I don't have any personal experience with voice recognition — other than Siri on the iPhone/iPad — but my son, a radiologist, uses speech recognition to dictate patient notes. Obviously recognition accuracy is of paramount importance in this setting. The notes use a limited albeit somewhat arcane vocabulary and generally are on the order of half a page in length. The system he uses is customized for the medical profession with particular attention paid to the specialized medical vocabulary. He dictates in a quiet room where the only background noises is the hum of the fan on a PC and he uses purpose built microphone. He always proof reads the dictation results and manually makes any necessary corrections. It is the rare note that does not require at least some correction. Sometimes he only has to manually correct a word or two, sometimes it is more expedient to hit the delete key and start over, and sometimes he resorts to the keyboard. By-the-way this system is not hosted on his PC. Instead the voice is digitized and sent to a dedicated host computer system.
The key factors in accuracy appear to be:
- No background noise
- very high quality microphone
- moderate to slow speech rate
- somewhat monotonic delivery
