Noise Reduction - How Much is Too Much?

Software plugins like Izotope RX, Cedar DNS One and Klevgrand Brusfri have saved Dialogue Editors and Post Production professionals hundreds of hours in ADR sessions over the years, by reducing the presence of excessive noise in production dialogue.

While these tools are extremely helpful, how far can you push them until they are more of a hindrance than help to your overall dialogue track?

How It Works

Noise Reduction (NR) plugins employ a variety of techniques to help remedy audio issues. This usually involves EQ, gating, expansion and phase reversal of unwanted portions of the signal. Results can vary extensively, which is why it is useful to try multiple passes of NR software on challenging material.

To Reduce or Not to Reduce?

With this technology available at the click of a button, it is tempting to apply NR to every bit of dialogue that comes your way. Unfortunately, anything above an 8dB reduction in noise (material dependant) usually results in artefacts or tonal changes that the audience may notice. Because of the processes involved in NR software, some of the intelligibility of dialogue can be affected, as we are usually left with more mid frequencies in our signal after processing.

Not only can this make the dialogue sound ‘washed out’, but it can also leave us with nasty resonances that are more damaging to the mix than the original noise. While it is possible to then correctively EQ these frequencies, you will then end up with dialogue with much less weight and clarity.

Alternative Options

You can help reduce noise more effectively with the use of multi-step processing. If you are time limited, this can just consist of multiple smaller passes of NR, for example 2 x 4dB passes rather than a single 8dB pass. This will usually introduce less artefacts. Ideally though, you will want to incorporate different plugins into the chain, so that you are using smaller amounts of multiple processes. For example, our chain often includes a high pass filter, high shelf EQ, expander, transient shaper, NR and volume automation to attenuate noise when speech isn’t present. Although much more automation and manual processing will be needed, you will find your NR is far more transparent, and won’t ruin your dialogue tracks for audiences.

Try building an NR signal chain into your dialogue editing template and see what you can come up with!

How much noise reduction is too much, and what are your favourite tips and tricks? Comment below.

‍