Document text
Forecast: It'll either be Craig or Donna By BILL KETTLER Mail Tribune Weather watchers won't have Sven to kick around much longer. Sven's not his real name. He doesn't have a real name. He's not even a real person, but his synthetic "voice" is familiar to everyone who listens to a National Weather Service weather radio. Medford forecasters nicknamed the voice "Sven" because its vocal inflections reminded them of somebody from Scandinavia. Sven replaced the voices of real forecasters, and he's been a source of complaints since he started work in December 1999. Some listeners complained that his voice was hard to understand. Others said he was just plain hard to listen to. Sven's days are numbered. The National Weather Service will introduce a new voice next spring to the weather forecasts that are broadcast by 120 stations across the United States. The Medford weather station and other local offices will choose between "Craig," a new male voice, and "Donna," a distinctly female voice. "They both sound very, very real," said Roger Williams, meteorologist in charge at the Medford station. "They're not perfect, but they're at least 80 percent better than Sven." Weather radios provide instant access to forecast data that is updated every six hours. The inexpensive, fixed-frequency radios are popular among outdoor enthusiasts, farmers and anyone else who wants up-to-the-minute weather information. Forecasts from the Medford weather station cover an area that extends north to Roseburg, west to the ocean, south to Mount Shasta and east to Klamath Falls and Lake County. In the early 1990s, the Weather Service decided to use computerized synthetic speech to free forecasters from endless recording chores. The computer behind Sven's voice had a limited vocabulary of sounds. Technicians had to spell some words phonetically to make Sven pronounce them properly. Oregon, for example, had to be spelled Orregun on the typewritten forecasts. Williams said the Weather Service initially intended to use real human voices for its second-generation of computerized speech. The plan called for building a library of spoken words, phrases and sounds that a computer would link together as it scanned a typewritten forecast. Agency managers abandoned that idea when they realized the sheer number of place names that would have to be recorded and stored. "You could come up with a basic vocabulary," he said, "but how do you deal with Snoqualmie Pass, or all the Indian place names in New England, or the Cajun names in Louisiana." The new computer-generated voices were produced by Siemens Information and Communication Network of Boca Raton, Fla., which won the $633,615 contract to produce new software known as Speechify. Williams said the Medford weather station initially planned to alternate the male and female voices in its forecasts, but the software is probably too complex to switch between voices. "Once you get the voice on the transmitter," he said, "you don't want to do it twice." Williams said choosing the voice could be difficult. "I was hoping we could make more people happy by alternating the male and female voices. "Some people are going to like a male voice. Some people are going to like a female voice. How do you choose? I don't know." Reach reporter Bill Kettler at 776-4492, or e-mail [email protected] Mail Tribune Home | Ottaway Newspapers, Inc. | Dow Jones & Co., Inc. | Privacy | Contact Us Copyright � 2001 Mail Tribune, Inc.