I really look for a way to do this: even out the levels of a recording, recorded on a device with Auto Levels on. I have some material from Zoom H1 and this is the exact problem. After some loud clip or noise the levels go down and this is undesirable. Manually it can be more less corrected but it seems it should be possible to make an automated process, similar to Sound Remover "Learn Sound Model" function. Learning the backround noise can be used as the required volume and the rest of the audio could be matched to that piece. I hope something like that could be done. Thank You.