This month's project is another addition to the Tanrak system: a narrow-band de-esser. Much used in professional recording, this sort of system can not only be used for limiting the absolute energy level of the sibilant content of vocals, but also by means of automatic level tracking, make sibilants less obtrusive over a wide range of vocal levels.

When trying to get the highest possible levels onto tape to make the best use of available dynamic range, a compressor is obviously the first item of equipment that would be pressed into service, especially on vocals, where levels are notoriously unpredictable. The degree to which the microphone signal can be compressed will depend firstly on how noticeable the limiting action becomes, and secondly how much the equipment, microphone and ambient noise is boosted by the compressor before it becomes a problem, although the latter can be combatted to a certain degree by using a noise gate after the compressor.

It's surprising just how much energy is concentrated in the upper reaches of the spectrum occupied by the sibilants 'f', 'sh', 'th', 'v', 'z', and most particularly 's'. The pre-emphasis used on tape machines compounds the problem, often making sibilants the the most energetic part of the vocal track and thus determining the operating level. Try talking into a Mic, and watch your tape machine VU meter peak when 'esses' are pronounced.

Sibilants can often become overly obtrusive in a mix, particularly when close miking is employed to maintain separation between musicians, or to reduce the effect of other ambient noise. Equalisation can help here, but to achieve any significant improvement the depth of correction would become obvious. Even if sibilance is not obtrusive as such, vocals often sound cleaner and smoother with sibilants attenuated.

A compressor with an equaliser patched into it's sidechain can be used to respond particularly to sibilants, but it will also tend to close down on the voiced sibilants 'v', 'z' and 'the', which would rather give the game away. The other deficiency with this solution is that the limiting threshold would be unvarying and absolute, leaving sibilants under the threshold untreated and thus nearly as prominent.

De-Essing

The principle of de-essing is to amplitude limit the band of frequencies occupied by the sibilants. The design presented here splits the 'ess' band from the incoming signal, using this not only for the sidechain signal to control limiting, but also in the VCA circuit so that only this band is attenuated, unlike the compressor-EQ solution. Figure 1 shows how the unit tailors the frequency spectrum at differing levels.

Dedicating a unit to de-essing also allows many of it's parameters to be optimised for this function, such as a very fast release time, (just 20mS in this case). To make operation fast and simple, all parameters have been fixed except the threshold level, and even this can be semi-automated by switching in the Track function. This causes the limiting threshold to be held relative to the average amplitude of the main signal, rather than at an absolute level. De-essing can then be achieved over a wide range of vocal levels, without requiring constant readjustment of the threshold control.

Circuit

The circuit diagram shown in Figure 2 reveals that the input signal passes through two inverting stages formed by IC1a and b to the output. The output is also routed to the sidechain circuit via two filters; a band-pass filter around IC3a, and a high-pass filter effectively formed by C5 and R11. The outputs from these two filters are mixed at the inputs of the OTA IC2a. When the OTA is energised by the sidechain, current feedback is set up around IC1a, reducing it's gain at frequencies within the band determined by the filters. Two filters are employed to give a response with a sharpish dip at around 6kHz to attenuate the 's' sound, superimposed with a less aggressive but wider band characteristic to attenuate the non-'ess' sibilants.

The output from the bandpass filter, IC3a is also used to feed the detection circuitry via C9 and the Threshold control, VR2. The filtered signal then becomes rectified by IC3b and is passed to the comparator, IC3c. When the rectified signal exceeds the threshold determined by R31 and 32, the output of IC3c pulses positively, charging C11 via R29. The voltage thus developed across C11 produces the control current for the aforementioned OTA via R21 and 22. The reduction in gain caused by the now increased transconductance of IC2a will then lower the output level to the point where the system is balanced. When De-essing is taking place, IC2c and TR1 cause the LED D1 to illuminate by driving it between the supply rails.

When the circuit is De-essing, the long time-constant of C11 and R21/R22 ensure that there is no ripple signal modulating the limited 'ess'. When the rectified signal falls below the threshold however, C10 is allowed to discharge, switching TR3 off and thus causing TR2 to discharge C11 fairly rapidly using a constant current, resulting in a very short release time.

In the Absolute mode, the feedback set up around the rectifier IC3b via the other half of the OTA IC2b is held constant by current injected into the control port via R36, R35 and TR4. In the tracking mode however, IC3d rectifies the untreated incoming signal onto C8. This voltage allows the gain of the rectifier stage IC3b to be varied in sympathy with the amplitude of the incoming signal by altering the control current via TR4. The threshold of De-essing is then a set fraction of the incoming signal, rather than being an absolute level.

