So far the award for Single Coolest Thing Done With Open Source at the Web 2.0 show has to go to Ifbyphone. These folks have used open source to make custom telephony applications as easy as designing a Web page. Easier, even.

When I sat down with Ifbyphone's Irv Shapiro, he started off by explaining Ifbyphone this way: "When connecting to the Web got cheap, and anyone could sign up and get a Web site, suddenly there was all this creativity in the Web -- all these developers building stuff for the Web. A lot of that went nowhere, but we also got the whole social networking / Web 2.0 bloom. Thing is, before the barriers fell, a kid in college couldn't have dabbled in this stuff and come up with a hit."

Now consider telephony, where nothing of the kind has really happened yet. Telephony is still closed, expensive, and slow. "If you want, say, a carrier-grade voice recognition system, you'd need thousands of dollars per port. The kid in the dorm, so to speak, isn't doing that."

Enter Ifbyphone, a "cloud telephone app platform." "Anyone who can build a forms-based Web app can now also build a telephony app." The whole system has been constructed along the lines of existing Web services, so it's all a bunch of APIs that use HTTP and XML.

The cost for entry also is absurdly low. Any developer can use 100 minutes of phone time a month just to play around, or charge 8¢ a minute after that to their credit card. (If you buy a contract, you can get as low as 3¼¢ a minute.) It's also available instantly -- no waiting for a port or switch to be provisioned.

The per-minute charge doesn't just include the connectivity, but also access to things like voice-to-text, text-to-voice, call menuing, call forwarding, message forwarding, and tons of other goodies.

"We built all this by taking the best from the world of open source and the best from the world of proprietary, and we glued 'em together in the cloud," Irv explained. "If we had used all of the latter, we couldn't afford to provide this freely to developers. We use Asterisk and OpenSER on the front, and the back end is licensed stuff like the text-to-speech engine. The two are glued together, but with the expensive stuff only used when needed. When your app does an API call, we look at what you have and place every piece of the call on the lowest possible cost components."

There was talk of using open source to do everything, but as it turned out many of the things they wanted (e.g., speech-to-text) weren't up to snuff in the open source world -- yet. By making each piece on the back end pluggable and discrete, they were able to only use the proprietary stuff only when absolutely required -- and retain the freedom to swap it out over time as open source replacements become available.

The core infrastructure for the system is about what you'd expect: a mix of Red Hat Enterprise and CentOS, with some of the proprietary stuff running on Windows. Apparently the Windows components pose the biggest bottleneck because they're all 32-bit only, which creates a real consolidation problem for them. Those pieces are going to be swapped out for open source substitutes as soon as humanly possible, and in fact they're already beta-testing a Linux edition of the same application as we speak. "Today," Irv says, "about 50% to 60% of the calls made through the system never touch a proprietary component."

Irv has coined his own phrase to describe what he's doing: phone mashups. An example:

The p2dir "phone mashup" provides point-to-point driving directions based on start and destination phone numbers. It combines telephony, speech recognition, reverse number, and driving-direction Web services; no Java, GPS, or cell towers needed. The complete source code for this sophisticated mashup is available in the phonemashup.com file repository.

With the cell system finally beginning to get a taste of badly needed freedom, this, plus features like geolocation and all the other Web 2.0 stuff we've already been hearing about, sounds like the way telephony should always have been since the dawn of the mobile age. Or even earlier.