Tales of RX: 3 steps to remove dialogue reverb

In the world of sound engineering, audio restoration is one of those skills that is so important to have and to keep improving on. It’s such a useful skill to master, that most companies will add knowledge of audio repair techniques as a requirement for recruits and, most times, they will specifically be referring to Izotope's RX.

Izotope RX is currently the industry standard for audio repair processing, offering an array of dynamic high quality plug-ins that will tackle any undesired sound from every possible angle. This makes them an incredible addition to your arsenal of plug-ins whether it’s for film, games, music or more.

It’s for all these reasons that I want to share some of the things I’ve learned while using RX in the hope that it will be of help to anyone reading this. So, without further ado, let’s see what we’re dealing with.

The Files..iles..iles..iles..

Recently I worked on an independent project in which the dialogue was purely voice-over. It is the nature of independent projects to have more restraints in terms of budget so, this meant the dialogue tracks that I received were not recorded professionally and therefore, had some quality issues. Then again, noisy dialogue occurs even in big blockbusters (See TRON: Legacy) and it is our job as dialogue editors to look past it and work towards the delivery with great quality in mind.

Upon first inspection, the dialogue sounded like it was recorded in the centre of a large concrete room. This meant it had a massive amount of reverb that crossed over every single spoken word. Now, In view of the fact that this dialogue was meant to be a voice-over, this amount of reverb conflicted with my sound references. What do I mean by references? Well, let me ask you this:

If I ask you to think of any film that has voice-over narration, can you remember what that sounded like? I was thinking of Psycho’s end-scene, The Shawshank Redemption or Gone Girl. They all have that very up-close and intimate narration that I was aiming for. However, the reverb is strong with this one, so let’s take it step by step and start from the bottom.

Step 1: EQ

After importing the dialogue file into the RX audio editor, the first thing I did was head over to the EQ to tackle the low end. As powerful as RX is, we want it to focus on what we want to hear, so we start by clearing those extra frequencies that we are not really going to need. Considering we are working with a female voice, most of the useful frequencies are usually above 150Hz approximately, so this is where we start using our ears instead of our eyes. It is very easy to set a high-pass filter at 150Hz and assume you don’t need the rest of the frequencies but, sometimes the difference between 150Hz and 100Hz can add or subtract a lot from the performance, so don’t take this as a rule of thumb and always trust your ears. In my case, after listening back a couple times I decided to set the high-pass filter at 115Hz.

Below you will see a picture of RX’s Spectrogram. This is an incredible feature that is worth looking into. I won’t delve too much into it here but, learning to understand what you are seeing on this spectrogram can make you a better sound engineer by having the ability to identify problems before you’ve even listened to an audio track (See More). Take this example:

Here is a picture of the dialogue's spectrogram before we’ve applied EQ to the low frequencies:

And here is after applying the 115Hz High-Pass Filter:

Notice how we’ve gotten rid of that long block of low frequencies that were across the bottom of the dialogue? If we consider that those low frequencies could have been caused by the early reflections of the actress' voice bouncing off and building up in the room, then we are already making our next step easier.

Step 2: De-Noise

When recording in any space, especially if it’s not acoustically treated, you will find that noise has made its way into your recording. This can be in the form of traffic, an A/C unit, or anything that will detract from the main focus which in this case is the dialogue. In our recording we have a strong amount of air, almost like a hiss that sits just above the spoken words. Therefore, what we aim to do on this step is clear most of the hiss in order to put the dialogue at the very front of it all, and to do that we will go to the Voice De-Noise.

For this particular example, I will be working on “Adaptive” mode but please be aware that I am not offering strict rules. Feel free to unclick “Adaptive” mode and manually focus on the frequency range you think works best. That being said, the reason why I’ve chosen to work on this mode is because, when active, it learns from the audio signal coming in and adjusts itself accordingly to any irregularities in order to produce a more consistent performance. If we take into account that in this case I am also considering reverb as noise, then "Adaptive" mode will work well with the dynamics of the human voice.

I usually set my “Threshold” to 0dB and my “Reduction” to 5db in order to preview how much I am affecting the dialogue. This gives me enough information to choose whether to increase or decrease the level of processing. In this example, I found that the sweet spot was around 10dB as it cleared most of the hiss that was over the dialogue, while also gently clearing some, but not all of the reverb.

One thing that I believe makes a big difference, is to split the work by doing multiple passes of small renders to the selected audio. This will give you the chance to see how far you are pushing the dialogue after every render, as well as introducing a smoother audio processing instead of a harsh one that can make the voice sound over-processed. Remember, it is very easy to go too far on the processing, so why not introduce it in small surgical doses?

With the sweet spot being between around 10dB, I did four renders of 3dB and then went back to just three renders. I decided to be cautious and consider that I may be taking too much out of the performance already, with still one other processing step left to do.

Step 3: De-Reverb

Now that we’ve cleared most of the sonic debris, for our final step we will be focusing solely on getting rid of the reverb and for that we will open the De-Reverb. Now, this step can be approached in the same manner as with the De-Noise. We don’t have adaptive mode but we have the “Learn” button. I like using this feature to have a starting point that I can then manually tweak but before that, I set my “Reduction” to 5dB and then click on “Preview”.

Now it is time to decide whether to up or down the "Reduction" levels accordingly and, in this particular case, is not an easy choice to make. The reduction levels at 9dB seemed to clear most of the reverb, but it also made the dialogue sound over-processed and compressed. In the end, I decided to do two renders of 3dB and one of 2.5dB, that way maintaining a more natural sounding performance.

At the start, the reverb was very strong and very noticeable compared to how it is now. However, in order to clean the dialogue of all the reverb I would’ve had to sacrifice sound quality, so the only choice I had was to keep some of the reverb in the dialogue. This is not ideal but at least it sounds natural and not overly compressed.

As sound engineers sometimes we have to make the best we can with what we have. We are expected to make choices and commit to them while working towards the best sound quality we can produce. In an ideal world, I would have chosen to record the lines again in a professional environment, but this would cost more money. It is part of our job to work within the budget and time frame of a project, and still meet the delivery deadline with our best work.

I hope you all find this helpful and will allow you to venture and experiment further with this amazing tool. Please remember to trust your ears and not take what I’ve written as a rule of thumb, because there is no such thing when it comes to sound.

Feel free to comment about your experiences with Izotope RX, what you would’ve done differently, or if I missed anything. I’m sure any information will be helpful to the community.