It's been over two weeks since I supplied my random rants and thoughts to this online journal. What kept me busy was a feeble attempt to fix the audio from the last comedy we shot. The audio was really, really bad because we didn't have an audio person and the producers kept the pressure up to shoot faster and faster. Of course something slipped. So because we didn't spend the time to get it right I had to spend two weeks cleaning up the aftermath. Frankly the audio still sucks, but as I put it to the producers. "The audio is still poo. It just sounds like poo that has the audio equivalent of Febreeze liberally applied." So tonight's technorati rant is going to cover a few things I did to help clean up the audio.
At first I tried ADR, but that was met with serious short comings - mostly because not all the actors were available the re-performance from those who were was hit and miss. ADR isn't easy. I surely couldn't do it, but then again I would be a terrible actor in general.
Finally I arrived at a workable solution, which at least produces a semi-tolerable soundtrack (almost, but still better than mumble-core). The first problem we had was not having real audio gear. We had to hide a lav in a candle pot at the center of the table. That was the best that was going to happen since we had six people to record and not enough time to switch mic's and test between takes. So much of the audio was recorded at too low of a volume which means, of course, hissy/noisy audio. The 5Dm2 has no real-time audio monitor or headphone out, so I thought we had the audio adequate...but no.
What I first did was try to mask the noise by shifting the volume of the audio clip down in between phrases from the actors. When the actors spoke, most of the time it would mask the excess noise. In some cases there was nothing I could really do but accept defeat, but a large majority of the time it worked as intended. The picture below shows the soundtrack with the individual audio clips and the volume shifts as the overlaid yellow lines.
Another quick technique I used was to clean up the background sounds with a series of notch filters. Often when recording in a home things (whatever they are) still buzz at 60/120/240Hz and other random frequencies. I'm not quite sure what we left running, but obviously it was still making noise. You can see the 120Hz and 240Hz spikes in the frequency response of the sound while no one is talking in the picture below. So I just use a set of notch filters at those two frequencies to remove the offending buzz. Simple and works almost every time, unless you have music in the background. Then you have to trial and error around this situation.
The last ditch effort involves using de-noising algorithms. Adobe Audition and Premier Pro use the same de-noiser as far as I can tell, so I prefer to apply it non-destructively in Premier. One of the main advantages of doing in there is that you can tweak the algorithms parameters such that the audio lays in correctly with the surrounding audio clips. I never trust an edit and forget approach. That often leads to one oddball audio clip that sounds correct in isolation, but is completely Borat between two Gloria Steinems. Apply this effect to taste, but over-using it gives an obvious robotic sound to dialog; which is even more distracting.
Some people swear by SoundSoap. I haven't used it personally, so I can't really say anything here about it.
What are my real lessons?
- Resist the demand to get everything fast instead of correct.
- Figure out how to get audio and picture fast and correct (working on it - new stuff coming).
- Help the producers understand that they must have the correct equipment and personnel, or they can spend more time - you generally don't get both.
- Again, do it right during production and post will be shorter.