When a suspended audio-source object is first created, it must be passed an IUnknown to a "real" audio source. The suspended audio source should then be passed to the speech recognition engine.
An application can make calls to IAudioSourceSusp to pause or resume the audio.