"The SpeechWorks division of ScanSoft has never been in a better position to bring its full speech capabilities to bear on the global enterprise and telecom market," said Daniel Hong, analyst for Voice Business at Datamonitor, a leading research firm. "SpeechWorks has further advanced its strategic relationships with standards-based platform providers by providing them with a truly global and feature-rich speech recognizer to develop and deploy customer self-service solutions worldwide. All of these assets, combined with its global resources and expert professional services team, place the SpeechWorks division in an elevated position in today's speech market."
OpenSpeech Recognizer 3.0 is tuned for speech applications such as auto-attendant, automated directory assistance and natural language customer self-service. It incorporates enhancements to support the specific recognition tasks often employed in these applications:
- SpeakFreely natural language capabilities enable applications such as call routing or customer self-service to respond precisely to callers' naturally phrased requests. SpeakFreely grammars are built automatically from textual examples of what a caller might say along with the meaning of each response. SpeakFreely can shift complex grammar development from speech specialists into the hands of application developers.
- OpenSpeech Recognizer 3.0 includes an expanded pronunciation dictionary in each language pack that provides comprehensive coverage of proper names local to that language. Ideal for applications such as Auto-Attendants, this feature boosts recognition accuracy while simplifying vocabulary maintenance.
- A new grammar file format accelerates compilation of large menu lists commonly used in automated Directory Assistance applications. This new feature speeds the preparation of updated grammars, thereby allowing more timely updates for more accurate listings, resulting in higher automation rates.
- An updated OpenSpeech Insight tool is included to assist with the development and maintenance of speech applications. The latest version offers increased support for analyzing high call volume speech deployments that require distributed call handling systems. With over 50 customizable reports, developers can easily determine which parts of a speech application may need attention to reach higher transaction completion rates and higher caller satisfaction.
OpenSpeech Recognizer 3.0 delivers broad language coverage through a global network of platform partners to serve organizations around the world. Customers and partners can deploy speech applications in many market using OpenSpeech Recognizer's global capabilities, including the following:
- OpenSpeech Recognizer 3.0 provides speech recognition in 44 languages and dialects and additional languages can be developed on a custom basis to meet the needs of specific applications.
- OpenSpeech Recognizer 3.0 offers full multilingual capabilities that allow any combination of languages to be mixed within a single utterance. This feature is valuable in bilingual markets where callers often switch between languages in a single response. For example, in Canada it might be used to seamlessly recognize responses combining English and French.
- Patented LEARN capabilities continually and automatically adapt the recognizer to each specific deployment. LEARN quickly adjusts acoustic models to local dialects, regional accents and telephony channel characteristics. The net result is measurably improved recognition accuracy without any additional effort, even if caller demographics shift over time.
OpenSpeech Recognizer 3.0 creates a unique best-of-breed design, drawing on carefully selected algorithms, development techniques, and components employed in the SpeechPearl® and OpenSpeech Recognizer product lines. This approach delivers a single, feature-rich speech recognition product to service the global speech recognition market, lowering development, sales and support costs for telephony platform developers. Enhanced features of OpenSpeech Recognizer 3.0 include the following:
- Core technology improvements have reduced recognition error rates by an average of 25% on a variety of tasks compared to OpenSpeech Recognizer 2.0. Fewer recognition errors lead to greater caller satisfaction, higher call completion rates and a faster return on investment.
- OpenSpeech Recognizer 3.0 incorporates techniques such as Finite State Transducers (FST) to reduce memory and CPU consumption without compromising recognition accuracy, increasing channel density and so lowering overall system cost. Savings of 90% can be realized for large grammars.
- OpenSpeech Recognizer 3.0 features new algorithms for more accurate confidence scoring, reducing the need to prompt for confirmation of inconclusive recognition results. This reduces individual call times to expand call handling capacity while raising caller satisfaction.
- OpenSpeech Recognizer 3.0 is available in all-in-one and client-server architectures sharing a common API that is backwards compatible with OpenSpeech Recognizer 2.0. In addition, support for MRCP is available from the newly updated SpeechWorks MediaServer 3.0, which also includes Speechify® text-to-speech, providing customers with more deployment options.
Pricing and Availability
OpenSpeech Recognizer 3.0 will be generally available on May 31, 2004 for Microsoft Windows 2000 and Windows 2003, Red Hat Linux 7.2 and Red Hat Advanced Server 2.1 operating systems. Updated support for SPARC Solaris 8 client will be available in July.