Nuance Unveils Nuance Recognizer, A Quantum Leap for Speech Recognition

Business Wire, April 24, 2007

New Release Delivers Unprecedented Performance, Manageability and Security; Transforms Caller Experience for Customer Care and Voice Search

BURLINGTON, Mass. -- Nuance Communications, Inc. (Nasdaq: NUAN), a leading provider of speech solutions, today unveiled Nuance[R] Recognizer v9, a new speech recognition engine that dramatically increases the performance and quality of speech-based customer care and voice search solutions. This development project, code named Quantum, culminates years of intense technology development and innovation in speech and combines the core values of six speech recognition technologies encompassing more than 400 person-years of research and development.

Nuance Recognizer delivers breakthrough recognition accuracy, enabling a more natural, flexible user interface that gives users increasingly unique and personalized experiences in an automated environment. With improved acoustic models and the industry's most advanced natural language support, Nuance Recognizer excels in handling a broad range of caller demographics in challenging applications with extremely large vocabularies, such as telephony customer self-service, directory services and mobile search.

"Nuance Recognizer signals a new era of performance for speech solutions," said Steve Chambers, president, speech division, Nuance. "Nuance Recognizer delivers the core technology and innovation breakthroughs that will enable improved caller experiences beyond anything the market has seen. We took the industry's leading speech recognition products and merged them to produce an even better successor. Based on benchmarks and testing against other solutions, we have exceeded our highest expectations."

By combining the best elements of the leading speech recognition engines brought together by Nuance over the past five years -- along with entirely new technologies developed in Nuance research labs -- Nuance Recognizer delivers unprecedented accuracy, with an average 27 percent error reduction rate across multiple tasks and languages over its predecessors. Fewer recognition errors lead to higher call completion rates and greater caller satisfaction. In addition, superior speech detection techniques employed by Nuance Recognizer offer a 45 percent improvement over previous versions, helping to ensure speech applications perform well for mobile callers in noisy environments, an increasingly common scenario.

Nuance Recognizer is designed to understand what users mean, not just what they say. The solution showcases major advances in semantic and language modeling to improve the ability to interpret a caller's open-ended response, as well as the enhanced capability of extracting multiple pieces of relevant information from that response. For example, if a caller states, "I'm moving to Atlanta and need a new number for my cell phone," the system can decipher several pieces of information from the caller's single statement (i.e., that the caller is moving, where the caller is moving and which account needs a new number). This capability reduces the number of follow-up questions needed to complete a transaction. Since up to 20 percent of retries and confirmations are from callers providing too much information at once, these advanced natural language capabilities deliver significant performance improvements, including increased automation rates and decreased call length.

"Nuance is responding to a marketplace that is becoming increasingly sophisticated," said Judith Markowitz, president of Judith Markowitz Consultants, and a recognized thought leader in the speech recognition industry. "Certainly, error reduction is welcome, as always, but the most significant improvements are those that move Nuance Recognizer beyond pure speech recognition and into the realm of spoken language understanding. This includes semantic language modeling and other enhancements related to the handling of unconstrained speech. These are the kinds of things that the marketplace is demanding, and these are the kinds of capabilities that are part of Nuance Recognizer."

Significant new features and technological advances of Nuance Recognizer include the following:

* Unprecedented Accuracy and Performance -- Nuance Recognizer delivers an average 27 percent error reduction across multiple tasks and language, and an average 45 percent improvement in detecting speech in noisy environments, particularly important for mobile callers.

* Natural Language Understanding -- Speech applications that incorporate open-ended questions are becoming mainstream, particularly for call steering, where a prompt such as "What can I help you with today?" is used to route calls intuitively among self-service systems or skilled live agents. Nuance Recognizer offers unique language understanding capabilities to make building such applications easier, even when processing lengthy responses to identify multiple meaningful phrases. Capturing more information from fewer responses streamlines the dialog and supports a sophisticated user interface.