What an impulse response physically is
An impulse response is a recording, not a design. You excite a room with something close to a perfect impulse, usually a starter pistol or a deconvolution click, and record the microphone. What you get is the direct sound followed by thousands of reflections arriving at slightly different times and levels.
Convolution works because a room is linear and time-invariant. Multiply the input signal by that recorded response and you get what the microphone would have captured if the music had been played there. That is why a good IR sounds like a place rather than a reverb preset.
| Part of the IR | Typical range | What it controls |
|---|---|---|
| Direct sound | 0 ms | Dry level before any room |
| Pre-delay | 5-80 ms | How much of the transient survives |
| Early reflections | 10-120 ms | Apparent size and source position |
| Tail and RT60 | 0.4-8 s | Decay time and perceived distance |
Pre-delay, early reflections and the tail
Pre-delay is the gap before the first reflection reaches the microphone, and it exists because sound needs time to travel. Put the source two metres from the nearest wall at 343 metres per second and the first return arrives about 5.8 milliseconds later. That gap is the reason a room can feel large without smearing the attack that triggered it.
Early reflections follow as discrete taps, and their pattern and spacing tell you the room's shape more reliably than the tail does. A plate with a very short pre-delay and hard reflections sounds close and aggressive; a hall with 60 milliseconds of pre-delay and soft taps sounds large.
The tail arrives once the discrete structure dissolves into a dense wash, and its length is quoted as RT60, the time for the level to fall by 60 decibels. A 1.4 second RT60 is a small booth, 2.2 seconds a live room, and anything past 4 seconds fights the bus unless the arrangement is sparse.
- Short pre-delay for drums and percussion, so the attack stays intact.
- Longer pre-delay for vocals and lead instruments, so the words stay intelligible.
- Watch the early reflection pattern when choosing between two IRs of similar length.
- Trim the tail to where the arrangement stops, rather than running the full IR length.
The CPU and mono-compatibility trade-offs
Convolution is a multiply and accumulate per sample per tap. At 48kHz a one second stereo IR is 48,000 taps per channel, so four parallel reverbs with three second IRs is over a million multiplies per sample position, and most plugins handle this with partitioned FFT convolution rather than running it raw.
The practical consequence is that IR length is a CPU dial. If your session stutters, shorten the IR before you start removing reverbs from the mix. Longer IRs also carry artefacts from whatever the room captured, including its own noise floor, so a nine second IR is mostly a record of the room's hiss.
Mono compatibility is the other cost and it is less discussed. A stereo IR decorrelates its two channels, and summing them for a mono folddown or a phone playback produces partial cancellation wherever the channels oppose. The result is a reverb that sounds wide in stereo and noticeably thinner with the low end partly gone in mono.
Choosing and checking an IR
Judging an IR by its name is the fastest way to end up with the wrong room. Read the waveform. A tight, dense block with a clear leading gap is a small controlled space. A long, sparse, slowly decaying shape is a large hall, and it will tell you immediately whether it belongs on the bus.
Two checks are worth making before you commit. Check the mono folddown, either with a correlation meter or by summing to a single channel and listening, and check what the IR does to your low end, because convolution applies the room response to everything you send.
A third question is whether you need convolution at all for that job. A plate on sends and a short algorithmic room on the snare will cover many production situations at a fraction of the cost, and will behave more predictably across playback systems. Convolution earns its price when the character of a specific place is the point.
- Read the IR waveform before loading it, and check pre-delay, early reflection spacing and tail length.
- Start pre-delay from the distance you want the listener to imagine, not from a preset.
- Measure mono compatibility before shipping, not after someone complains about the phone playback.
- Shorten the IR to the shortest length that still reads as the space you want.
The bottom line
An impulse response is a recording of a room's answer to a single sound, and reading one tells you more than any reverb preset name will. Pre-delay keeps the transient yours, early reflections set the size, and the tail does the rest. The costs are real and worth measuring: CPU scales with IR length and sample rate, and a stereo IR can lose low end when summed to mono.
Hear the room, then check the weight
Retuning moves every reflection with the source. 432Hz MASTER runs the whole space at one ratio so the tail stays attached to the material.
Frequently asked questions
What is an impulse response physically?
A microphone recording of a short, loud broadband sound played in a room, usually a starter pistol or a deconvolution click. What you record is not the sound itself but the sum of every reflection arriving at the microphone, which is why it describes the room.
What is the difference between pre-delay and early reflections?
Pre-delay is the silent gap before the first reflection reaches the microphone, caused by the distance between source and wall. Early reflections are the discrete taps that follow it, usually within the first 50 to 120 milliseconds. Gap first, then discrete echoes, then the dense tail.
Why does reverb disappear in mono?
A stereo IR decorrelates the left and right channels, and summing them creates partial cancellation at whatever frequencies the two channels happen to oppose. The low end is usually the worst casualty. Check the correlation meter, or fold the reverb down early and listen on a mono foldback.
Is longer IR always better?
No, and cost goes up with length. A three second stereo IR at 48kHz is roughly 288,000 taps per channel, so four of them in parallel is a real CPU load. Use the shortest IR that gives the size you need, and trim the tail where nothing is playing.