The latest Dragon NaturallySpeaking software offers up to 99 percent out of the box accuracy, according to the developer, Nuance Communications. In practice, accuracy varies depending on factors such as how much a user uses the software, their type of voice or accent, and the quality of their computer microphone.
Speech recognition packages such as Dragon NaturallySpeaking use machine learning algorithms that improve the accuracy of the software by analyzing a user's speech patterns and adapting to them during use. With Dragon software, it is also possible to improve overall accuracy by adding commonly used words, such as names of people, to a dictionary of known utterances. Even after accounting for errors, Dragon was able to provide a 60 percent increase in input speed compared to typing, notes PCWorld.com.