Comparison between two methodologies of testing isolated word speech recognizers