{"course":{"slug":"motion-and-audio","title":"Motion, Sound and Music","url":"https://addaly.com/learn/motion-and-audio","level":"intermediate","track":"create"},"lesson":{"slug":"gain-structure-and-headroom","title":"Where the level should sit","url":"https://addaly.com/learn/motion-and-audio/gain-structure-and-headroom","position":49,"of":72,"minutes":10,"answer":"Clipping replaces the waveform's peak with a flat line carrying harmonics that were never in the source and cannot be recovered.","keyIdea":"Clipping replaces the waveform's peak with a flat line carrying harmonics that were never in the source and cannot be recovered, while recording quietly only moves you closer to a noise floor that 24-bit has room to spare above, so the two ends of the meter are not symmetrical and headroom is cheap."},"body":{"format":"markdown","text":"## Where the level should sit, and why\n\nSet your recording so **peaks land between −12 and −6 dBFS**, with the average of speech somewhere around −18. That is the whole recommendation, and the reasons for both ends of it are worth knowing because they are not symmetrical.\n\n**The top end is a hard wall.** 0 dBFS is the largest number the format can represent. Exceed it and the waveform's peak is replaced by a flat line at the maximum value. That flat top is not a slightly-too-loud version of the sound; it is a different waveform, containing high-order harmonics that were never in the original. No processing recovers the shape, because the information is not attenuated: it is absent. Clipping is permanent.\n\n**The bottom end is soft.** Recording 12dB quieter than you could have does not destroy anything. It uses fewer of the available code values, and in 24-bit there are so many that it does not matter, for reasons covered in the next lesson. The cost of recording too quiet is that you are closer to the noise floor, and the noise floor of a decent preamp is a long way down.\n\nThe asymmetry is the point: **too loud is fatal, too quiet is an inconvenience.** Leave headroom.\n\n## Where to apply gain matters\n\nSignal chain: microphone → preamp → converter → file → software.\n\nGain applied at the **preamp** amplifies the microphone signal before the converter, so the signal arrives at the converter well above the converter's own noise. Gain applied **in software afterwards** multiplies the signal *and* everything the preamp and converter added to it.\n\nSo the rule is: get the level right at the preamp. A recording made at −40 dBFS and normalised to −12 afterwards has 28dB more of the preamp's hiss than one recorded at −12 in the first place. This is the single most common cause of a hissy recording made with adequate equipment.\n\nThe opposite error exists too: cranking the preamp to the point where it distorts, then turning it down digitally. The preamp's distortion is already in the signal.\n\n## What a meter is telling you\n\nTwo kinds of number, and people confuse them constantly.\n\n**Peak** meters show the highest sample value. This is what you set to avoid clipping.\n\n**RMS or loudness** meters show an average over time, which corresponds much more closely to how loud something sounds. Speech typically has 12 to 18dB between its average and its peaks (the crest factor) which is why speech peaking at −6 sounds much quieter than music peaking at −6.\n\nAnd a sample-peak meter can miss a peak. The analogue waveform reconstructed between samples can overshoot the highest sample, by a decibel or more. **True peak** meters model that reconstruction. It matters at the delivery stage, where a file peaking at −0.1 dBFS by sample measurement can clip a lossy encoder's decoder, and it is why the loudness lesson later specifies a true-peak ceiling of −1 dB or lower.\n\n## 32-bit float, and the caveat nobody mentions\n\nRecorders advertising 32-bit float claim you cannot clip. The claim is half true and the half that is false is the important one.\n\nIn a 32-bit float file, values above 1.0 are representable, so a signal that would have clipped in a fixed-point file is stored intact and can be pulled down afterwards. That part is real.\n\nBut the **analogue front end** (the microphone's own maximum sound pressure level, the preamp's supply voltage, the converter's input stage) still has a physical ceiling. Recorders achieve this with dual converters at different gains, combined in the file, which extends the range enormously and does not make it infinite. Overload the capsule or the input stage and the distortion is captured faithfully into a float file that cannot be repaired.\n\n32-bit float is an excellent insurance policy against a sudden shout. It is not a reason to stop setting gain.