Speeding down the voice track

Union Pacific Corp., the largest rail system in North America, has installed a high-security speech-recognition system that corporate customers, such as The Home Depot Inc., Monsanto Co. and Dow Chemical Co., are using to schedule rail car movement in and out of train yards.

The first of its kind for e-commerce in the train industry, the speech-recognition system converts voice-based requests phoned in by the railroad company’s customers and converts this spoken information into Web text, automatically updating Union Pacific’s Web server of rail car information. The speech-to-Web application also has a “voice biometrics” security feature to check the caller’s voice pattern against a digital voice sample stored in the so-called speaker verification database. IT managers at Union Pacific say the application, based on SpeechWorks International Inc.’s SpeechSecure, is much safer than phoning in rail car scheduling information to Union Pacific’s call centre.

Voice biometrics makes faking the identity of a legitimate person for purposes of sabotage “impossible,” says Paul McGee, Union Pacific’s senior manager of IS. “You’d have to be holding him hostage.”

With heightened security concerns following the Sept. 11 terrorist attacks, the IT department found Union Pacific upper management very responsive to the idea of adding the voice-to-Web application as an alternative to the traditional method of phoning in rail car scheduling requests to a call center. About 500 of Union Pacific’s largest customers have started using it to schedule movement of about 1,500 rail cars per week.

“We’re shipping hazardous materials and military supplies,” McGee says. “So security is important.”

Union Pacific, which won’t disclose the cost for the voice-recognition project, says factors other than security were also important in driving the decision to use speech-to-Web technology.

It’s a more efficient way to have rail car supervisors in the field phone in their instructions on what rail cars they want to have available for loading goods or sent to a destination point. The speech-to-Web system banishes the commonplace errors – which have expensive consequences in misdirected rail cars – that are made at call centers when call service representatives are listening to instructions over the phone and keying it into a computer, McGee says.

Union Pacific’s first wave of customers using it say they have found it easy to use.

“It works great,” says John Buchmann, distribution center manager for The Home Depot in San Antonio, Texas, who oversees rail car unloading. “It also faxes a confirmation of everything you’ve said after it’s done.” The speech recognition’s machine voice repeats back the rail car information given to it and asks if it’s correct.

“We do 130 cars a month,” says Buchmann, adding he likes the voice biometrics security. “No other associates working with you can say your password because the password is linked to your voice.”

How it works

After authorized personnel authenticate identity through voice biometrics, they speak the alphanumeric rail car numbers along with instructions for their release. The SpeechWorks’ application repeats it back for accuracy, and then converts the digital voice message to text to update the Union Pacific Web server, which runs a proprietary application called Car Release. Rail cars don’t move until the customer has given the instruction to do so.

Customers can also input rail car instructions directly into the Union Pacific Web server over the Internet, but this is not always possible for distribution supervisors when they are out in the rail yard all day.

“The real target for this speech-recognition system is the group sitting on the dock that doesn’t have Web access,” says Union Pacific’s director of e-commerce, Charlie Duckworth.

Information, converted from speech to text, is stored as HTML data in Union Pacific’s Web server, which is also a front end for the corporate mainframe. Once the mainframe has the rail-car-related information, it will schedule a task for railroad personnel, such as pulling an empty rail car from a customer’s loading dock so it can be available to pick up goods elsewhere.

The rail-car-release information is important because it not only sends rail cars on their way, but it also automatically triggers mainframe-based invoicing and billing functions.

Union Pacific plans to expand its use of speech-to-text for use by truck drivers arriving at rail facilities to pick up goods, McGee says. “They’ll speak their ID to validate who they are,” McGee says.

The SpeechWorks-based system was chosen after a competitive bidding process in which SpeechWorks and competitor Nuance faced off. While there wasn’t much difference in the prices, according to McGee, the deciding factor was perceived ease of integration with Union Pacific’s Web server.