Would also like to know this. It looks like they're processing the audio file in 30 second chunks, so a naive approach of keeping a buffer of 30-second input stream chunks and just continually writing to an output .mp3 could work...