\n\n## Practical routine\n\n1. Ask the speaker to talk at their loudest realistic level (the laugh, the emphatic bit) not their test-voice.\n2. Set the preamp so that peaks land around −9 dBFS.\n3. Check the quiet passages are comfortably above the noise floor; if they are near −50, either get closer or accept a quieter recording and a noisier lift.\n4. Record ten seconds of silence for room tone before anybody speaks. It is free, it takes ten seconds, and every edit later will want it.\n5. Watch the meter through the first minute. People get louder once they relax.\n\n## Limiters on the way in\n\nA limiter at the input stops a peak from clipping by reducing gain fast when the level approaches a ceiling. Many field recorders include one.\n\nFor unrepeatable recordings (an interview, a live event) turn it on with a ceiling around −6 dBFS. The occasional couple of decibels of gain reduction on a laugh is far better than a clipped take. For controlled studio work, leave it off and set gain properly; a limiter is an insurance policy with a small premium, not a replacement for a level.\n\n## The free path\n\n**Audacity** shows peak and RMS on its meters and has a clipping indicator you should turn on. **Ardour** and **Reaper** (Reaper is paid but has an unlimited evaluation period and a very low licence) offer full metering. **ffmpeg** can tell you what a finished file actually did:\n\n```bash\nffmpeg -i take.wav -af \"astats=metadata=1:measure_overall=Peak_level+RMS_level\" -f null -\nffmpeg -i take.wav -af ebur128=peak=true -f null -\n```\n\nThe second prints integrated loudness and true peak, both free, both exactly what a delivery specification asks for.\n\n## The limitation\n\nMetering describes a file, not a performance. A recording at a perfect −9 dBFS peak of a voice 80cm from a microphone in a hard room is a well-levelled bad recording, and no meter says so. Level is the easiest thing to get right and the least important of the things that matter.\n\nThere is also no universal standard for what a meter shows. Different applications use different ballistics and different reference points, so two meters can legitimately disagree by several decibels on the same file. The numbers in this lesson are dBFS peak, which is unambiguous; anything described as \"VU\" or \"dB\" without a suffix is not."},"check":{"question":"A recordist raises the preamp until occasional peaks distort, then reduces the track's level in software by 6dB and reports that it sounds fine. What is wrong with this reasoning?","options":[{"text":"Reducing level in software applies to the whole track, so quiet passages are now 6dB closer to the file's noise floor than they need to be.","correct":false,"tempting":"A real consequence of digital attenuation, but minor compared with what has already happened to the peaks."},{"text":"The software reduction is applied after the fader's dither stage, so the 6dB cut reintroduces quantisation error in the reduced signal that the original recording did not have.","correct":false,"tempting":"Dither is applied at final bit-depth reduction, not at every fader, and modern hosts work in float where a 6dB cut is lossless."},{"text":"The distortion happened in the preamp before the converter, so it is part of the recorded signal and turning the whole thing down turns the distortion down with it.","correct":true},{"text":"A 6dB reduction in software is not equivalent to 6dB less preamp gain, because preamp gain is applied logarithmically to the microphone's voltage while digital attenuation scales sample values linearly, so the two operate on different curves.","correct":false,"tempting":"Both are multiplications by a constant factor; the decibel is a ratio in either domain, so this describes a difference that does not exist."}]},"prev":{"slug":"the-room-not-the-mic","title":"Direct sound, reverberant sound, and the distance between them","url":"https://addaly.com/learn/motion-and-audio/the-room-not-the-mic"},"next":{"slug":"sample-rate-and-bit-depth","title":"Two numbers doing two different jobs","url":"https://addaly.com/learn/motion-and-audio/sample-rate-and-bit-depth"},"authorship":"Founder and series editor: Dr Mustafa Mun. Written by AI models to a written standard. Exam keys checked blind by a second model: 3,665 of 3,666 agreed. Academic review invited.","review":{"status":"Draft, open for academic review"},"cite":"Addaly (2026) 'Where the level should sit', in Motion, Sound and Music. