Computerized weather voice undergoes fine-tuning

Print More

(Host) Many Vermonters use specialized weather radios to tune into forecasts broadcast 24 hours a day from the National Weather Service office in Burlington. Three years ago, the Weather Service began using a computerized voice in its broadcasts. The change freed forecasters from the chore, and enabled the Weather Service to get information out faster.

As VPR’s Steve Zind reports, the effort to improve the computerized system means weather radio listeners shouldn’t get too attached to the voice bringing them the forecasts.

(Zind) He’s been one of the most listened to radio voices in Vermont. Familiar to farmers, boaters, campers, gardeners – anyone who tunes into one of the three weather radio frequencies in the state. At the National Weather Service office in Burlington, they call him Arnold, after the similarly Slavic sounding film star:

(Arnold) “Broadcasting from a frequency from the summit of Mount Ascutney in southern Vermont.”

(Zind) Arnold isn’t exactly Mr. Personality. When the weather service made the switch from a human forecaster to Arnold’s computer-generated voice, many Vermonters complained. Bill Grady is with the Burlington weather bureau:

(Grady) “One person did write and tell us that she used to give weather radios to her friends and she took the batteries out and threw the radios in the trash can.”

(Zind) But weather radio listeners don’t have Arnold to kick around anymore. Meet Craig:

(Craig) “Today, partly sunny, high forty to forty-five.”

(Zind) Over the past several months, Craig has taken Arnold’s place as the new computer-generated voice of the weather service. The switch was made because research showed that Craig is more intelligible than Arnold. He certainly sounds friendlier. If you didn’t like Arnold’s metallic, no nonsense delivery, you’ll probably find Craig’s smoother, more inflected voice an improvement.

(Craig) “High pressure from central Canada will move southeast across the region through much of the week.”

(Zind) Craig is an off the shelf, voice-to-text computer program designed for a variety of business uses. It’s taken hours to customize the software to recite weather forecasts for Vermont. That’s been the job of meteorologist Jason Neilson.

To get Craig to pronounce a word or phrase properly, Neilson carefully types it out phonetically. He sits at a keyboard, experimenting with different letter combinations to find the most natural sound.

(Neilson) “Like the words, ‘as a low.’
(Craig) “As a low moves across the forecast area.”
(Neilson) “As you can see here, I’ve designated it as ‘a-z-z-a-l-o-e,’ to make it sound. Instead of just three separate words, I made it sound as one word.”
(Craig) “As a low moves across the forecast area.”
(Neilson) “Much better as far as your hearing it, than it sounds originally.”

(Zind) Neilson also had to enter local place names into Craig’s vocabulary. There’s a lot of trial and error involved. Craig might say a word differently from one day to the next as Neilson continues to tweak the program.

One problem Neilson discovered is that Craig seems emotionally invested in certain words. Craig sounds sad when he uses a word like “cloudy.” A fair weather term like “clear” makes him downright giddy.

(Craig) “For tonight, CLEAR, with lows in the twenties.”

(Grady) “It does put an inflection in on certain words. This is one problem that’s created in that you don’t want it in there all the time.”

(Zind) To do away with Craig’s editorial slant on the weather, Neilson adds a single letter to the phonetic spelling of the word “clear.”

(Neilson) “Just one letter – I put in an ‘e’ between the ‘a’ and the ‘r’ and .”
(Craig) “For tonight clear, with lows in the twenties.”
(Neilson) “And that happiness is all gone.”

(Zind) Craig has a female counterpart named Donna. Both voices are in use at some weather service bureaus.

Eventually, the Burlington office will introduce a female voice, but it probably won’t be Donna. In the next year, she and Craig will be history; replaced by new computer voices that will sound even closer to the real thing. Grady says the day will come when it will be impossible to tell the difference between a human voice and an artificial voice.

As that day draws near, it’s anyone’s guess where a talking computer might turn up in the future.

(Craig) “For Vermont Public Radio, I’m Craig in South Burlington.”

Comments are closed.