Reply to thread

Message: [QUOTE="Queen, post: 90470, member: 27"] An impulse response is not a recording of a room, but a measurement of how that space reacts to a single instant of sound. Get the measurement wrong, and every reverb built from that file carries the error. Nothing downstream fixes a bad capture. The blunt method is to make a loud, short noise and record the decay. People use balloons, starter pistols, and hand claps. ISO 3382-2 actually allows it, so the approach is not fringe. It is still the weakest option on the list. Balloons and clappers do not put enough energy below 125 Hz to give a trustworthy low-frequency decay, and no two pops come out identical. [HEADING=2]A rising sine sweep beats a balloon pop every time[/HEADING] The standard method now is an exponential sine sweep. You play a tone that climbs slowly through the audible range and record what the space does with it. Because you set the length, you set how much energy goes in. That energy is the whole point. A sweep gives you 80 to 100 dB of usable decay range, against roughly 30 to 50 dB from a balloon. Play the same digital file twice, and the excitation is identical, so repeat measurements actually agree. Length is the cheap fix for a noisy room. A longer sweep pushes more total energy into the space without making it louder, which is how people measure churches and stairwells without annoying everyone in the building. [B]Why balloon pops make weak impulse responses[/B] comes down to physics rather than technique. A bursting balloon radiates unevenly and never repeats, while ISO 3382-1 asks for an omnidirectional source holding within ±1 dB across directions. [HEADING=2]Deconvolution pulls the room out of the sweep[/HEADING] What the microphone captures is not an impulse response yet. It is your sweep with the room smeared through it, so the sweep has to come back out. The maths is a division. Both the recording and the original sweep move into the frequency domain, and then the recording is divided by the excitation, which leaves the transfer function of the space. Plain division misbehaves outside the sweep's bandwidth, where there is barely any signal, and dividing tiny numbers blows up the noise. Measurement software uses regularised inversion, holding back the correction outside the band you actually measured. Skip it, and hiss ends up sitting in the tail forever. Exponential sweeps shrug off loudspeaker distortion better than the older maximum length sequence method. They also handle a room that shifts slightly mid-measurement, which matters when air conditioning or a passing van refuses to cooperate. [HEADING=2]Convolution assumes the room never changes[/HEADING] Every convolution reverb rests on one assumption. The space responds the same way at any level and at any moment, which acousticians call linear and time-invariant behaviour. Air in a room mostly obliges. Spring tanks and plate reverbs do not, because their response changes with level, and one static file cannot carry that. It is also why a captured tail sits so still. Nothing inside it drifts, so a long decay can feel frozen against an arrangement that keeps moving. Producers chasing movement reach for [B][URL='https://goldmidi.com/community/threads/gooey-audio-brought-albatrosss-warped-vintage-effects.75874/']plugins that dirty a captured space on purpose[/URL][/B], layering wobble and grit over a fixed tail. If you capture your own, write everything down. Source position, microphone position, height, and gain all change the result, and ISO 3382-1 sets the receiver at 1.2 m because that is roughly where a seated listener's ears sit. Storage matters once you move past plain stereo. Spatial and binaural room impulse responses have a standard format of their own under AES69, published in 2020 and revised in 2022, which keeps the geometry attached to the audio instead of stranded in a filename. [/QUOTE]

Name