The Sonnox Oxford SuprEsser was designed primarily to be the last word in ‘de-essing’ applications. While we worked on creating the best possible de-essed sound, we found that we needed more control than was available on conventional De-Essers. Most de-essing work can be carried out in ‘simple’ mode but sometimes, to do the job properly, it is necessary to utilise all the controls of the underlying engine — a full-blown dynamic EQ, or frequency- conscious compressor.
Thus a new concept was born — a simple and intuitive De-Esser, plus access to a much more sophisticated frequency-specific compressor, giving complete control over aggressive frequencies wherever they may be in the audio spectrum. With the Oxford SuprEsser you can take out unwanted frequency peaks as and when they occur — unlike applying a fixed EQ to the whole channel.
As a De-Esser, the Oxford SuprEsser is designed for the treatment of sibilance and fricatives in vocals, and the treatment of unwanted whistles and ’spirant’ artefacts associated with wind instruments. However, it can equally be applied to removing low-end plosives and thuds from over-close vocal work, without affecting nearby components in the frequency spectrum, keeping the low-end intact. For the most natural sounding results, the Oxford SuprEsser de-esses only the frequency band you select — so you won’t end up with an over-de-essed lisp- like voice with all the high frequencies gone!
Detailed visual feedback is provided by a highly intuitive graphical display, allowing quick identification of the frequencies that need treatment, and where to set the threshold. The threshold level and peak-hold level of the user-definable band are shown on the graph, alongside the FFT (Fast Fourier Transform) display of the narrow band signal, which includes retention of the peak level and the frequency containing the most energy.
At the heart of the Oxford SuprEsser is an enhanced version of the compressor section of the Sonnox Oxford Dynamics plug-in, which is renowned amongst professional users for its consistent delivery of the precise and transparent control of peak signals. Around this is built a pair of crossover filters to make the compressor react only to the defined frequency band. These linear-phase filters are modelled on the filters from the Sonnox Oxford EQ, making the Oxford SuprEsser useful for precise mastering as well as mixing work.
Three listen modes allow the user to listen to the Mix, the output of the bandpass filter (Inside), or the output of the band-reject filter (Outside).
Careful thought has been put into making the Oxford SuprEsser extremely easy and quick to use. The screenshot on the title page shows the plug-in as it appears when first activated. Once the frequency band has been defined, simply lower the threshold fader until the gain reduction meter starts to kick in. The plug-in then automatically tracks the general signal level and the threshold follows accordingly, so that it gives the same relative amount of gain reduction as the signal level rises or falls. This is perfect for vocals where, for example, a vocalist is louder in the chorus than the verse, but you want to apply the same relative amount of de-essing, but don’t want to over de-ess in the chorus. This Auto-Level-Tracking mode lets the Oxford SuprEsser do all the hard work. It can also be switched off for a fixed threshold.
In its default mode of operation, the Oxford SuprEsser feeds the defined Band signal to the compressor to affect only this narrow-band signal. The result is that the compressor reacts only to specific frequency components when they reach a specific threshold, and applies compression only to these specific frequency components, leaving the rest of the spectrum untouched. Audio Wide mode can also be selected to allow full-band compression reacting to only specific frequencies, or any combination of Band and Wide defining when the compressor reacts, and what it compresses.
The Oxford SuprEsser comes complete with many presets to act as good starting points; alternatively, the advanced section provides full access to all controls for precise correction, or even creative use.
Highly featured professional De-Esser
Linear-phase Dynamic EQ
Transparently controls aggressive frequencies
Automatic level-tracking follows energy level (eliminating the need to automate threshold)
Large intuitive graphic display makes finding frequencies very easy
Full spectrum operation (20Hz–20kHz)
Three different ‘Listen’ modes
Very easy to use
Advanced mode for ultimate control of the Dynamic EQ
Many creative as well as corrective uses
Presets provide good starting points
Basic SuprEsser signal flow
The Oxford SuprEsser contains a pair of mutually opposing filters — by default one is a narrow bandpass filter, the other is the complementary narrow band-reject filter. This results in one signal path containing just the contents of the band of interest, and another signal path containing the input signal with this band entirely removed; when mixed back together in equal ratios, you get the original signal.
