I don’t have a lot of experience with the messier side of audio processing, and I’m trying to deal with about a hundred hours of recorded speech that needs some work before putting it on the Web. Most of it is pretty much OK as is, and some of it has some clipping. One or two hours of the material is absolutely horrible, and I have no idea what happened with this audio or whether there’s any shot at recovering a listenable signal.
Here’s a link to a short sample. Obviously the signal is clipped at -5.5 dB, and a quick look at a spectrum plot (with large enough windows) shows there’s some funky stuff going on with harmonics of something around 185Hz or so. But this is still quite a puzzle to me.
If you’re recording on a Windows computer, it sounds like you’re running into many of these problems…
Using a Windows machine as a tape recorder can be an adventure.
It sounds and looks like a comb filter …
If you feed-back the output to the input with a slight delay that can create comb filter effect,
e.g. if you’ve forgot to uncheck “software playthrough” when recording “stereo mix” (aka what-U-hear) …
Aha! That makes sense, and Audacity’s built-in autocorrelation stuff shows very strong autocorrelation at 5.3 ms (which would be a fairly normal playback delay). I’ll putz around with Octave a little and see if I can’t improve the signal somewhat with a deconvolution.
I should say-- I didn’t make these recordings; if I’d been on hand I would have known better than to include playback in the recording mix. I received the hundreds of hours of material from someone else and am left to guess at what to do to deal with their problems.