Playing a random sound during another stimuli is not a problem. I would probably just randomly select it from a stim file. Have an embedded silence in the sound file to create an onset and then, without delay, present the sentence. Then take two RTs to the sentence--one for the sound and a second one for the sentence itself--using separate valid keys for each. There may be other ways, but that's off the top of my head. You can also do a search on here for "probe" -- you might find some useful links with that--both for DirectRT and MediaLab.