Discrete speech systems for command and control usually have the highest recognition accuracy due to the limited vocabulary of commands. As the vocabulary increases (e.g. for dictation systems), the level of accuracy generally reduces.

Background noises can adversely affect recognition accuracy of any speech system. Sounds from background noises may be mis-interpreted as commands/speech.

How to Determine Accuracy
Read or dictate at several paragraphs of text into Microsoft Word using your speech recognition program without correcting any errors that the program makes. Determine the total number of words dictated by using the "Word Count..." option on the Tools Menu or the Properties/Statistics option on the File Menu. Then count the number of errors made and compare that to the total number of words spoken. To determine an accuract score as a percentage of the total words dictated, divide the number of words dictated correctly by the total number of words dictated. For example: two errors out of 100 words would be (98 correct / 100 total) = 98%.

Key Steps to High Speech Recognition Accuracy (http://www.emicrophones.com/docdetails.asp?documentid=38)
From emicrophones.com