The bandpass signal is usually fed to the compressor, both to the sidechain and to the main signal path, so that it is just this signal that triggers gain reduction, and it is just this signal that is affected by any gain reduction.
The bandpass/reject filters will switch automatically to other EQ types when conditions warrant it, giving a total of four EQ types. For example, when the ‘Width’ control narrows the bandpass filter to its minimum, a High-Q notch filter is invoked that provides a much better reduction of a very narrow band of frequencies. When either side of the band window touches the end stops (20Hz or 20kHz), the bandpass filters change to LF-Cut or HF-Cut as necessary.
In addition to the four EQ types, there are four different compressor modes that are concerned with which signal is passed to the sidechain and which signal is passed to the main input of the compressor. See Section 2.3 Advanced Operational Modes for more information on this.
The following diagram provides a more complete signal flow to illustrate the elements required for advanced modes of operation:
Advanced SuprEsser signal flow
The ‘Wet/Dry’ control is a frequently requested feature that allows you to add back the uncompressed signal to a highly compressed signal to add in some punch. As you can see from the diagram above, the Wet/Dry blend control is implemented inside the dynamics module, and therefore only operates as expected when you are listening to the ‘Mix’, or ‘Inside’.
The linear-phase filters used by the Oxford SuprEsser require an ‘Impulse Response Kernel’ to model the response of the internal Oxford Filters. The size of this kernel (as measured in samples) determines both the plug-in delay and the accuracy of the model, which in turn affects the performance, especially at lower frequencies.
Better performance at low frequencies, but longer overall plug-in delays
Adequate performance at high frequencies, and shorter plug-in delays
While small kernels are adequate for high frequency work such as de-essing, the performance at low frequencies will cause poorly defined filter slopes, and poor separation of ‘Inside’ from ‘Outside’.
We have found that host applications are generally not able to readily adjust their delay compensation engines if the kernel size is adjusted dynamically at run time. The Native Oxford SuprEsser is therefore released as three separate plug-ins; each with a different fixed kernel size.
This is the standard version, and has a kernel size of 2048 samples. This is suitable for work across the entire frequency range at 44.1kHz sample rate, and thus is the standard plug-in to reach for, especially for the mixing stage of a project. Depending on your audio buffer size, the delay or latency will be somewhere in the region of 1044 to 3092 samples. See the section below for information on reducing the delay.
This is the large kernel version, with a kernel size of 8192 samples. This version gives superior resolution at the lowest frequencies, and thus is particularly suitable for low frequency mastering work. Remembering that, as you increase the sample rate, the resolution at the low end will be correspondingly reduced, this version is particularly suitable for use at higher sample rates, especially 176.4kHz or 192kHz. This version will have a very significant delay that can be beyond the ability of Pro Tools HD to automatically compensate for. The true delay is reported correctly in the track delay information, and depending on your audio buffer size, the delay will be somewhere in the region of 4116 to 12308 samples. See the section below for information on reducing the delay.
This small kernel version has a kernel size of 512 samples.This version gives superior performance in terms of having a small delay, and thus is more suitable for live de-essing work at low sample rates, and for the tracking phase of a project, where you are laying down new tracks from midi instruments, and you don’t want a significant delay between what you are playing and what you are hearing. The smaller kernel size means the resolution at low frequencies will be poor, and so cannot be used much below 400Hz. Depending on your audio buffer size, the delay or latency will be somewhere in the region of 276 to 3072 samples. See the section below for information on reducing the delay.
This version has an even smaller kernel size of 128 samples, in order to further reduce the processing delay. Due to this, the filter range has been reduced to 1 kHz minimum, in order to prevent poor low frequency performance from producing unexpected results.
For further information, see Section 5.
The plug-in delay produced by the Oxford SuprEsser depends both on the kernel size/resolution and the audio block size.
