When the consequences of errors are not significant, the speech amplitude and rejection levels required for input should be user-adjustable.