R2 is used only during the setting-up procedure, allowing a fixed bias current to be modulated by the incoming signal via the calibration links so that any offsets on IC2a can be nulled out.

(Click image for higher resolution version)

Construction

Building the De-esser using the high quality kit should present no problems, especially since, by exclusive use of PC mounting connectors, switches and potentiometers, there's no interwiring to do. The first step in construction is to insert, solder and crop the resistor leads, populating the PCB according to the parts list, and the overlay printed on the PCB itself. Bending the leads outward at 45° prior to soldering will hold the components in place without running the risk of shorting together a pair of pads. Solder the 14 links in place using resistor lead off-cuts, at the positions shown dotted on the overlay. At the two positions marked 'CAL', use looped links which can be easily removed later. Do not link the position marked 'UNCAL'. Taking care with orientation, locate and solder the diodes D2-7 and transistors. The IC sockets come next, making sure that they are pressed down onto the PCB whilst soldering, but leaving the ICs themselves out until later. Now insert and solder the capacitors, taking care with the polarity of the electrolytic types. Solder in the preset VR1, leaning it back slightly to ease access later. The buss connector and the two jack sockets can then be soldered whilst holding them firmly down onto the PCB. A piece of foam rubber laid on the bench comes in handy for holding connectors and the like in place on up-turned PCBs during soldering.

Trim the pot shaft to 8mm from the bush using a hacksaw whilst holding the shaft in a vice, or just use a pair of cable cutters. Fit a PC bracket to the pot and locate into its PCB position, but don't solder at this point. After determining the correct orientation of the LED, bend it's leads down at right angles, 4mm from its body and locate into the PCB without soldering. Screw one nut onto each toggle switch and locate into the PCB, again without soldering. Place shakeproof washers on the switches and pot, then offer the front panel up, feeding the pot and switch bushes and LED dome into the appropriate panel apertures. The panel is then fixed in place by means of the pot nut which should be fully tightened. Only finger tighten the front switch nuts however, leaving the final securing to the rear nuts, which should be screwed up against the rear of the panel. The pot, bracket, switches and LED can now be soldered, after making sure that they are all fully home, and that the panel is square to the PCB.

Spend some time now to check over the assembly very carefully, especially on the track side where dry joints and solder splashes are all too common, even for the experienced constructor. When you are completely satisfied with the assembly, load the ICs into their sockets, being careful with orientation. Note that IC3 is fitted around the other way from all the other ICs. Finally, fit the knob and cap so that the marker line of covers the scale evenly, with equal 'dead-band' at each end, then push on the toggle switch lever covers.

Setting Up

With the CAL links in position, slot the completed module into the sub-rack with the module to it's right removed. Apply the rack power, turn the Threshold control fully clockwise and inject a constant sound signal of some kind, such as from a synthesiser at a fairly high level. Monitor the output and carefully adjust the preset VR1 until the level of the distorted signal breaking through is at a minimum. It should be possible to almost completely eradicate any signal breakthrough. Once this has been done, the module can be withdrawn from the sub-rack and the CAL links removed. The UNCAL link must now be soldered in place. The unit should now be ready for use.

In Use

Connections to the unit can either be via the sub-rack's linking system or by means of jack leads in and out. The position of the module in the sub-rack is not critical, although it would make sense to place it to the right of the Comp/lim module, before any effects. De-essing could quite advantageously be used to supplement normal compression; indeed the combination of Comp/lim and De-esser would make sense for every vocal recording session or live performance.

It must be stressed that this is a voice-only device for use either on a pre-amplified mic signal (such as on a desk insert point) during recording, or perhaps less ideally on mixdown, again using a desk insert point. Sounds other than voice present in the signal would not only confuse the de-esser, but would make it's action audible.

If the module is required just to extend headroom by limiting the maximum sibilant level, then the Absolute mode should be selected and the Threshold control adjusted so that the loudest 'esses' cause the indicator to light fully. Lower level sibilants will then be found to cause less bright illumination of the indicator, and will be attenuated to a lesser degree.

For general sibilant reduction, the Tracking mode will be found easier to use. Again, the Threshold control is adjusted for a bright indicator during 'esses'. Now however, 'esses'will be attenuated regardless of the vocal amplitude level.

The price inclusive of VAT and postage (within the UK) is £41.95 in kit form, or £58.95 ready assembled and tested. Further information on the modular rack system can be obtained from the address below. The de-esser module is available from: Tantek, (Contact Details).