• Henri Falk

De-Noising Recordings

One of the most important tools for me in the process of making this spatial audio installation for headphones has without a doubt been the de-noising plugin. When making recordings in an inherently noisy urban environment, or when cranking up the input recording volume to catch some quieter sounds which, using cheap-mid range recorders, will lead to a rather loud noise floor from the mics' preamps, then a de-noiser is invaluable to transform the recordings into clean, high-fidelity samples.

I have been using the spectral de-noiser included in the Izotopope RX 7 suite. RX7 includes a number of sophisticated audio repair tools and comes in three different editions; elements, standard and advanced. The de-noiser is included in both the latter ones and for the standard version they are currently charging 399 USD. The Advanced version gets very pricy at 1,199 USD, but it is quite incredible what these tools can do. Except getting rid of noise, clicks, reverb, isolating vocals etc., a world of possibilities open up also for (mis)using these tools to mangle and twist audio samples in a creative way.

I will focus on the Spectral De-Noiser in this post to have a closer look at how it works and what it can do.

The Spectral De-Noiser plugin included in RX7

This tool is designed to remove stationary or slowly changing tonal noise and broadband hiss by learning a profile of the offending noise and then subtracting it from the signal when the amplitude drops below the specified threshold. It features separate controls for tonal and broadband noise, management of denoising artifacts, and an editing interface for controlling reduction across the frequency spectrum.

With the Learn button engaged, the plugin listens to a select piece of the audio, preferably a couple of seconds long section of the audio containing the noise only, to form a static profile of what it will subtract from the signal. This works well if the noise profile and level is rather constant throughout the audio sample. The amount of reduction can then be fine tuned using the threshold and reduction sliders. By checking the Output noise only box, one can listen to the delta noise which is being subtracted from the signal. The threshold slider controls the amplitude separation of noise and useful signal levels (0 dB is a good default setting), while the reduction slider controls the desired amount of noise suppression in decibels.

Checking Adaptive mode, makes the noise profile change based on the incoming audio. This is a useful option when the undesired noise changes a lot, for example in outdoor environments. The adaptive mode is heavy on memory and CPU, so bouncing these processed samples is a good idea in order not to have several of these processes going on in the DAW.

There is also a quality slider, ranging from A to D, with increasing algorithmic complexity, processing time and CPU load. More information about these algorithms can be found in the instruction manual. If real time processing is not necessary, one might assume that the highest quality renders the best results. However, I would encourage users to compare results. Depending on the sample, I have sometimes found a lower quality setting sounding better or more musical than a higher quality one.

The Artifact control slider determines how much noise reduction will depend upon either spectral subtraction (lower values) or wide band gating (higher values). Lower values can more accurately separate noise from the desired audio signal, but can produce musical noise artifacts. Higher values will have fewer musical noise artifacts, but sound more like broadband gating, resulting in bursts of noise right after the signal falls below the threshold.

With Reduction curve enabled, the Smoothing slider controls the amount of interpolation between the reduction curve points, allowing for sharper or more gradual slopes between edit curve points. This makes it possible to customize the amount of noise reduction being applied across different frequency regions. Higher edit point values result in less noise reduction in the associated frequency region, while lower values result in more noise reduction in the associated frequency region.

There is a number of additional advanced settings, some of which are altered when the quality slider is changed. I recommend reading through the instruction manual to get an understanding of what the different settings do and then experiment and listen to different results. If necessary, more manual carving out of noise and setting the noise profile can be achieved by using the RX7 spectral selection tools in the application. This can not be done in the plug-in.

I hope that this post provides an overview of what can be accomplished with de-noising and what this plug-in is capable of. It is a very sophisticated and useful tool if you are working with field recordings where unwanted noise is often hard to avoid. It comes with quite a hefty price tag, but I would recommend giving it a try.


Be in the loop (infrequent updates)

©2020 by Henri Falk