Today I have been inspired to share a little about the ‘dark art’ of dialogue noise suppression. I mean this in the sense that we’re shooting in the dark a lot when it comes to noise – there is no absolute template or ultimate solution to the problem which works every time, and different tools have their own unique merits and shortfalls. I’d like to focus on one largely-overlooked tool, the Waves C4 Multiband Processor, while additionally touching on other tools and techniques which have rarely been uncovered.
The most appropriate place to start is with the basics of the human voice, and noise in general. We do not know where we’re going if you do not know where we come from. So what is there to know about the human voice? Let us begin with a breakdown of its parts, pieces, and terminology – it’s like an anatomy class but less gross! Through my own experience and philosophizing (gold star to the individual who can name the movie that made philosophizing famous) with colleagues over the past 7 years, I have arrived upon this general breakdown of the male human voice:
- 20Hz – 80 Hz = NULL (except for James Earl Jones et al)
- 80Hz – 130Hz = Proximity
- 130Hz – 400Hz = Resonance (usually unwanted around 250Hz – 350Hz )
- 400Hz – 750 Hz = Presence/saturation/room (sometimes unwanted)
- 750Hz – 1000Hz = Nasal (usually unwanted)
- 1000Hz – 3500Hz = Intelligibility
- 3500Hz – 6500Hz = Clarity
- 6500Hz – 9000Hz = Transparency
- 9000Hz – 15000Hz = Air/Life
The female voice follows a similar scale, yet is shifted slightly higher (naturally). The listed boundaries are not hard-and-fast rules either; a gradual transition crossover of about 50-100 Hz, sometimes more in the upper registers, is a more accurate assessment. When this terminology is used however, more often than not these are the frequency ranges and spectral regions which are being referred to.
Noise, is well, everything else. The dread ballast hum from outside and around the corner, the fridge buzz during an intimate love scene, and the traffic drone when we’re supposed to be out in the middle of nowhere in a 127 Hours-esque locale. Some of these noise types are purely broadband in the spirit of (naturally-occurring) white noise thanks to the cosmic radiation background of the universe. Other noises are purely harmonic such as buzzes and hums. Both though are not wanted. The production sound mixer will use a variety of mic’ing and recording techniques to baffle such noise, yet there is only so much a sound mixer is able to do with the tools they have and the circumstances of the location/production. Because of this, we rely upon dialogue editors to first massage the sonic ‘mess’ of mosaic sound edits into a a fluid context to both support the story and detract attention from its technical impurities. From there, the role of the dialogue premixer (usually the dialogue re-recording mixer) focuses upon clarifying the dialogue edit through various techniques of equalization and notching out harmonic noise, dynamics control, and most important, broadband noise suppression. I am always a strong believer that the more time spent in the edit to carefully blend backfill noise and edit out ticks, pops, lip smacks, etc results in less perceived noise and a dialogue track with which the mixer can spend more time creatively clarifying it rather than having to address nit-picky technical hurdles. Yes it results in much more dialogue editorial time, although it pays off big on the stage. In many cases, the golden rule “add noise to remove noise” rings very true – by adding physical noise (such as carrying all the noise floors from jarring A/B/C angle edits), you will lower the level of perceived noise. Its a psycho-acoustical phenomenon that I milk for all it’s worth and have no shame in disclosing it. Mix in the backgrounds – and it works even more effectively than hearing the naked dialogue. It truly works wonders on even the most facepalm of scenes. And, it even allows more easier broadband noise reduction in some cases because you are playing with a global noise signature and ducking that around dialogue lines rather than dancing between three different noise signatures.
So, the $64,000 question… how do you suppress the broadband noise? For starters, I’ll mention the Cadillacs of noise suppression, the CEDAR DNS (Dialogue Noise Suppressor) and the Dolby Cat .43 (which utilizes Dolby-A Noise Reduction algorithms). A CEDAR DNS2000 hardware/software package easily costs $7,000. Outrageous? Not so much. It is capable of some amazing feats that even I was floored. Give it a variety of atrocious noise and the CEDAR says “sure, I know what to do with this, no problem” In a nutshell, both of these tools operate on the basis of multi-band expansion with a series of 4-5 faders (one per band) which are played like a mixing console. And, they both operate as hardware units, while the CEDAR hardware ties directly into ProTools in the form of an RTAS plugin so that automation and settings can be saved and restored. While these tools are great, they’re not easily-available nor realistic options on an “in-the-box” budget. That’s where the Waves C4 comes into the picture.
The C4 is a slick tool in a variety of ways, and with Waves’ new re-pricing plans, this sucker can be picked up for a mere $99. When you work its magic for dialogue noise suppression, I feel this plugin is easily worth somewhere in the $1500-$2000 range. The trick to this plugin is how you configure it. Upon loading its default settings when you launch the RTAS, anybody who has used the CEDAR or Dolby Cat .43 will feel right at home. This plugin is a near-zero-latency multi-band compander – or, compressor/expander – all-in-one. It to uses 4 bands (the new C6 operates on the same algorithm as the C4 yet has 2 more ‘floating’ bands). In it’s default state, the plugin is prepared to do compression.
