Sci-fi may be showing the way again- subvocalization voice recognition or ‘mental’ speech recognition seem the obvious medium term answers.