What group delay actually is
Group delay is how long a frequency component waits inside a filter before it comes out. It is measured in milliseconds, it varies with frequency, and when it varies the relative alignment of the partials in your signal is no longer the alignment you recorded.
The arithmetic is simple. Phase deviation in degrees equals 360 multiplied by delay in seconds multiplied by frequency in hertz. A one millisecond delay is nothing at 100Hz, which is 36 degrees of shift, and enormous at 10kHz, which is 3600 degrees, or ten full cycles.
That is why filters sound fine on a bass note and wrong on a hi-hat. The same delay applies across the passband, but the phase error it creates grows with frequency.
| Processor | Typical group delay | What it costs |
|---|---|---|
| IIR shelf, gentle high-pass | Under 1 ms | Almost nothing, slight phase slope |
| Fast FFT equaliser | 2-8 ms | A little smearing above the cut |
| Linear phase equaliser | 15-60 ms | Full latency, clean transients |
| Multiband compressor | 5-20 ms | Latency, plus band-edge phase shifts |
| Lookahead limiter | 1-10 ms | Latency proportional to lookahead |
Linear phase and minimum phase
A minimum phase filter puts its change as early in the waveform as the mathematics allows, concentrating delay where the filtering happens, so a bell cut at 2kHz drags the leading edge of the transient with it. That is why minimum phase is the default: least delay, lowest CPU.
A linear phase filter takes the opposite approach. It applies a uniform delay across the entire bandwidth, so every partial of a drum hit shifts by the same amount and the transient shape survives intact. That uniform shift is the latency you pay.
Neither is more correct. They are different trade-offs, and which one wins depends entirely on what you are processing and how much latency the rest of your session will tolerate.
- Minimum phase on individual tracks keeps the session tight and keeps latency off the recording path.
- Linear phase on the mix bus and mastering chain is cheap in risk because nothing downstream needs to re-record.
- Stacking five minimum phase dynamics plugins is often worse than one linear phase compressor, because the delays add in series.
- Any plugin that claims to be transparent is trading something, usually latency, and the manual should say how much.
The price is latency, and it is real
At 48kHz one sample is about 20.8 microseconds. A linear phase equaliser set to its longest setting can add somewhere between 20 and 60 milliseconds, which is roughly 960 to 2900 samples of delay sitting between your input and your output.
That matters in three specific situations: when recording live, when splitting an overdub, and when any two tracks must line up sample accurately. It does not matter when the chain is at the end of the session, which is most mastering work.
Latency also has a perceptual cost. Play the same mix through a fixed delay of thirty milliseconds and it will sound slightly detached and slightly thin on transients, because the ear fuses very short delays but starts treating longer ones as a separate event.
What phase error actually costs you
Here is the part worth internalising. Two signals that differ only in phase carry the same energy over the same time period. An integrated loudness meter will barely register the difference. What changes is the peak, and more importantly the slope of the waveform as it climbs.
Definition is largely about how quickly a transient reaches full level. A kick that rises in three milliseconds sounds like a kick. The same kick, smearing over fifteen milliseconds by summing a minimum phase and a linear phase copy of itself, sounds like a soft drum sound with a low end that is hard to locate.
So when people say phase problems make a mix sound dull, they are usually describing this loss of transient definition, not a level change. Judge any filter on a solo kick and a solo snare, in bypass against active, before you judge it on the full mix.
- Solo one source and listen to the transient alone, not inside the arrangement.
- Bypass the suspect plugin and compare the waveform outline, not just the meters.
- Check summed pairs of doubled parts for comb filtering before blaming EQ.
- Move to linear phase only where the transient is worth the latency you are adding.
The bottom line
Phase is measurable, and the measurement is group delay in milliseconds. Linear phase buys correct timing at the cost of latency, and minimum phase buys latency at the cost of a smeared front edge. Phase error costs you transient definition far more than it costs you level, so judge filters on attacks rather than on how loud they are.
Check what the whole chain is adding
A full retune pass adds its own group delay. Listen through 432Hz MASTER with the bypass engaged so you are comparing phase and not tuning.
Frequently asked questions
Is linear phase always better?
No. A linear phase filter moves the whole signal by a fixed number of milliseconds, so the waveform shape survives, but everything it processes arrives late. For mastering and mix bus work that trade is usually worth it. For a live vocal chain it usually is not.
Does phase affect loudness?
Barely. Two signals that differ only in phase have the same energy over time, so an integrated loudness meter barely moves. What changes is peak level and how sharply the waveform reaches full amplitude, which is exactly what the ear reads as definition.
What causes phase problems in a mix?
Doubling a part with a fixed time offset, miking the same source from two distances, parallel processing with different algorithms, and high-pass filters with gentle slopes all add group delay. The offset doubles and miking usually do the most audible damage.
How do I see phase problems in a session?
Look at the waveform of the summed pair rather than at each track alone. If the peaks of two supposedly identical takes do not sit on top of each other, they are comb filtering, and nudging one by a few milliseconds until the sum thickens is faster than any plugin.