Nuance Is the Talk of the Town for New York’s Subway System
The Metropolitan Transit Authority (MTA) of New York City today launched an automated version of Trip Planner Voice in conjunction with Nuance Communications, Aspect—an unified communications software and consulting firm—and Trapeze Group, a transportation solutions provider.
Trip Planner is a 24/7 itinerary service aimed at New York City residents that lets users plan routes over public transport within any two municipally located points. The service was first launched as an online service three years ago and grew to include a telephone version, Trip Planner Voice, which began beta-testing in September of 2008.
By April of 2009, an average of 18,045 customers used Trip Planner daily, an increase of 172% over 2008’s 6,640 daily average. The call center version in particular, Travel Information Center, has received an average of 4,560 calls daily. These calls will now be fielded by an automated system with speech recognition powered by Nuance. The routes themselves will be mapped by Trapeze and linked with the speech components by Aspect.
Users will be able to get results by making their way through a series of prompts that allow them to set points of interest, intersections, or subway/bus stops as their beginning and end points. Once that has been entered, they are read back directions by the system through a combination of live recording and text-to-speech (which handles the name of streets). Should users have trouble with the automated system, they will be automatically routed to a live agent for additional assistance.
According to a prepared statement, today’s release represents both an effort to reduce call time and dependence on agents (and thereby costs) and to increase convenience for callers.
From a speech recognition perspective, one of the biggest challenges facing the project was the sheer number of street names in New York City. They city is home to tens of thousands of different streets, points of interest, and hundreds of bus and subway stations—all of which had entries that posed specific challenges for accurate speech recognition.
“Often there are many subway stations with the same name,” says Bobbi Manian, senior project manager for Nuance in its professional services division. “If a person said they wanted to plan a route and that they were starting at 14th Street, but they didn’t say ‘14th Street, Union Square’ they just said ‘14th Street,’ we had the intelligence to recognize that, but the back end database didn’t know what to do with that because it could have been 14th Street and Seventh Avenue, or Eight Avenue—so we made enhancements to say Which one do you want?”
Manian also points to difficulties with pronunciation. Houston Street, famously pronounced by most New Yorkers as “House-ton Street” is sometimes pronounced as if it were the Texan metropolis. Hoyt–Schermerhorn, an intersection and train stop in Brooklyn similarly has at least three different widely accepted pronunciations.
To deal with such complications, Nuance had to build its voice user interface to use graduated prompting and ask for clarification and get more precise information—offering more and more examples until a user is able to furnish the information the system needs. As the system continues to get more calls it is expected to become more accurate through the accumulation of data and resultant tuning.
This project represents the first joint venture between Nuance and the MTA. After tuning is more complete, the speech company’s role in the project may recede.
Currently, Nuance is working on a Spanish language version of the application. In the future, MTA may expand Trip Planner Voice to include other features such as text messaging, but no plans have been made definite.