This happens when a page uses webkitAudioContext at the same time as the speech recognition API. This is likely a common need, for processing audio graphically while also checking for speech matches (e.g. speech volume meter, spectrum analysis, etc.) It should only ask once for both.
Is it something WebKit level or Safari level? Plus it is mentioned 'Chrome', so it might not be applicable. @ap - any idea? Should we close it because it mentions 'Chrome' and I haven't came across this error myself.
Please provide a test case if this is still an issue (and sorry for ignoring this for so long, not sure how this fell through the cracks so completely).