Cleaning dialogue for fun and profit

As a segue from my previous post, this one is about dealing with the results from a bad location recording. The damage is done and the sound post team has to address it.

Most of the time, a spotting session is done to assess all the problems in the recordings. This session is attended by the director, the dialogue editor(s) and the supervising sound editor, or the re-recording mixer, if he is supervising the sound for the project. It is at this point that the dialogue lines that need to be replaced with ADR are flagged and logged.  The reasons for doing ADR are either extreme noise on the line or a performance issue that the director feels can be improved. 

The projects for which ADR is not done are documentaries and studio shows taped in front of a live audience. In these cases, we're stuck with the sync sound and it is up to us to make it presentable. The problems and methods to deal with them include:

• Incidental noises: these include stuff like audience coughs, cell phones, crew footsteps, lavalier microphones rubbing against clothes, very loud lip smacks, etc.  

They need to be removed or minimised as much as possible, as they are very distracting to the viewer. In many cases, they happen between words so they can easily be edited out in ProTools and filling the gap with ambient noise from the same microphone. Many times, though, they will occur in the middle of words. The only options are to either try to EQ the noise out by pulling down its main frequency range, or finding a word or syllable from the same person in the same scene to replace the unusable one. 

• Broadband noise: this is constant noise over the whole line. It can be an electric generator, distant traffic, open ocean ambience, tape hiss, steady wind, etc.

The best way to deal with this kind of noise is by using multiband expanders (Waves' C4, WaveArts Multidynamics 5), noise reduction plug-ins (Waves' X-Noise, WaveArts MR Noise) and parametric EQs to find and cut the most offending frequency range. 

Multiband expanders work by taking anything below a certain level in a certain frequency range and making it quieter. So, for example, if you have noise that lives in the low frequencies, you would set your processor's threshold and expansion ratio fairly high for that range, and it would only lower the level for those frequencies whenever the signal goes below the set threshold. 

Noise reduction processors work by analysing a sample of the noise and substracting that noise "signature" from the rest of the signal, leaving anything that isn't noise largely untouched. The problem is, if you go overboard with it, the dialogue itself will be affected and you'll get "artifacts", digital imperfections on the signal that will make it sound unnatural and boxy.

Parametric EQs are the most straightforward but less effective processors for dealing with this kind of noise. Basically, find the most offending frequency range by selecting a frequency with a medium bandwidth, boost it way up and sweep the entire frequency spectrum until you find a spot where the noise is at its worst sounding. Then turn down that frequency range until the noise is minimised without affecting the dialogue too adversely.

It is best to use a combination of these judiciously to clean up broadband noise – you'll get much better results than by picking just one and overusing it.

Some lines with some broadband noise may not be flagged for ADR in live action shows because the noise is determined by the supervising sound editor to be not too severe and it is decided that it can be cleaned up in the mix.