Lower the global gain to about -6dB, and raise the global Range to about +6db, effectively resulting in the top of the purple bar to rest at the zero point. Pull the global Threshold to about -50 for starters. What you now have is a multi-band downward expander. It works like this: when the dialogue output is below the Threshold levels (e.g. noisefloor), the will be expanded (suppressed) by 6-dB or more if you choose more – see band 1 in the image. As soon as the dialogue pushes past those Thresholds (e.g. spoken dialogue or movement), it will immediately output at unity gain – see band 3 in the image. Play the Thresholds to control how open or closed your output is for each frequency band! Adjusting the Qs between the bands, as well as the Knee and individual Attack/Release of each band, affects how jarring and tight or fluid and smooth the bands operate – different noise situations call for different degrees of manipulation. Heck you can even alter the Gain/Range of each band separately, allowing some bands to expand more or less than others on their own compared to the global range. And that’s what makes this plugin such a slam dunk! Unlike the CEDAR, Dolby Cat .43, and even Waves’ very own WNS and W43, the C4 allows complete finite control over every aspect of how this multi-band expansion operates. Granted, WNS offers some bare-minimum options, it’s nowhere near as comprehensive as the C4. The band control is highly useful, sometimes even more automating around tricky noise problems, and that’s where the frequency breakdown at the top of the post becomes very useful. As you will note, I have used this information to set up Band 2 to control the muddiness and roominess (and traffic tone), where Band 3 focuses exclusively on Intelligibility. This plugin will do some serious broadband noise suppression when you learn to work with it – I’ve spent about 4 years intimately working with the plugin to get to a point where I feel comfortable in what I can make it do in dialogue premixes, and am still learning more about it as I go and try to push its limits. Don’t just take my word for it, dialogue re-recording mixer Yuri Reese has used this plugin in his dialogue chain too. And I’m sure there are others who have taken it on as well.
Here are some dialogue samples to show the rockin’ power of this plugin, with special thanks to Johnny Lee director of “The Judas Cradle” for graciously allowing these for demonstrative purposes. I have chosen to demonstrate this because not only was the dialogue marred with noise inconsistency and unacceptable noise levels when the raw tracks were gain staged to a proper dialnorm level , it also demanded a squeaky-clean dialogue mix because of both the film’s location and the fact that we sucked out the backgrounds in one section – so the dialogue had to stand on its own in naked form, there was no leverage for creative sound masking. It had to play right. This image at the right displays my meticulous dialogue edits within the mix session (bussing etc) and zooms in on the section for which the below audio samples are provided for some visual context. This is the usual level of complexity that my dialogue edits on shows entail, and I have learned to cut this way at a rapid pace. As I said, it’s a lot of up-front work on the edit, yet it pays dividends on the stage. As you will notice by the presence of edits on my X tracks, I even had to embark on some FFT noise suppression courtesy of iZotope RX in order to get the dialogue into a workable, yet still unacceptable, signal-to-noise ratio.
Below is before and after the C4 was used in the premix sidechain. In both of these, all premix chain plugins are running however, on the first sample only the C4 is disabled so differences in before/after can be isolated solely to what the C4 brings to the table. I highly recommend listening to this at a proper -20dBFS = 85dBSPL reference level on monitors to hear the noise issues in proper context, or use a set of headphones.
Premix chain, no C4:
Premix chain, with C4:
Amazing huh?
That’s the gist of this plugin. Why the small explanation about it? Because it’s that simple to work with! Yes, it takes a long time to truly learn its capabilities and cajole it into working for you in the best way possible. Yet, it’s price and functionality cannot be beat for its’ quality. I have sampled the WNS (Waves Noise Suppressor), based upon the CEDAR, and have found nearly no difference between it and the C4 – the only difference to me is that the C4 comes in radically cheaper in price and loads you with a wealth of customization controls that the WNS doesn’t provide. I may need to spend more time with the WNS to confirm my thoughts. However, in brief side-by-side processing tests of both the C4 and WNS on the same piece of dialogue, I was not impressed whatsoever with the WNS – I’m happy as can be with my lil ol’ C4.
I have just learned about Algorithmix reNOVAtor, made famous in “TRON: Legacy” by its use at Skywalker Sound in pulling out strong harmonic noise generated by the light suits that were worn by the actors. Supervising Sound Editor Gwen Whittle shared briefly about this tool and the multi-pass dialogue premix process that Gary Rizzo underwent at the Q&A I attended at the film’s screening at the AMPAS theater. Alas, this tool will have to addressed at a different time when I have an opportunity to play with it and put it to work, when I can afford it. All that I know is that it works on spectral analysis principals, and has me very intrigued. More shall be revealed.
And with that, I bid you good evening and happy noise wrangling!