
ON A cold winter鈥檚 day in December 2006, Guruduth Banavar鈥檚 team gathered up some workers at a bustling marketplace in New Delhi, India, and cajoled them, each in turn, into a car.
The team was from the IBM India Research Laboratory (IRL) in New Dehli. They had come to the market to test an alternative to the internet for India鈥檚 rural population. The system is based on the cellphone, though, and so the din of hawkers selling vegetables, and shoppers looking for everything from jewellery to electronics, made conversation impossible.
Once inside the car, however, 10 of the 12 volunteers 鈥 who had never before interacted with a speaking computer 鈥 were able to create their own voice-based website, or VoiceSite, in just under 4 minutes apiece. The first trial of the 鈥渟poken web鈥 was a success.
Advertisement
The spoken web is an attempt to bring the power of the internet to rural India. Despite India鈥檚 highly touted knowledge economy, around 70 per cent of its billion-strong population lives outside its cities, and most of them earn just $4 per day or less. And even for those who can afford computers, attempting to access the internet remains a futile exercise, either because they cannot read or write or the information on the web is simply not relevant to them.
The answer, says Tapan Parikh of the University of California, Berkeley, is speech. 鈥淎n audio format would provide much more access and opportunity for local people to contribute,鈥 he says. 鈥淲hile a farmer may not be able to write a memo, or an email, or a summary of his work, he can easily talk about it.鈥
鈥淲hile a farmer may not be able to write about his work, he can easily talk about it鈥
Nearly 300 million people now use cellphones in India, up from zero a little more than a decade ago. Their use has spread to semi-urban and rural areas, and about 8 million new subscribers sign up each month. 鈥淚n most cases [they] use very low-end mobile phones, not iPhones and BlackBerrys,鈥 says Banavar, director of IBM IRL. 鈥淪o we asked ourselves: how can we enable this vast population to do things on the phone in a manner that does not assume a PC-based internet?鈥
To answer that question, Banavar and his team interviewed workers across India 鈥 including housemaids, milkmen, plumbers, and electricians 鈥 to understand their level of technological literacy. What they found was that while most people used cellphones, they rarely use text messaging or even the address book on their phones. 鈥淭hey would rather call a friend to find another person鈥檚 phone number than actually save the number in their address book,鈥 says Banavar.
It was clear that any new internet-style technology for rural India could not be like the web, which is essentially a literacy-dependent visual interface. 鈥淭hat led to the spoken web,鈥 says Banavar.
Conceptually, the spoken web is a network of VoiceSites, just as the internet is a network of websites. A VoiceSite can only be accessed by a phone, and only requires the user to be able to speak and listen. Callers can create their own VoiceSites or access those of others. They can also surf the spoken web, jumping from VoiceSite to VoiceSite using speech.
In developing the system, the team took their cues from the growth of the internet. 鈥淭he thing that really fuelled the adoption of the World Wide Web is the ease with which you can create websites,鈥 says IBM IRL鈥檚 Amit Nanavati. So they focused on developing a simple way to create VoiceSites.
Say a plumber wants new customers. First he calls a number and software called VoiGen then guides him, in his local language, through the process of setting up a VoiceSite. Relevant information, such as a welcome greeting and the plumber鈥檚 contact details are recorded. Behind the scenes, the VoiGen system then creates a VoiceSite. A phone number, analogous to a URL, is then assigned to the plumber.
Anyone who calls the plumber鈥檚 VoiceSite number is greeted with his welcome message, and then given help to navigate the information. For example, the caller can say 鈥渁ddress鈥 to hear the plumber鈥檚 address, or 鈥渁ppointment鈥 to request a call-out with him.
The way a caller navigates the VoiceSite is based on a template, and the team has created templates for different types of site, such as advertising wares or conducting an auction.
A caller鈥檚 experience of an individual VoiceSite is similar to the interactive voice response (IVR) systems that customers encounter when calling, say, an airline or their bank.
However, where the spoken web differs from these systems is that different VoiceSites can be linked, just like in the internet. To create links, VoiGen prompts the user at predetermined points to provide the phone number and a brief description of related VoiceSites 鈥 of a friend鈥檚 hardware store, say. Then, a caller to the plumber鈥檚 site who chooses to listen to the plumber鈥檚 recommended links will hear the description and can press a key or say a word to be transferred to the hardware store鈥檚 VoiceSite. 鈥淭his takes you beyond the IVR stuff,鈥 says Nanavati (see 鈥淕o with the audio flow鈥).
To manage these transfers, IBM has developed a new protocol, hyperspeech transfer protocol (HSTP), which is analogous to the hypertext transfer protocol (HTTP) that takes people from one website to another.
Banavar says the team avoided the use of complex voice recognition software. 鈥淚magine India, with the number of languages, dialects and number of different demographics鈥 Forget it, we would never be able to address it,鈥 he says. Instead, the spoken web relies on a very small vocabulary and the structured interaction means the system always knows the context of the spoken words.
Preliminary studies of the system have already thrown up some surprises. 鈥淥nce you give people the capability to start putting information on these sites, their imagination starts running quite wild,鈥 says Banavar. For instance, people wanted to use VoiceSites for matchmaking, or for negative advertising 鈥 asking callers not to believe the information on another VoiceSite, he says.
Despite such unexpected usage, Banavar thinks that the spoken web will serve local communities well. For instance, an electrician who cannot afford his own storefront can use a VoiceSite to attract customers and set up appointments, and so not miss out on business when he is out working; villagers can check a VoiceSite to find out if the mobile hospital is coming to their village on any given day; a farmer could set up a VoiceSite to sell his daily produce to anyone within a 50-kilometre radius. 鈥淭he farmer can get the best price, as opposed to the current model, where they have one broker who does everything the way he likes. The broker maximises his profits and the farmer is left in the dirt,鈥 says Banavar.
Parikh of UC Berkeley, who has been working with the IBM team, is using the spoken web technology for a pilot project in the state of Gujarat in western India. The local radio station archives its content on a VoiceSite, which villagers can access and leave comments on. Others can listen to these comments, and future broadcasts incorporate them into their discussion. 鈥淲e are using the spoken web tool to make community radio more interactive,鈥 says Parikh.
With increasing cellphone coverage, a technology like the spoken web could prove to be a valuable resource for the developing world, he says. 鈥淐onventional approaches have only looked at taking the existing web and making it available on mobile devices,鈥 he says. 鈥淭his is an opportunity for making an entirely new kind of web.鈥
Go with the audio flow
If the spoken web is to become a true internet on your cellphone, users will need to be able to 鈥渟urf鈥 easily through different sites.
So a team at the IBM India Research Laboratory in New Dehli is working on a simple audio browser to make this possible. The browser is a special VoiceSite that maintains a directory of all other VoiceSites and is the first number that the user calls. As the user moves through the spoken web, the browser creates a record of their path. This represents their browsing history, which allows the system to respond to requests like 鈥済o back鈥 or 鈥済o forward鈥. This history is only stored for the duration of the call.
A permanent record is kept of the user鈥檚 bookmarks. At any time during the interaction with the spoken web, the user can say 鈥渂ookmark鈥, which will add the site to their database.
Making the spoken web as easy to browse as the internet will be challenging, though, as audio is much harder to scan, search and index than text. 鈥淪ince audio information is laid out in time, it is difficult to quickly find relevant information or scan over a document, or pieces of content,鈥 says Tapan Parikh of the University of California, Berkeley. With some technical advances, it should be possible, he says.