voice search, and report on case studies and performance statistics from AT&T's business listing Speak4it application.

Based on 15+ years of research and development in Arabic natural language processing (NLP), the Sakhr Mobile S2S Translator is the world's first and only live mobile Speech-to-Speech (S2S) application. The Sakhr Mobile S2S Translator is easy to use, yet provides instant and accurate translations - anywhere you want. Speak English or Arabic into a mobile phone like the iPhone or Blackberry, hear the audio translation, see the text translation. The application leverages server-based speech processing and cloud computing.

One of the fundamental measures for a successful deployment of a call steering application is maximizing the accurate routing of callers so callers can use self-service completely in the IVR. In this session Medco Health will present a close look at deploying, maintaining, and measuring success rates of a call steering application. We will also present research on call steering with natural language speech recognition versus live-agent assisted routing. Comparison of performance statistics and of productivity gain/loss between the two methodologies will be presented.

Designing, building and tuning large scale, enterprise speech applications is a complex task. When you throw natural language (NL) speech into the mix, the complexity can increase several-fold. However, complexity can be managed with rigorous, comprehensive project plans, strong project leadership, and thorough understanding of key components of the project life cycle of an NL application. This presentation will cover key processes for a successful NL project, potential issues and pitfalls to avoid, and will give real examples from some of the country's largest live production NL deployments.

This presentation will focus on the ways in which speech technology has been leveraged to transform the Australia Post NSW/ACT CCC over 18 months from a reactive call processing center to a proactive service and revenue-generating focused center. All voice services are now managed in real time via a web portal, which is managed by the call center administrators. The call center is using a mixture of inbound and outbound services which it can configure and turn on and off as required. We will show how this technology has allowed us to focus on revenue-generating activities and describe automation rates and customer satisfaction results achieved.

Milwaukee-based Integrity Construction Group faced a problem common to many modern businesses: Despite having a great phone system loaded with features, it couldn't take advantage of many of them without relying on a complicated cheat sheet. With about 80 users-plus an IVR and call center that allows customers to check job status in real time-remembering the sequence of keys required to transfer calls between extensions at four different locations proved impossible without the aid of cheat sheets taped to phones. By partnering with Schmooze Communications, Integrity Construction was able to solve these problems and increase its productivity. This has allowed for increased efficiency and productivity in every aspect of its business.

As voice and self-service applications move increasingly into the cloud and to IP communications, what do you need to be concerned about with regard to the security of hosted solutions? If you grow to trust the cloud, how can you be sure it will be there for you? What protections can you put in place? What backup plans can you establish? What questions should you ask potential hosted/cloud vendors? In this session, security professional Dan York will walk you through the basic risk areas of voice-over-IP security, explain how those relate to both hosted and hybrid configurations and leave you with a concrete list of questions to consider in considering hosted/cloud options.

Creating an automated access control system that will perform as good as-or better than-a human guard is challenging, especially when the system needs to be both preventive and user-friendly. We selected voice, as it is the most natural human machine interface while concurrently providing biometric information on the speaker. The fusion of speech recognition and speaker verification creates the ability to know who you are.