New Release Delivers 50% Density Improvements and Up to 15% Accuracy Gains; Positions ScanSoft as a Complete Supplier of ASR and TTS Engines to IVR, Carrier and Integrator Organizations Worldwide

PEABODY, Mass., May 27, 2003 - ScanSoft, Inc. (NASDAQ: SSFT), a leading supplier of imaging, speech and language solutions, today announced the immediate availability of ScanSoft® SpeechPearl™ XML 1.0, a new release of its best-in-class automatic speech recognition (ASR) engine. SpeechPearl XML 1.0 simplifies integrating automatic speech recognition with telephony platforms based on VoiceXML 2.0 and SALT standards and is designed to significantly reduce the cost of development and deployment of those solutions.

The new product release expands ScanSoft's commitment to open standards, and complements ScanSoft RealSpeak, the leading text-to-speech (TTS) solution which already has support for XML. It also follows ScanSoft's earlier announcement that it will make certain of its patents available to the W3C in support of the advancement of VoiceXML and SALT standard efforts.

Developed for a wide range of telephony speech applications, SpeechPearl is a leading software-only ASR solution designed for voice-processing platforms. With highly accurate speech recognition and natural language processing, SpeechPearl is ideally suited for today's increasingly sophisticated call center applications. In addition to providing superior performance on grammar-based tasks, SpeechPearl provides natural, 'say what you mean' commands with its True Dialog™ technology and optimized searching of large lists with its unique FindIt™ features.

SpeechPearl is targeted for system integrators and platform providers that deploy robust natural language speech applications. By incorporating the latest best-of-breed speech technology and offering a world-wide language portfolio, SpeechPearl technology boosts the capabilities and the world-wide reach of those speech providers and enables them to build up a highly competitive speech offering in the market place.

SpeechPearl offers a single, standardized application programming interface (API) that supports the creation of speech-enabled natural language applications in more than 45 languages all around the globe, thus the widest language portfolio in the telephony speech industry.

Increased Languages, Accuracy and Standards Support

To help partners provide their customers with the most accurate and flexible telephony speech solutions, ScanSoft has added a number of key enhancements and capabilities in SpeechPearl XML, including:

• Direct support for VoiceXML and SALT
  SpeechPearl XML is designed for efficiently loading, sharing and resolving the highly dynamic VoiceXML grammar configurations. It has integrated caching techniques for reducing response latency. In the dynamic environment of Web-based VoiceXML applications, the SpeechPearl engine has been optimized for efficiently handling dynamic application scripts. SpeechPearl XML incorporates an optional grammar loader, which can load grammars from file and via http.
  
  
• Support for the W3C SRGS grammar format
  The ScanSoft solution supports the XML version of the W3C speech recognition grammar specification version 1.0 (SRGS) for supporting VoiceXML 2.0.
  
  
• Enhanced out-of-the-box accuracy
  SpeechPearl XML comes with the latest speech recognition engine technology, dramatically enhancing the out-of-the-box performance by up to 50%.
  
  
• Accuracy improvements for tuned applications
  This release further enhances recognition accuracy on tuned applications between 10-15% on large sets of standard applications, thus enhancing recognition quality and robustness.
  
  
• New generation of grapheme to phoneme transcription module
  A new generation of grapheme-to-phoneme technology has been included, leading to a significant improvement in statistical auto-transcription accuracy.
  
  
• Robust channel density and scalability
  ScanSoft's latest speech recognition technology leads to a breakthrough in real-time performance, increasing channel density by a factor of two for typical field applications.
  
  
• Further optimized handling of largest vocabulary up to 2+ million words
  ScanSoft has further improved the handling of large vocabulary sizes, one of the key strengths of SpeechPearl, making it the market leading solution for large vocabulary speech applications such as automated Directory Assistance services.
  
  
• Offering more than 45 languages, the largest language portfolio in the industry
  SpeechPearl XML 1.0 supports more than 45 languages around the globe, including US English, UK English, Indian English, Australian English, Spanish, Catalan, Latin American Spanish, Argentine Spanish, Chile Spanish, Columbian Spanish, Mexican Spanish, Puerto Rico Spanish, US Spanish, German, Austrian German, Luxemburg German, Swiss German, French, Canadian French, Belgian French, Luxemburg French, Swiss French, Portuguese, Brazilian Portuguese, Danish, Dutch, Welsh, Finnish, Flemish, Swedish Finland, Swedish, Norwegian, Italian, Greek, Turkish, Japanese, Mandarin China Mainland, Mandarin Taiwanese, Cantonese, Hungarian, Polish, Russian, Slovakian, Slovenian, Czech and Czech/Slovak.
  
  
• Built-in grammars for VoiceXML 2.0
  The seven standard built-in grammars specified in the VoiceXML 2.0 specification are included in the standard language offering.
  
  
• Toolkit for application workflow
  SpeechPearl XML comes with an advanced toolkit that performs grammar syntax checking, and grammar compiling as well as lexicon generation and an auto-tuning and evaluation tool for tuning the speech recognition parameters.
  
  
• Client/server architecture and resource manager
  SpeechPearl XML comes with a client/server architecture and a resource manager for performing load balancing and distributing the recognition resources over a network of speech ASR servers.
  
  
• Intelligent barge-in capabilities and speech detection
  The SpeechPearl XML SDK is shipped with SpeechDetectorTM, a state-of-the-art voice activity detector for performing robust separation from speech against background noise and for robustly detecting the beginning and end of an utterance. SpeechPearl XML 1.0 offers highly reliable barge-in capabilities, including intelligent, context-based barge-in.

"As call centers continually look for technologies that increase productivity and customer satisfaction, state of the art voice and speech processing applications will need to continually evolve to reduce the time and costs of deployment and integration," said Robert Weideman, chief marketing officer, ScanSoft. "The enhancements and features in SpeechPearl XML demonstrate ScanSoft's continued commitment to the telephony voice application market."

About ScanSoft, Inc.

ScanSoft, Inc. is a leading supplier of imaging, speech and language solutions that are used to automate a wide range of manual processes - saving time, increasing worker productivity and improving customer service. For additional information on the company, please visit www.ScanSoft.com.

Trademark reference: ScanSoft, SpeechPearl, and RealSpeak are registered trademarks or trademarks of ScanSoft, Inc. in the United States and other countries. All other company names or product names referenced herein may be the trademarks of their respective owners.

The statements in this press release that relate to future plans, events or performance are forward-looking statements that involve risks and uncertainties, including risks associated with market trends, competitive factors, and other risks identified in ScanSoft's SEC filings. Actual results, events and performance may differ materially. Readers are cautioned not to put undue reliance on these forward-looking statements that speak only as of the date hereof.

Contact Information