Approaching dialogue recovery with clipped audio, what to do first? (DOCUMENTARY)

Question

hi everyone,

i am currently preparing the audio post process for a (low budget) documentary. the doc isn't finished and still needs a lot of work, but i was eager to see and hear some footage. the director and had a screening (on her laptop(speakers)) and talked about the sound and her ideas, so far so good... well i also asked for the omf and am now listening through the material in my studio.

my first conclusion is that it's pretty bad soundwise. the director filmed everything herself, but is not at all audio savy out of the 18 minutes, at least 6 minutes of the audio is distorted, not very badly, but defenitely noticeable in about 30% of the peaks per region.

the soundtrack consists mainly of an interview with one subject, over the course of three days. everything was recorded using lavelier mic (and transmitter) direct to DVcamera. the other part of the soundtrack is ambience and city sounds. everything is mono and there are no alternate takes available. some DV tapes have ok sound, others are distorted.

to get a feel of how to approach this little monster i decided to do some tests. i've started declipping with izotope rx to get an idea of how bad it is and must say i'm quite pleased with the results so far. but i'm still wondering how and where to start when i get the locked picture and omf.

normally i don't get audiorecordings this bad and restoring bad takes is only a small part of the job. i start with dialogue editing: cleaning up the tracks (ticks, clicks, pops etc) and a few passes have them leveled out evenly. (this is overtly simplified ofcourse) but now i face the challenge of having a lot of clipped audio and some reasonably good audio. cleaning up dialogue within the clipped regions is not an option, everything is to loud and if i bring down the volume i'm still left with a audio image full of clipped peaks, distorting my vision/hearing.

i managed to boil it down into two options:

a. start first with the reasonably good audio takes, cleaning them up and getting a feel for the dialogue flow (and it's levels) and then work your way down to the bad parts

b. start with the bad parts and use ie. izotope to declip the regions first* and then start the cleaning up of all the dialogue in one pass and go from there.

*i mainly ask this because i don't know how much izotope (in my case) changes in the audio besides extrapolating the headroom, is the rest of the audio left 'untouched'? i don't think so, but will it matter with audio in this saturated state?

am i still thinking straight? what would be your approach to this?

any help would be greatly appreciated!

greetings,

arnoud

UPDATE: well the documentary premiered a few weeks ago and the director (and me) are very pleased with the results. in the end it didn't take as much time as thought necessary. the clipping was fixed very well with RX2, I love it! There was only 1 sentence that sounded a bit ugly but no one actually noticed, so that's good to know. The doc is being well received and as a result will be screening around Europe on several festivals, so I am quite happy! Thanks again for all the advice. If you want to I could post some examples (soundclips).

Answer 1

The way most of the digital plug-in de-clippers work is that they try to re-draw the waveform how the waveform looked had it not been "clipped" at the top.

If your dialogue is full-on clipped at digital 0, the first thing I do in this case is gain down the audio at least 3 or 4 dB to give it the headroom the plug-in needs to do it's job better. You can't re-draw waveforms without headroom above the clipped cycles and this, in my humble experience, helps it work and sound much better.

Answer 2

I'm assuming you're asking about this, because you're trying to optimize your efforts. If the audio is that badly distorted, any side effects of the Izotope Declipper are going to be more tolerable/acceptable than the raws.

I'd start with the problem pieces, because consistency of sound is more important in this situation. Jarring jumps in audio quality are distracting and draw attention to themselves. The more consistent the audio (including if it's all crappy sounding), the more likely the viewer is to acclimate to and overlook its problems. If you've got one piece of pristine sounding audio in a sea of distortion, people will instantly begin recognizing that distortion...whereas they'll probably have ignored it previously.

I'm going to make one suggestion here though. Ignore the visual waveforms.

They're going to affect your percption of the problems. If you're working in Pro Tools, switch from "waveform" to "blocks" view. Use your ears, independent of your eyes, and determine where the most egregious problems are. Figure out which segments are completely intolerable, and focus on fixing those first. As best you can, get them to behave similarly to the cleaner audio clips. You're probably going to run out of time to work on the better pieces, but that's something you're may just have to accept. It's OK. Your return on investment of effort will be far greater this way.

If you start by working from the other direction, you'll more likely run out of time to work on the crappy audio...meaning you've lost that consistent sound, and drawn attention to the program's weaknesses. That's something you're client will not be happy with; whether they recognize it consciously or not.

Answer 3

You do what you can do! That's all you can do, really. I've done the same thing. But it was a commercial thing, the audio was distorted heavily all the time through the whole show and they still wanted to air it. I just took the X-noise, X-hum and X-click (Waves) through the whole mixing process and did all the EQ and trimming of levels POST FX. That way I could get a sense of how bad it was and how un-bad I could make it.

So for your project, I'd do the same; make one "clean" audio region with izotope, mix it if you have to with the original material to get some presence or whatever. But if it's clipped, not much to do than just work with it. Make it sound a bit smoother. But don't overdo it, trying to repair sound is like trying to glue an old vase together. If people pay attention they can clearly see the cracks.

Good luck!

Answer 4

Looks like Shaun and Olle have covered everything, but just about Izotope DeClipper: It has a threshold setting, and i'm pretty sure that only audio above the threshold is extrapolated. So if you put your threshold right at the point where the audio is clipped, you should be fine.

Also, you haven't mentioned it, but if you have a camera mic track, you can use that (with some eq treatment) to paste over very short clips.

Good luck!

Answer 5

I would just like to add to what other's have already said: that once you have declipped things, I find using EQ, SPL Transient Designer and Waves C4 to help shape the tone can make things sound a bit more natural.