I realize you didn't create that graph, but I'm disappointed you would reproduce it given that it is highly misleading in several ways... 1. It uses the good old trick of manipulating the y axis to suit whatever interpretation of the data is wanted. Replot that data on a linear instead of logarithmic scale and voice recognition software suddenly doesn't look so dismal. 2. The data set shown here in red was cherry-picked from several found on the original NIST plot. When the number of words that need to be recognized is fixed and limited (such as commands that can be issued to a computer), voice recognition performs quite well. 3. The last data point on the plotted set is from over 8 years ago. Lack of data since then in no way implies that no progress has been made.