A behind-the-scenes look at new apps under development in the vendor test labs.
VoIP can save money (maybe), simplify network administration (maybe) and improve productivity (maybe), but let’s face it, a major reason people are interested is that VoIP has the potential to do some really cool stuff.
Like figuring out where you are, how you are accessible and whether the caller is important enough to interrupt you.
Like determining whether you are who you say you are before granting you natural-speech access to data such as your bank account information.
Like giving you intercom-like connections to just about everybody so when you want to reach someone, you just start talking.
These are just some of the new features being developed in VoIP labs. Just wait awhile and your vendors and service providers will be offering them.
Nortel builds caller-ID bot
Nortel is creating a trainable system that over time learns more about how end users want to handle calls, according to Phil Edholm, CTO and chief architect, enterprise networks at Nortel.
Today, you can use caller ID to determine who is calling and whether you want to answer the phone, but that requires stopping what you are doing, looking at the phone and then picking it up or not. What if the phone network could act like a well-informed administrative assistant and intelligently decide for you what to do with a call? Using Session Initiation Protocol (SIP ) and a server-based user software agent, Nortel is working on just that.
If a user is listed on a SIP presence server as available at a certain phone number and gets a call from a customer who represents a $1 million account, the call agent would put that call through. If, however, the agent determines the call is from an annoying router salesman, it can be dropped to voice mail.
After handling the call, the software agent queries the user to find out if it acted properly. It might ask the user to rank the importance of the caller, whether the call should have been put through, whether time of day was a factor, or whether the user’s activity at the time was a factor.
“You get an IM from your personal agent, and it asks you to classify the individual so it can understand where you lump that person and then how the call was handled,” Edholm says. “It builds a database of how you thought the agent handled the caller at different times and with different types of device availability.”
Based on these questions and answers, the agent develops its skill for screening calls for a particular user. “If you had a secretary, and she forwarded a call to you without asking, and you hung up and said, ‘If that guy calls back in the future, query me before you send him in,’ – that’s a similar feedback mechanism,” he says.
Some work still needs to be done to improve the agent’s learning curve and streamline the user interface, but this feature should be available in 2006 or 2007, Edholm says.
AT&T is testing voice biometrics
AT&T is working on network VoIP smarts that can listen to callers and answer their questions without human intervention. Called multimodal, natural-language customer care, the system includes some current technology and some new developments, says Behzad Nadji, vice president of AT&T research and architecture.
If a user needs bank account information, he could go to a Web site, click to log on and set up a VoIP call. “It allows you to log in by speaking your name and PIN number and uses the spoken word to verify your PIN number, and it has a signature of your voice. It uses biometric technology to verify you are who you say you are,” Nadji says.
Rather than answer a series of yes-and-no questions as interactive voice response systems do today, users state their problem in conversational speech. The system’s VoiceTone speech engine converts what is said to text, analyzes it and fishes the answer to the question out of available databases. It converts the text from the database into a spoken answer using AT&T’s Natural Voices software. “When you hear these engines you don’t feel as if it’s a robot on the other end,” Nadji says. “It’s pretty natural.” It also has a 220,000-word vocabulary, about four times the vocabulary of the average adult.
VoIP helps integrate voice with data because they share a common packet infrastructure, which also opens the possibility of leaving multiple calls open all the time, Edholm says. In a circuit-switched network, a person makes a call, a circuit is created then torn down when the parties hang up. With VoIP, no packets are passed across the network when no one is talking, so connections could be kept open using low-bandwidth keep-alive packets until the connection is needed.
“You don’t have to go through the call setup and tear down. By clicking on a connection and opening that voice path you can open up an existing connection. You can have lots of intercoms,” Edholm says. So an administrative assistant for a department could work from home but still be available to respond to spontaneous voice questions. “It’s a lot like having your admin sitting outside your door and saying, ‘Hey Brenda, can you check something for me?'” he says.
Of course, it’s possible that several people might want to talk at once, he says, but Nortel is working on that, too. One possible solution is a dashboard that alerts recipients that someone else is trying to get through, and an automated response back to the talker saying, “hang on a second, they’re already talking to somebody else.”
With these types of applications and call features being written for SIP-based networks relatively quickly, VoIP could crumble under the weight of the sheer number of them if there is no way to make sure they don’t interfere with each other. For example, a conferencing feature could automatically call a participant at 2 p.m., but the participant isn’t at the phone and has activated call forwarding that will try several other numbers.
When the conference server calls and the find-me feature answers, the conferencing server asks for the participant’s PIN. The find-me feature doesn’t respond. It is looking for the participant. When it finally finds him and he connects to the conferencing server, it is waiting for the PIN. The participant doesn’t know this, hears nothing and hangs up.
AT&T is offering testing technology called distributed feature composition that would let applications explain themselves to each other so they don’t trip over each other causing negative interactions, Nadji says, eliminating potential management nightmares as cool new applications proliferate.
