NWS hires fresh talent for weather radio
NWS hires fresh talent for weather radio
Can you say, 'meteorology'? Voice synthesis takes to the air to present the latest forecasts
- By William Jackson
- Jan 04, 2002
The National Weather Service is firing Igor and hiring Donna and Craig to take his place forecasting the weather.
Igor is the digitized voice that has been reporting current conditions and warnings since 1997 on National Oceanic and Atmospheric Administration Weather Radio. It did an acceptable job but didn't score well with listeners in the personality department. NWS director John J. Kelly began a search for his replacement in 2000.
'Voice quality had to be the No. 1 criterion,' said Lawrence Lehmann, manager of the Office of Science and Technology's voice processor project.
Donna and Craig, selected in August, aren't just another pair of pretty faces. In fact, they don't have faces at all. They are concatenated phonemic voices compiled by Speechify software from SpeechWorks International Inc. of Boston.
Igor is a computer-generated artifact of the DECtalk text-to-speech ISA card. State of the art a few years ago, he sounds woefully old-fashioned by today's standards.
NOAA Weather Radio began broadcasting in 1975 and now has 600 stations, each controlled by a local NWS office. The broadcasts serve 85 percent of the U.S. population 24 hours a day with current local conditions, forecasts and severe weather warnings.
For two decades the local forecasters wrote, recorded and broadcast these messages themselves'a time-consuming business that sometimes delays important announcements, Lehmann said.
When a 1995 modernization program cut the number of local offices from 216 to 121, each office became responsible for as many as 13 weather radio transmitters.
To offset the increased workload, automated text-to-speech technology replaced the human-recorded broadcasts.Now hear this
NOAA Weather Radio added the automated voice as part of its Console Replacement System, which updated the hardware and software used by NWS offices to produce broadcasts.
The DECtalk boards generate speech from ASCII text files. Force Computers Inc. of San Jose, Calif., acquired DECtalk technology when Compaq Computer Corp. bought Digital Equipment Corp. in 1998.
The Console Replacement System consists of a main processor at each local office, usually a 200-MHz PC, which schedules and manages messages. Up to three other PCs each have up to five DECtalk cards, one for each radio transmitter.
When the automated voice system began operating, forecasters could type scripts and quickly schedule where and when they would be played. A job that once took 15 minutes of every hour was reduced to 15 minutes each shift, Lehmann said. The broadcasters learned some lessons, too.
'There ain't no substitute for good English' in scripts, Lehmann said. 'If you write a run-on sentence, this baby will take a deep breath and go on forever.'
Many radio listeners said they were unhappy with the synthetic voice. Igor's pronunciation is sometimes difficult to understand and decidedly nonhuman.
By the time the search for a replacement began, voice synthesis had improved dramatically. Two pioneering technologies, computer voice generation and sentence-building from a dictionary of recorded words and phrases, had been replaced by a hybrid. The current state of the art fashions words and phrases from recorded speech elements, called phonemes. By blending individual sounds, software can create natural-sounding inflections.Sibling act
NWS was looking for a male and a female voice, so that different types of messages could have different presentations, and it wanted technology that would work with the Console Replacement System. The agency posted voice samples from five products on a Web site and asked for public response.
In a tally of the 19,000 comments received, Speechify's Craig got top marks for the male voice, and his sister Donna won as the female voice.
NOAA in August awarded a $633,615 contract to Siemens Information and Communications Networks Inc. of Boca Raton, Fla., to implement Speechify. The contract calls for 128 Speechify licenses for female and male voices in U.S. English, with one year of upgrades and an option for Spanish.
Speechify runs under Red Hat Linux 6.2 or 7.0 and will reside on a separate computer at each office to generate the sound files. The main processor will continue to handle scheduling, and the DECtalk cards on the other PCs will continue distributing the voice files to the radio transmitters.