Separate the band from the event
Saying that a vocal hurts around 3.5 kHz can describe two very different problems. The entire recording may be thin and aggressive, or a resonance may jump out only on a few vowels and consonants. Static EQ is the more honest answer to the first problem. In the second case, leaving a cut in place can remove clarity from every syllable that was already working. Before choosing a processor, ask whether the frequency itself is wrong or whether an occasional event at that frequency is wrong.
FabFilter’s official documentation draws a useful distinction. A conventional dynamic EQ band moves the gain of the whole band according to its trigger. In Spectral mode, Pro-Q 4 reacts to specific frequencies inside that band when they cross the threshold and leaves the other frequencies alone. The screen still shows one bell, but responsibility is divided into smaller pieces underneath it. The practical promise is not more reduction. It is less disturbance to material that did not need repair.
Do not begin by looping only the harshest syllable. Loop that syllable with the healthy words immediately before and after it, then use a static bell to locate a candidate range. If a continuous cut of two or three decibels improves every phrase, spectral processing may be unnecessary. If the harsh moment settles but ordinary phrases step backward, return the bell to zero and try Spectral mode with a small negative dynamic range. This sequence keeps the diagnosis ahead of the feature.
Judge the first result in the sentence, not in solo. A consonant can become smoother while the front of the word opens late, or while breaths and room noise appear to move forward by contrast. Put the vocal back inside the instrumental and listen through the word after the problem. A useful setting lowers the abrasive event without changing the speed or expression of what follows.
Density defines responsibility, not prestige
The name Spectral Density can make a higher value look automatically more advanced. The manual’s definition carries no such value judgment. At lower settings, relatively broad frequency regions respond together; at higher settings, the reacting areas become narrower and more specific. Density is therefore not a resolution score. It decides whether one problem will be treated as a broad body or as a collection of thin events.
Break wide sibilant energy into very high Density and many narrow reductions can enter and leave quickly, making the surface feel finely agitated. Treat one narrow guitar resonance with very low Density and the pick attack and nearby harmonics may bow with it. BGC starts at the default, then deliberately auditions both extremes. Once you hear the point where the low side darkens the whole syllable and the point where the high side granulates the texture, a useful middle range becomes easier to identify.
In Auto mode, Pro-Q 4 derives threshold, attack, and release behavior from the incoming audio; the expanded panel lets you adjust them manually. Automatic does not mean final. Start by limiting the dynamic range to a small value, perhaps one or two decibels, and alternate between three passages that contain the fault and three that do not. If gain reduction remains busy during the healthy passages, question band placement and width before threshold. A sophisticated detector cannot rescue a badly chosen target without creating a more complicated mistake.
Set manual attack and release by the timing of pronunciation, not by the shape of the meter. An attack that is too fast can blunt the consonant’s beginning; one that is too slow lets the abrasive front pass. A long release may darken the next vowel, while a very short release can make the resonance tail appear to jump twice. For vocals, ask whether the word arrives at the same speed when you stop reading the lyric. For guitar, ask whether the sustain after the pick remains one breath. Those are more useful clocks than the graph.
Linear phase is not free, so decide inside the session
Pro-Q 4 automatically uses linear-phase processing for a Spectral band. Other ordinary bands can remain in the processing mode selected in the bottom bar, while spectral bands receive a separate Processing Resolution control. FabFilter’s manual explains that higher resolution extends low-frequency capability while adding latency, and that linear-phase processing can introduce pre-ring. The moment Spectral Dynamics is enabled, the audition must expand from the frequency curve to the time domain.
There is no reason to select a high resolution by reflex when treating upper-frequency harshness. FabFilter advises that Low or Medium resolution will usually be suitable for spectral work above 1 kHz. In a session where a singer is monitoring or a software instrument is being played live, the difference returns through performance response rather than a specification sheet. Before stacking instances across a mix, check whether the lowest practical resolution already solves the problem.
Make the comparison in three passes. First, use the plug-in bypass to alternate between faulty and healthy passages at the same output level. Next, play transient-rich neighbors such as drums, lead vocal, and an aggressive synth together and listen for any change at their front edge. Finally, close the plug-in window and play the song from beginning to end. A smoother analyzer trace matters less than whether the voice still walks forward when the chorus arrives.
Successful spectral dynamic EQ is rarely conspicuous. The lasting impression should not be that the harsh moment has been dramatically removed, but that no other moment paid for the correction. If that result does not arrive, compare clip gain, an edit, multiband compression, or a plain static EQ before pushing Density higher or range deeper. Precision is not measured by how much of a tool is used. It is measured by how much of the performance the tool knows to leave alone.