The Premises:QuickTake, conductor and compiler of online surveys.
Entering survey questions and managing target lists online, getting instant feedback
- NetByTel's (Boca Raton, FL - 561-988-5050) ASP speech platform
- Intel/Dialogic's (Parsippany, NJ - 973-993-3000) voice and Speech DSP cards
- SpeechWorks' (Boston - 617-428-4444) speech recognition, VoiceXML interpreter, and Speechify Text-to-Speech
- Custom C, C++ Programming
If we accept VoiceXML as the HTML of voice applications, it takes very little imagination and knowledge of web history to picture the next step: a markup language further simplified into templates for generic voice applications. Remember that tools such as Microsoft Front Page and Macromedia's Dreamweaver quickly took the pain (and some of the control) out of HTML and went further, releasing wizards and templates for sample websites (event sites, company sites, personal pages, etc.). The hosts themselves, such as Earthlink, GeoCities, and Yahoo, did something similar on an ASP basis.
To see how this model begins to play out in the speech-hosting universe, check out NetByTel.
Voice-web host NetByTel is not yet using VoiceXML. They're now working on integrating SpeechWorks' VoiceXML interpreter into their platform, but for now, their template applications are written in C and C++. But like the website templates on the HTML visual side, NetByTel has gone VoiceXML one better by offering fill-in-the-blank CT applications on its hosted platform.
A basic e-commerce, order-by-number application is a template-filling proposition. Input the client's product database, fill in the prompts, and send the ordering data back via XML to the client's back-end processors. And provide the speech interface: NetByTel relies on SpeechWorks' speech recognition, and is in the process of moving from L&H text-to-speech to SpeechWorks' Speechify. It backs up its ASR with touch-tone and live-agent options.
I sampled the ordering application used by Office Depot, which allows account customers to order products by spoken number; the same number that appears on its e-commerce website. It's realizing the voice-web dream of voice-enabling web-resident data. NetByTel customized several modules from both its CRM and revenue-generating suites for Office Depot in a matter of months.
For the XML-challenged, NetByTel will perform ODBC-XML conversions. They've also done screen scrapes from HTML agents, a more kludgy way of voice-enabling the web experience. Looking further ahead, "We will do VoiceXML where we find a client who has VoiceXML-enabled their site," comments Dewey Anderson, CTO of NetByTel .
An even more forward-looking template application, however, is an IVR already familiar to readers of this column: survey-taking.
Here is the web-template IVR idea taken to its furthest, most user-friendly extreme: Surf to the template site, enter your questions and possible answers, click to show branching, input your outdial list, give your credit card number, and press Submit. The website sends the entered info via XML to NetByTel's voice platform, whose custom programming takes over. It turns questions into TTS-delivered speech and possible answers into more TTS for prompts and recognition grammars to understand spoken replies. Phones start ringing, data starts accumulating, XML flows back to the survey server, and bar charts of real-time results start publishing themselves on the client's password-protected web page.
This is the latest service being offered, via NetByTel, by QuickTake (Wilton, CT - 203-834-8585). QuickTake, launched in February 2000, is the e-survey offering of Greenfield Online, a full-service market research firm. QuickTake's chief proposition is still text-based surveys conducted via the web, which are brought to respondents' attention via email links. QuickTake clients design surveys online (with professional help from QuickTake). They use radio buttons or check boxes, design question flows, and input lists of email addresses or pager numbers that stay resident for future surveys. After the survey is launched, they get web-based, real-time compiled results.
The voice-enabled survey offering is still in customer trial, but it is presented as just one of three choices of survey media: In addition to the "traditional" HTML-based, questions can be conveyed in text via two-way pager, cell phone, or PDA; and in voice.
The template changes with the medium chosen: An HTML-based survey, for example, can augment its questions with charts and graphics and present much longer questions and answer choices. ASR-dependent surveys limit possible replies to 30 characters, since each a), b), c), and d) must be compiled into recognition vocabularies.
QuickTake will also divide its voice offering into three levels of automation: The cheapest way is to enter your questions in text to be delivered through TTS. This is not recommended unless you have a cooperative target list of those willing to listen to synthesized speech, which, no matter how mellifluous, will never sound as clear or warm as human speech. Second, you as survey maker can dial into a number and record your questions yourself. Third, you can pay them to record your questions professionally, even by your company spokesperson, say.
NetByTel attaches an 800 DNIS number to the application for an inbound survey (sometimes using the clients' pre-existing 800 number), or receives an outdial list for outbound survey. The Voice ASP's cross-application modules use automated agents to interact through the Internet with its customers' infrastructure via XML.
For v-commerce, it accesses the client's order-processing back end. For now, plain Internet connectivity suffices for the QuickTake beta testers, though they hope this will justify dedicated lines in the future.
TTS for Advisory Board
The market research client closest to speech-enabled IVR that QuickTake could tell me about was a farm equipment manufacturer, which had assembled an "advisory board" of farmers and given them two-way pagers. A good bet for wireless, it lets farmers read questions on their pagers and reply while in the field.
Under development is a pharmaceutical survey that would replace the diaries commonly used by patients in drug trials. Arthritis patients - presumably an older, less-email-accessible population - can call in and answer questions on drug reactions; questions that the company can easily modify from day to day, as results come in.
NetByTel's platform in Boca Raton relies on Dual Pentium Catalyst servers, into which three DS3s are broken into T-1s, using Adtran multiplexers. Now running on NT with Intel/Dialogic D/240 and D/480 voice cards and Antares speech DSP platforms, plans are to move to Oracle on a redundant Solaris configuration for ASR production databases, data mining and reporting, and Linux or Solaris for the front end, to use existing speech hardware. The web server itself will move from NT to a Unix-based Apache server.