The block size is the size of the sample buffers passed to the plug-in by the host, and is usually specified in your audio hardware preferences/configuration/setup page. The reason that the plug-in delay depends on the block size is that the plug-in must accumulate a whole kernel sized block of samples before it can process them.
In order to ensure the minimum plug-in delay, make sure the block/buffer size is the same as or greater than the kernel size/resolution setting!
For example, if the kernel size is set to 512, i.e. you are using the Low Latency version, and if your block/buffer size is 512 or 1024, this will ensure that the plug-in produces the minimum delay.
If you use a smaller block size than the kernel size, the overall delay of the plug-in will go up, not down.
The Oxford SuprEsser uses a process called convolution to implement filtering, a process that is expensive on CPU resources. When using very large kernel sizes, in combination with small audio buffer sizes, it is possible for the plug-in to take longer than an entire audio block to complete processing, with the result that (from a monitoring point of view) the playback breaks up. This can manifest as obvious clicks and pops, or more subtly as occasional quiet clicks.
If you experience clicks, keep in mind that these are monitoring only. They will not be present when bouncing down your mix.
If you experience clicks, increasing your audio buffer size will resolve the issue. Once again, the best buffer size for the Oxford SuprEsser is the same as the kernel size (or resolution). Generally speaking, the only reason for small audio buffer sizes is for tracking, or live working requirements.
The plug-in issues a red warning label next to the Sonnox button when you are using a combination of small buffer size and large kernel size that we have found to generally result in clicks. This is not a reliable indicator, however; it completely depends on your system CPU speed, sample rate, and other factors.
There are four major operational modes selectable via the TRIGGER and AUDIO buttons in the Advanced Mode (MORE:) section, as described below.
Gain ducking occurs only in the narrow band, triggered by the narrow band.
In this mode (the default) both the trigger signal (to the compressor sidechain) and the main compressor signal is the narrow-bandpass-filtered signal.
The output of the compressor is then fed to the crossover block (the ‘Listen’ section) for mixing back in with the version of the input signal that has been filtered with the corresponding narrow-band-reject filter. The result is that the plug-in affects only a narrow frequency band, triggered by that same frequency band, and does not affect the rest of the frequency spectrum.
In this mode, when you also use the Auto Level Track function (AUTO IN), the threshold follows the general level of the signal post the band-reject filter, (ie. everything except the troublesome frequencies).
Gain ducking of (wide) input signal, triggered by narrow band signal. Equivalent to compressor in side chain EQ mode.
In this mode, the trigger signal (to the compressor sidechain) is the narrow-bandpass-filtered signal, and the main compressor signal is the (wide) delayed input signal.
This means that when the narrow band signal triggers gain ducking, the ducking occurs over the whole frequency spectrum, as in traditional sidechain EQ compression.
This mode offers one major advantage over the equivalent mode in the Oxford Dynamics compressor — you have a WET/DRY control on the output stage, so you can mix the original signal back into the compressed signal.
In this mode, when you also use the Auto Level Track function (AUTO IN), the threshold follows the general level of the signal post the narrow-band-reject-filter, (ie. everything except the troublesome frequencies).
Furthermore, the crossover controls (in the LISTEN section) have no effect on the signal because the output of the compressor is already a complete wide-spectrum signal.
Gain ducking occurs only in the narrow band of the spectrum, triggered by the wide input signal.
In this mode, the trigger signal (to the compressor sidechain) is the delayed input signal, and the main compressor signal is the narrow-bandpass-filtered signal.
This mode is useful, for example, when a broad-spectrum trigger is required to trigger reduction of a specific frequency. One example could be to process a kick drum/bass combination to reduce a rattle or click whenever the kick occurs.
In this mode, when you also use the Auto Level Track function (AUTO IN), the threshold follows the general level of the delayed input signal.
Equivalent to compressor with no side chain EQ.
In this mode, the trigger signal (to the compressor sidechain) and the main compressor signal are both fed from the delayed input signal. In other words, you now have an ordinary compressor with no frequency specific sidechain processing.
This mode offers one major advantage over the Oxford Dynamics compressor — you have a WET/DRY control on the output stage so that you can mix the original signal back into the compressed signal. This can be used to create that unique sound in which a fully compressed signal (with no dynamic headroom) has some punch added back into it.