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/motion-and-audio/gain-structure-and-headroom (Accessed: 13 October 2026).","licence":{"name":"CC BY-NC-SA 4.0","url":"https://creativecommons.org/licenses/by-nc-sa/4.0/"},"reader":{"schema":1,"title":"Where the level should sit","courseTitle":"Motion, Sound and Music","module":{"index":6,"slug":"capturing-sound","title":"Capturing sound"},"answer":null,"keyIdea":"Clipping replaces the waveform’s peak with a flat line carrying harmonics that were never in the source and cannot be recovered, while recording quietly only moves you closer to a noise floor that 24-bit has room to spare above, so the two ends of the meter are not symmetrical and headroom is cheap.","hero":{"kind":"plate","image":null,"alt":null,"caption":null},"blocks":[{"t":"h2","id":"where-the-level-should-sit-and-why","runs":[{"text":"Where the level should sit, and why"}]},{"t":"p","runs":[{"text":"Set your recording so "},{"text":"peaks land between −12 and −6 dBFS","b":true},{"text":", with the average of speech somewhere around −18. That is the whole recommendation, and the reasons for both ends of it are worth knowing because they are not symmetrical."}]},{"t":"p","runs":[{"text":"The top end is a hard wall.","b":true},{"text":" 0 dBFS is the largest number the format can represent. Exceed it and the waveform’s peak is replaced by a flat line at the maximum value. That flat top is not a slightly-too-loud version of the sound; it is a different waveform, containing high-order harmonics that were never in the original. No processing recovers the shape, because the information is not attenuated: it is absent. Clipping is permanent."}]},{"t":"p","runs":[{"text":"The bottom end is soft.","b":true},{"text":" Recording 12dB quieter than you could have does not destroy anything. It uses fewer of the available code values, and in 24-bit there are so many that it does not matter, for reasons covered in the next lesson. The cost of recording too quiet is that you are closer to the noise floor, and the noise floor of a decent preamp is a long way down."}]},{"t":"p","runs":[{"text":"The asymmetry is the point: "},{"text":"too loud is fatal, too quiet is an inconvenience.","b":true},{"text":" Leave headroom."}]},{"t":"h2","id":"where-to-apply-gain-matters","runs":[{"text":"Where to apply gain matters"}]},{"t":"p","runs":[{"text":"Signal chain: microphone → preamp → converter → file → software."}]},{"t":"p","runs":[{"text":"Gain applied at the "},{"text":"preamp","b":true},{"text":" amplifies the microphone signal before the converter, so the signal arrives at the converter well above the converter’s own noise. Gain applied "},{"text":"in software afterwards","b":true},{"text":" multiplies the signal "},{"text":"and","i":true},{"text":" everything the preamp and converter added to it."}]},{"t":"p","runs":[{"text":"So the rule is: get the level right at the preamp. A recording made at −40 dBFS and normalised to −12 afterwards has 28dB more of the preamp’s hiss than one recorded at −12 in the first place. This is the single most common cause of a hissy recording made with adequate equipment."}]},{"t":"p","runs":[{"text":"The opposite error exists too: cranking the preamp to the point where it distorts, then turning it down digitally. The preamp’s distortion is already in the signal."}]},{"t":"h2","id":"what-a-meter-is-telling-you","runs":[{"text":"What a meter is telling you"}]},{"t":"p","runs":[{"text":"Two kinds of number, and people confuse them constantly."}]},{"t":"p","runs":[{"text":"Peak","b":true},{"text":" meters show the highest sample value. This is what you set to avoid clipping."}]},{"t":"p","runs":[{"text":"RMS or loudness","b":true},{"text":" meters show an average over time, which corresponds much more closely to how loud something sounds. Speech typically has 12 to 18dB between its average and its peaks (the crest factor) which is why speech peaking at −6 sounds much quieter than music peaking at −6."}]},{"t":"p","runs":[{"text":"And a sample-peak meter can miss a peak. The analogue waveform reconstructed between samples can overshoot the highest sample, by a decibel or more. "},{"text":"True peak","b":true},{"text":" meters model that reconstruction. It matters at the delivery stage, where a file peaking at −0.1 dBFS by sample measurement can clip a lossy encoder’s decoder, and it is why the loudness lesson later specifies a true-peak ceiling of −1 dB or lower."}]},{"t":"h2","id":"32-bit-float-and-the-caveat-nobody-mentions","runs":[{"text":"32-bit float, and the caveat nobody mentions"}]},{"t":"p","runs":[{"text":"Recorders advertising 32-bit float claim you cannot clip. The claim is half true and the half that is false is the important one."