How to EQ your microphone

This guide uses the parametric EQ in patchd, a free effect on every channel, and this guide walks it the way patchd organizes its other processors, beginner to intermediate to advanced, so you can stop at any tier and have a working curve. If you want the concept first, start with what an EQ is, then come back here and point one at your voice.

The default bands and the four voice moves

A fresh EQ in patchd ships with three bands: a high pass at 80 Hz (Q 0.707), a peak at 3 kHz (gain 0, Q 1.0), and a high shelf at 10 kHz (Q 0.707). That is not an accident. Those three bands sit on the three spots where a spoken voice most often needs help, and together with one optional cut they cover the entire recipe:

• Rumble: high-pass at 80 to 120 Hz. Desk thumps, air conditioning, traffic, and the hum your mic arm transmits all live below your voice. A high pass removes everything under its frequency, and since almost nothing useful in speech lives down there, this is the closest EQ gets to a free win.
• Boxiness: gentle cut at 250 to 500 Hz. If your voice sounds like it was recorded inside a cardboard box, the box is here. Cut 2 to 4 dB with a medium-width peak band and the cardboard opens up.
• Presence: boost at 3 to 6 kHz. Consonants and intelligibility live in this range. A wide boost of 2 to 4 dB is the difference between "audible" and "clear," which is why the default peak band parks at 3 kHz waiting for you.
• Air: high shelf around 10 kHz. A shelf raises everything above its frequency by the same amount. A couple of dB up here adds the expensive-studio sheen; more than that adds hiss with a sheen.

The working method is three steps. First, sweep to find the problem: push a peak band up 8 dB or so, narrow its Q, and drag it slowly across the range while you talk. The problem frequency will jump out and announce itself. Second, cut narrow: flip that boost into a cut right where the problem lives. Third, boost wide: broad, gentle boosts in the presence and air zones sound musical; narrow boosts sound like a phone call from 1996. Test the result against your own ears with a quick mic test.

Q, gain, and band types

The intermediate skill is knowing what each control trades away. Every band has three: frequency (20 Hz to 20 kHz), gain (plus or minus 18 dB), and Q (0.1 to 10).

• Q is the width knob. Low Q means a wide, gentle hill that affects an octave or more; it sounds musical and forgiving. High Q means a surgical spike that touches almost nothing around it. The rule from the beginner tier generalizes: cuts tolerate high Q because the ear forgives a missing sliver, but boosts want low Q because the ear immediately notices an added one. The 0.707 on the default high pass is the Butterworth value, the flattest passband a biquad can give you.
• Gain goes to 18, use 4. The range is plus or minus 18 dB so you can hear extremes while sweeping. Finished voice EQ almost never leaves single digits; if a band needs more than 6 dB, the real fix is usually mic position, not the EQ.
• Band types are shapes. A peak is a bump or dip at one frequency. A shelf raises or lowers everything past its frequency, in either direction. A pass filter removes everything on the wrong side of it entirely, no gain knob involved. patchd also offers notch, bandpass, and allpass bands for surgical work, but the voice recipe above uses only passes, peaks, and shelves.
• Cut before you boost. Cutting a problem and raising the channel level does the same job as boosting everything around the problem, but with less added gain and fewer chances to hit the ceiling. Boosts spend headroom; cuts refund it.

The math, straight from the engine

Each band is a biquad filter built from the RBJ cookbook formulas, and the first thing the engine does with your gain knob is turn it into a linear amplitude:

Your frequency and Q become the other two ingredients. The center frequency turns into an angle around the digital unit circle, and Q sets alpha, which decides how wide the band is:

From A, w0, and alpha the engine assembles five coefficients per band (for a peak, b0 = 1 + alpha * A and a0 = 1 + alpha / A, and so on down the cookbook). To draw the curve you see on screen, the app evaluates each band's magnitude at every frequency along the axis:

The part worth knowing: the frontend and the audio engine compute their coefficients from the same formulas, kept in lockstep by construction. The curve you drag is not an artist's impression of the filter, it is the filter, evaluated point by point. What you see is literally what you hear.

Where the EQ sits in the chain

EQ earns its keep in two different chair positions. As cleanup, it goes early: the high pass removes rumble before noise suppression, the gate, and the compressor ever see it, so none of them waste effort reacting to a truck driving past. As tone polish, a second EQ can sit after the compressor, because compression changes your frequency balance and the polish should judge the finished product. Either way, the high pass comes first in the chain, always: everything downstream works better when the rumble is already gone.

In patchd the EQ is a free node in the effects rack on every channel. Drag bands directly on the curve, hear the change in real time at single-digit milliseconds of latency on a small ASIO buffer, and the result is what every destination hears: your headphones, your stream bus, all of it. patchd is in development now. Join the waitlist and we will tell you the moment it is ready to install.