In this mode, the LISTEN section has no effect on the signal because the output of the compressor is already a complete wide-spectrum signal.
Auto Level Tracking mode is enabled by selecting the LEVEL TRACKING/AUTO IN button. This is ON by default.
The purpose of this mode is to automatically adjust the threshold level to follow the general signal level of the wide-band input. This means that if some vocals wander from loud to quiet, the same amount of relative gain reduction is applied when a transient peak occurs above the general level.
Generally speaking, once you have isolated as narrowly as possible a troublesome sound in the frequency spectrum, you will see the red peak-hold marker, inside the bandpass filter on the graphic display, indicating the peak level of the troublesome sound. Now enable Auto Track mode, and bring the THRESHOLD level down until its corresponding line on the graph is below the red line. You will then see the gain reduction meter (ATTENUATION) starting to show gain reduction.
If the general signal level of the material changes, the threshold will now follow that level, so that the gain reduction meter should continue to indicate the same amount of reduction of the peaks.
The algorithm to track the general signal level is not as simple as it appears. First, one of the most important characteristics of speech is its staccato nature, and the frequent pauses or silence. If the threshold truly followed the signal level, each time there was a pause broken by a word starting with a consonant, the threshold would have fallen too low, and the consonant would be over gain-reduced before the threshold recovered its normal level.
To get around this challenge, the algorithm implements a working range of 24dB, and any sample value outside this window is not included in the calculation of the general signal level. If you were to intermittently mute the input signal, you will see that the threshold effectively just stays where it was when there was an active signal, and continues tracking again when the active signal is restored. Yet if you slowly fade out the input signal level, the threshold will follow, only giving up after 36dB of gain reduction.
Auto Level Tracking therefore works well with continuous material, or with conversation broken up with frequent pauses.
By default, the Oxford SuprEsser uses a bandpass filter (along with its inverse filter, the band-reject filter) to isolate the audio in the frequency range selected. The FREQUENCY and WIDTH controls defines the low and high edges of the filters used, and the SLOPE/Q control defines how quickly a signal blends from gain-ducked to non-gain-ducked in the frequency spectrum. The effect of less steep slopes is to give a smoother blend between gain- reduced and non-gain-reduced regions of the frequency spectrum, at the cost of a less specific trigger.
In this mode, the bandpass filter is created by using pairs of Oxford R3 Filters, giving a total of up to 72 dB per octave of separation.
There are three other modes possible, which automatically switch into operation in certain situations:
When you reduce the WIDTH control all the way down to 0.2 octaves, you activate High-Q Notch mode. In this mode, instead of using R3 Filters, the plug-in uses four Oxford EQ Type-2 Filters, giving a total of 80dB of gain reduction at the centre point of the notch. This mode is useful when the band of energy you are interested in reducing is extremely narrow, or virtually a single frequency, like a whistle.
In this mode, if desired, you can reduce the Q of the filters by reducing the SLOPE control. The effect of this is to give a smoother blend between gain-reduced and non-gain-reduced regions of the frequency spectrum, at the cost of a less specific trigger.
When you reduce the left edge of the bandpass filter all the way down to 20Hz, you activate LF-Cut Filter mode. This means that the lower edge of the bandpass window is effectively at 0Hz, and this is useful, for example, when working on eliminating sub-bass plosives and thuds/booms that contain DC components.
In this mode you can think of operation as an LF-Cut Filter that activates only when the signal reaches a certain threshold, in which the cut-off frequency is defined by the upper edge of the bandpass window.
When you increase the right edge of the bandpass filter all the way up to 20 kHz, you activate HF-Cut Filter mode. This means that the upper edge of the bandpass window is effectively at infinity Hz, and this is useful, for example, when you want to duck the entire HF part of the signal.
In this mode you can think of operation as an HF-Cut Filter that activates only when the signal reaches a certain threshold, in which the cut-off frequency is defined by the lower edge of the bandpass window. This is how some more primitive De-Essers work.