From the introduction of NOAA Weather Radio until the late 1990s, nearly all the voices heard in the broadcasts were those of the staff at local National Weather Service (NWS) offices. The messages were manually recorded, first on tape cartridges and later digitally, and placed in the broadcast cycle.
As part of the NWS Modernization during the 1990s, many local offices were closed and their NOAA Weather Radio consoles were moved to the new or enhanced Weather Forecast Offices. This was also the start of a period of rapid expansion of the Weather Radio network. What had been about 400 transmitters in 1990 grew to near 600 by the end of 2000 and is now (at the end of 2006) over 960 transmitters across the 50 states, Puerto Rico, Guam, and American Samoa.
To cope with the increasing number of transmitters at each office, and to speed the overall delivery of warning messages to the public, the Console Replacement System (CRS) was deployed at NWS Weather Forecast Offices in the late 1990s. CRS introduced a computerized voice nicknamed “Paul” using the DECTalk text-to-speech system. DECTalk grew out of research by the late Dr. Dennis Klatt of MIT. While CRS greatly enhanced the speed of delivery and scheduling of Weather Radio messages, there was some dissatisfaction with Paul’s voice.
The National Weather Service embarked on a Voice Improvement Processor (VIP) program in late 2000, and implemented newer text-to-speech voices nationwide in 2002, nicknamed “Donna” and “Craig”. A year later, further updates were made. The “Donna” voice was improved, “Craig” was replaced by “Tom”, and a Spanish voice “Javier” was added at a few sites.
All of the VIP voices have been produced using the Speechify text-to-speech system. (The official Speechify name for our “Donna” voice is “Mara”.) Speechify was originally a product of the Speechworks company, based on technology developed by AT&T. Speechworks was purchased by Scansoft in 2003, and Scansoft merged with Nuance in 2005.
The VIP voices generally have been better received by the public than “Paul” was. There is a better capability to fine-tune the pronunciation of words and phrases along with controls to adjust the volume and rate of speech. These all help to make the voices more understandable when it really counts – in warning situations.