Automatic Speech Recognition Fine-Tunes Self-Service.

Imagine you are a customer and you have just completed a call with your insurance company. For the first time you can recall, you hang up smiling. You have received all the requested information about your policy quickly, without spending an eternity on hold or having to bear those long, annoying touch-tone menus -- and the call was completed without ever talking to a live agent.

Does this sound like science fiction? Welcome to the new millennium, where more of the classic touch-tone IVR systems that have dominated the enterprise for 30 or more years are being provided with major face lifts. More to the point, perhaps it's a new set of vocal chords.

Voice recognition technology has come of age and, when integrated into business-critical systems such as IVR and office automation systems, it can provide a new level of service at a surprisingly reasonable cost. Replacing functions such as the basic hierarchical dual-tone multi-frequency (DTMF) menus and complex dialogs employing natural language understanding, automatic speech recognition (ASR) is finding its way into enterprises conducting e-business and carriers deploying voice-activated dialing and automated directory assistance.

Thanks to language modeling, sophisticated grammars and accuracy tuning tools, speaker-independent ASR engines can attain an accuracy rate of 97 percent or better, rivaling that of a live agent. Combined with natural language understanding, this allows callers to navigate an application without having to follow a strict menu structure common in a typical IVR system. For instance, a caller who wants to transfer $100 between his or her bank accounts need not listen to a series of prompts such as "for transfer, press one" and "for checking account, press two."

All that is required is a verbal caller request such as, "I'd like to transfer $100 from my savings account to my checking account, please." The application responds by prompting the caller to articulate the account number and, upon validation, handles the transaction appropriately.

ASR is only part of this remarkable achievement. Text-to-speech (TTS), an application that uses basic computer ASCII text and simulates speech, has produced quality very close to the natural human voice. This human-like automated response allows callers to listen and understand with ease rather than struggle through the tedious, monotone sound that has been the hallmark of TTS for 35 or more years. TTS technology is capable of simulating actual speech while maintaining the appropriate prosody, speed, voice inflections and other characteristics that are important to human communication.

While businesses experiment with voice technology, the industry itself has accelerated the development efforts of linguists, dia log designers and speech technology engineers to create a broader selection of vastly improved products. To aid this development effort, tools are available that range from grammar and vocabulary to call flow and dialog design, enabling the creation of extensive, complex and accurate applications. Additionally, speaker verification, which is a biometric technology, provides unsurpassed security over the telephone when coupled with traditional passwords or account numbers.

In this global business environment, supporting multiple languages is vital. Currently, most of the languages spoken in North and South America, Western and Eastern Europe and Asia are supported through speech recognition, while developing areas such as India, the Middle East and Africa are either available or in development.

VoiceXML:

The Emerging Standard

Voice technology is bridging yet another gap to access the enormous expanse of information contained on the Web. VoiceXML, a scripting language born of the same family as HTML, allows voice applications to be served up to speech browsers in the same way that HTML pages are served up to the traditional Web browsers. The similarity of VoiceXML to HTML makes it easy for developers to create voice applications that can leverage the existing Web infrastructure and enable companies to use existing investments with voice access to information on the Web.

VoiceXML is a particularly compelling, emerging technology for voice applications and it represents the first potential standard for voice applications. Originating with the VoiceXML Forum, a consortium of companies that includes Motorola, IBM, Lucent and AT&T, the VoiceXML standard is now the responsibility of the Worldwide Web Consortium (W3C), an organization with a long record of establishing technological standards. The involvement of the W3C, coupled with widespread developer acceptance of the VoiceXML specification, will promote interoperability among diverse voice applications and businesses around the globe.

Connecting The Voice Application

Voice recognition over the telephone has created some challenges for telephony equipment vendors. Often when a caller's voice is transmitted over the traditional public switched telephone networks (PSTN) to an ASR engine, it can be garbled and difficult to understand. Additionally, satellite repeaters and other telephone equipment can introduce echo, static or noise, which negatively affect the accuracy rate of the speech recognition engine.

Voice over IP (VoIP) technology has become prevalent in large enterprises and when introduced, can cause packet loss, latency and jitter that affect the voice sample. To counter these negative effects, equipment vendors are designing telephone network interfaces that provide superior echo cancellation, noise filters, jitter buffers and caching to improve voice quality and deliver excellent speech recognition. The accuracy of the ASR, of course, remains the major factor in user adoption of voice applications.

The Session Initiation Protocol (SIP) standard is also emerging throughout VoIP networks to handle call control in a distributed network. Easier to use and more flexible than other protocols, such as H.323 and Megaco, SIP is becoming the preferred protocol for voice application developers. SIP has won widespread adoption by high-profile organizations such as Microsoft, which has integrated the SIP standard into its latest operating system.

Adoption Is The Key

Voice recognition over the telephone has reached a critical mass, demonstrated by the response of businesses to dramatically increased end user adoption. For service providers, voice applications provide a competitive differentiation that can drive revenue. Additional benefits to wireless carriers include promotion of the safe use of cell phones while driving and the increase of usable "minutes" that the carriers sell on their networks. Carriers recognize that speech technology represents more than another enhanced service. Speech recognition improves the usability of existing services and allows for the expansion of new, revenue-generating applications, such as instant conferencing and instant messaging services.

An increase in employee productivity, customer satisfaction, sales automation and more efficient service centers all contribute to the bottom line of any enterprise. Currently, there are a wide range of revenue-generating voice applications available that include customer relationship management systems (CRM), sales automation, e-business systems such as stock trading and voice banking, and more sophisticated interactive voice response (IVR) systems.

IVR vendors are now scrambling to integrate speech recognition into their products, propelled largely by competitive pressures, although this differentiation could be short-lived once it becomes commonplace in the IVR. Many IVRs have reached a limit on functionality constrained by DTMF. Fortunately, speech can broaden the services that an IVR can provide.

Voice recognition and voice-enabled applications have hit the mainstream, and we should expect to see an explosive growth of these types of services now and in the near future.

Steve Parsons is director of product management for the New Network Services division of NMS Communications (formerly Natural MicroSystems, www.nmss.com). In this position he is responsible for product marketing and management of HearSay, the company's high-density voice portal platform, integrating NMS telephony hardware with best-of-breed speech products.