If you have ever used Altiverb, Waves IR1, or Logic's Space Designer, you have used convolution reverb. The technology is powerful enough to make a vocal recorded in a bedroom sound like it was performed in a concert hall in Vienna. It does this not by simulating the physics of a room with delay lines and filters, but by literally recording the room's response to sound and applying that response as a filter.
The concept is elegant. The execution is where things get interesting.
What Is an Impulse Response?
An impulse response (IR) is a recording of how a system reacts to a brief, sharp sound. In acoustic terms, it is the sound of a room "answering" when you fire a starting pistol or pop a balloon. That answer contains everything the room does to sound: the early reflections off nearby walls, the buildup of reverberant energy, the gradual decay as that energy is absorbed by surfaces, and the frequency-dependent damping that gives each space its unique character.
Mathematically, an impulse is a signal that is zero everywhere except at one instant, where it has infinite amplitude and unit area. In practice, we approximate it. A balloon pop, a starter pistol, or a sine sweep deconvolved back to an impulse all produce a usable IR. The quality of that approximation determines the quality of the reverb.
Think of it this way: the impulse response is the room's fingerprint. No two rooms have the same one. A cathedral and a parking garage might both have long reverb times, but their impulse responses will look and sound completely different because the shapes, materials, and geometries that produce those reverbs are different.
How Convolution Works
Convolution is a mathematical operation that combines two signals to produce a third. In audio, we convolve the input signal (your vocal, your guitar, your drum track) with the impulse response (the room's fingerprint). The output is the input signal as though it were played in that room.
The math is computationally expensive. For every sample of input audio, the convolution algorithm multiplies it by every sample of the impulse response and sums the results. A three-second IR at 44.1 kHz contains 132,300 samples. That means 132,300 multiplications and additions for every single sample of output audio. Modern CPUs handle this with fast Fourier transform (FFT) techniques that work in the frequency domain rather than sample by sample, but the principle is the same.
The key insight is that the IR contains all the information about the room. There is no reverb algorithm to tune, no pre-delay knob to adjust, no damping parameter to set. The room did all of that when the IR was recorded. What you hear is what was captured.
Convolution does not simulate a room. It applies the room's actual acoustic response to your audio. The room is the algorithm.
Convolution vs Algorithmic Reverb
Algorithmic reverb, the kind built into most DAWs and plugins, constructs reverb from building blocks: delay lines for early reflections, allpass filters for diffusion, lowpass filters for high-frequency damping, and feedback loops for decay tail. The developer tunes these blocks to sound good. Valhalla Room, Lexicon PCM Native, and FabFilter Pro-R are all excellent algorithmic reverbs that give you extensive control over the sound.
Convolution reverb trades control for authenticity. You cannot adjust the room size or the damping in a convolution reverb, because those properties are baked into the IR. What you get instead is the exact sound of a specific, real space. If the IR was recorded in the Boston Symphony Hall, the reverb sounds like Boston Symphony Hall. Not approximately like it. Exactly like it.
Each approach has its place. Algorithmic reverb excels at creating sounds that do not exist in reality, or at giving you musical control over the reverb character. Convolution excels at realism, at placing a sound in a real-world context, and at recreating the sound of hardware that would be impractical or impossible to own.
Recording Impulse Responses
The quality of a convolution reverb is entirely dependent on the quality of the impulse response. Recording a good IR is a craft that requires patience, specialized equipment, and an understanding of acoustics.
There are three primary methods for capturing an IR:
1. Transient Method
The simplest approach. Fire a starting pistol, pop a balloon, or clap a clapper board, and record the result. The recording is the impulse response. This method is fast and intuitive, but the "impulse" is not a true mathematical impulse. The frequency content of a balloon pop is not flat, which means the resulting IR has a coloration that does not accurately represent the room's frequency response. It works for quick capture but is not suitable for professional libraries.
2. Sine Sweep Method
The industry standard. A sine wave sweeps from the lowest to the highest frequency the speaker can reproduce, typically over 10 to 30 seconds. The sweep is played through a speaker in the room and recorded with a microphone. The recording is then deconvolved mathematically to produce the impulse response. Because the sweep spends equal time at each frequency, the resulting IR has a flat frequency response and accurately represents the room. The long duration also provides a high signal-to-noise ratio, which is critical for capturing the full decay tail.
3. Maximum Length Sequence (MLS)
A pseudo-random noise signal that is played through the room and recorded. The IR is extracted by cross-correlating the recording with the original MLS signal. This method is less common today because sine sweeps generally produce cleaner results, but it remains useful for measuring systems where sweep playback is impractical.
Capturing Hardware Reverbs
Recording the IR of a hardware reverb unit like a Lexicon 480L or EMT 252 is a different challenge. You are not capturing a room, you are capturing a signal processor. The process involves sending a stimulus signal through the unit, recording the output, and deconvolving the result.
The critical decision is whether to use the unit's analog or digital I/O. Digital capture is cleaner and easier, but it bypasses the unit's D/A and A/D converters. Those converters contribute significantly to the character of vintage hardware. The Lexicon 480L's converters add a particular warmth and grain that is part of its sound. At Acousticas, we record all hardware IRs through the analog path, using the unit's own converters, to preserve that character.
This means the stimulus signal must be converted to analog, sent through the unit's analog inputs, and the output converted back to digital. Each conversion adds noise, and that noise must be carefully managed. We use high-quality interfaces and average multiple takes to reduce noise.
For units like the Lexicon 224XL, the challenge is even greater. The 224XL's converters respond differently to different stimulus signals, so custom sweeps must be designed for each unit. Recording the 224XL library took two months of work.
Using IRs in Your DAW
Once you have an impulse response, using it is straightforward. Load it into any convolution reverb plugin. Most modern DAWs include one: Space Designer in Logic Pro, Pro Reverb in Pro Tools, and Reverb in Cubase all support WAV-format IRs. Third-party options like Altiverb, Waves IR1, and LiquidSonics Reverberate offer more features and larger bundled IR libraries.
A few tips for getting the best results:
- Choose the right IR for the source. A cathedral IR on a lead vocal will create a wash of reverb that buries the vocal in the mix. Use shorter IRs (plates, chambers, rooms) for lead elements and save the long halls for orchestral elements or special effects.
- Use pre-delay. Most convolution plugins allow you to add pre-delay, which separates the dry signal from the reverb onset. This improves clarity without reducing the perceived reverb amount.
- EQ the reverb return. Sending the convolution output through an EQ lets you roll off low frequencies that muddy the mix or reduce high frequencies that sound harsh. This does not change the IR, it shapes how the reverb sits in the mix.
- Try multiple IRs. A common technique is to use a short room IR for intimacy and a longer hall IR for space, blended together. The room provides the close perspective, the hall provides the depth.
The Future of Convolution
Convolution reverb has been around since the 1990s, but the technology continues to evolve. Machine learning models are beginning to synthesize impulse responses from text descriptions. Real-time convolution with moving sources and listeners is becoming feasible. And as VR and spatial audio grow, the demand for accurate, position-dependent room responses is driving new capture techniques that go beyond stereo to full spherical impulse responses.
But the fundamental principle remains unchanged. The room is the instrument. Convolution is how we play it.
At Acousticas, we will keep capturing rooms and hardware, one impulse at a time.