system. The workers can now buy a prepaid con- [ 6], which lets you create your Voicesite just by
nection over the counter, thereby gaining inde- making a phone call.
pendence from the shopkeepers. They get assign- For these micro-business freelancers, a missed
ments by word-of-mouth and through inexpensive call is missed revenue. Now, suppose our free-
advertisements in the local yellow pages. lancers could have their Voicesites—this would
In almost all developing countries around the mean an online presence for them. What if a
world, Internet penetration is much lower than potential client could reach a plumber’s Voicesite
that of the mobile phone, and the rate of increase and schedule an appointment with him? We cre-
of mobile-phone penetration far exceeds that of ated a template for a plumber, which included
the Internet. This fact, coupled with the obvi- questions such as “Enter your welcome message,”
ous preference of speech interfaces over textual “What are your working hours,” and “Would you
ones, led us to the vision of the Telecom Web [ 3, like to mention references for your work?” The
4]. The Telecom Web is a worldwide network of plumber’s answers are recorded by VoiGen and
Voicesites, just as the World Wide Web is a net- used to create a Voicesite so that when a poten-
work of websites. A Voicesite is a voice-driven tial client calls up the plumber, he hears the
application that consists of voice pages (say, plumber’s voice taking the client through various
VoiceXML files) that are hosted in the telecom possible interactions with the Voicesite. The sys-
infrastructure. tem can be set up such that when the plumber is

The Telecom Web exists and operates on the unable to pick up the call, the call is redirected telephony network. People browse Voicesites by to his Voicesite, or alternatively, all calls first get talking with them, traverse from one Voicesite to directed to the Voicesite, and you are connected another via VoiLinks, and even conduct transac- to the plumber only if you need to speak with tions over voice. The Telecom Web figure shows him. VoiGen becomes the equivalent of a “talking several Voicesites connected to each other via HTML editor” for creating a Voicesite.

VoiLinks, which make it possible to move from Just to try this with real targets, we sampled 12 one Voicesite to another by uttering commands freelancers in South Delhi. None of them had ever or keywords. This introduces a “browsing-by- interacted with an IVR before, let alone browsed talking” experience that includes the possibility the Internet. We explained the whole idea of hav-of supporting “back buttons” (“go to the previ- ing a Voicesite to them, and also the mechanism ous Voicesite”), bookmarks, etc. The Voicesites of creating one. Ten out of those 12 were able to can be identified by phone numbers playing the create their Voicesite in under four minutes (this role of URLs. When one traverses a VoiLink to go includes the time it took us to explain things), from one Voicesite to another, this is more than a which means that the concept of a Voicesite and simple call transfer—the context of the conversa- the user interface to create it were reasonably tion also needs to be transferred along with the compelling and intuitive. Two of them could not: call [ 4]. The very first interaction was in a noisy environ-

A common objection to the general acceptance ment, and the user did not have the patience to of such an approach is the frustrating experi- repeat what he was supposed to say. To reduce ence we’ve had so far in using voice applications. noise, the interaction venue was shifted to a car. However, we believe that there is a reason for Another one failed to create his Voicesite because cautious optimism: In already developed regions, he thought he was interacting with a human alternatives to voice have been available, and so at the other end and assumed that free speech expectations are different. For our targets, this would work. will enable them to do things they have never In several parts of the world where Internet been able to do, and by starting out with small access is deep and literacy is not an issue, the applications [ 5], we might find the right way to World Wide Web suffices. There are several ongo-use voice. Just as the proliferation of the World ing efforts to make the Web accessible over voice; Wide Web hinged upon the simplicity of creating the notion of a Telecom Web in such regions is a website (HTML), so will the proliferation of the superfluous. And yet in regions where the tele-Telecom Web depend upon the ease of creation of phony (largely mobile) penetration is far higher, Voicesites. We have built a system called VoiGen and rising faster than Internet penetration, the