Speech recognition software has made impressive progress in the past year, but it is not yet suitable for widespread use. While it is easier to use and has better recognition capabilities than a year ago, it still is only 90 percent reliable unless the user spends a great deal of time with the software to teach it to recognize a specific individual's voice almost all of the time. Five hours of reading canned paragraphs and correcting errors will typically increase reliability to 95 percent. Output from speech recognition requires very careful proofing, since even wrong words will be spelled correctly and thus will not be easily sighted. Overall, data entry by a good keyboarder is less time consuming than using speech recognition and correcting the bloopers.
The application that appears to be most widely and successfully used is data entry and retrieval for physically handicapped persons, including persons with severe carpal tunnel syndrome.