}]},{"t":"p","runs":[{"text":"In a 32-bit float file, values above 1.0 are representable, so a signal that would have clipped in a fixed-point file is stored intact and can be pulled down afterwards. That part is real."}]},{"t":"p","runs":[{"text":"But the "},{"text":"analogue front end","b":true},{"text":" (the microphone’s own maximum sound pressure level, the preamp’s supply voltage, the converter’s input stage) still has a physical ceiling. Recorders achieve this with dual converters at different gains, combined in the file, which extends the range enormously and does not make it infinite. Overload the capsule or the input stage and the distortion is captured faithfully into a float file that cannot be repaired."}]},{"t":"p","runs":[{"text":"32-bit float is an excellent insurance policy against a sudden shout. It is not a reason to stop setting gain."}]},{"t":"h2","id":"practical-routine","runs":[{"text":"Practical routine"}]},{"t":"ol","start":1,"items":[[{"text":"Ask the speaker to talk at their loudest realistic level (the laugh, the emphatic bit) not their test-voice."}],[{"text":"Set the preamp so that peaks land around −9 dBFS."}],[{"text":"Check the quiet passages are comfortably above the noise floor; if they are near −50, either get closer or accept a quieter recording and a noisier lift."}],[{"text":"Record ten seconds of silence for room tone before anybody speaks. It is free, it takes ten seconds, and every edit later will want it."}],[{"text":"Watch the meter through the first minute. People get louder once they relax."}]]},{"t":"h2","id":"limiters-on-the-way-in","runs":[{"text":"Limiters on the way in"}]},{"t":"p","runs":[{"text":"A limiter at the input stops a peak from clipping by reducing gain fast when the level approaches a ceiling. Many field recorders include one."}]},{"t":"p","runs":[{"text":"For unrepeatable recordings (an interview, a live event) turn it on with a ceiling around −6 dBFS. The occasional couple of decibels of gain reduction on a laugh is far better than a clipped take. For controlled studio work, leave it off and set gain properly; a limiter is an insurance policy with a small premium, not a replacement for a level."}]},{"t":"h2","id":"the-free-path","runs":[{"text":"The free path"}],"role":"free"},{"t":"p","runs":[{"text":"Audacity","b":true},{"text":" shows peak and RMS on its meters and has a clipping indicator you should turn on. "},{"text":"Ardour","b":true},{"text":" and "},{"text":"Reaper","b":true},{"text":" (Reaper is paid but has an unlimited evaluation period and a very low licence) offer full metering. "},{"text":"ffmpeg","b":true},{"text":" can tell you what a finished file actually did:"}]},{"t":"code","lang":"bash","label":"Bash","text":"ffmpeg -i take.wav -af \"astats=metadata=1:measure_overall=Peak_level+RMS_level\" -f null -\nffmpeg -i take.wav -af ebur128=peak=true -f null -"},{"t":"p","runs":[{"text":"The second prints integrated loudness and true peak, both free, both exactly what a delivery specification asks for."}]},{"t":"h2","id":"the-limitation","runs":[{"text":"The limitation"}],"role":"limit"},{"t":"p","runs":[{"text":"Metering describes a file, not a performance. A recording at a perfect −9 dBFS peak of a voice 80cm from a microphone in a hard room is a well-levelled bad recording, and no meter says so. Level is the easiest thing to get right and the least important of the things that matter."}]},{"t":"p","runs":[{"text":"There is also no universal standard for what a meter shows. Different applications use different ballistics and different reference points, so two meters can legitimately disagree by several decibels on the same file. The numbers in this lesson are dBFS peak, which is unambiguous; anything described as “VU” or “dB” without a suffix is not."}]}],"check":{"question":[{"text":"A recordist raises the preamp until occasional peaks distort, then reduces the track’s level in software by 6dB and reports that it sounds fine. What is wrong with this reasoning?"}],"options":[{"text":[{"text":"Reducing level in software applies to the whole track, so quiet passages are now 6dB closer to the file’s noise floor than they need to be."}],"correct":false,"tempting":[{"text":"A real consequence of digital attenuation, but minor compared with what has already happened to the peaks."}]},{"text":[{"text":"The software reduction is applied after the fader’s dither stage, so the 6dB cut reintroduces quantisation error in the reduced signal that the original recording did not have."}],"correct":false,"tempting":[{"text":"Dither is applied at final bit-depth reduction, not at every fader, and modern hosts work in float where a 6dB cut is lossless."}]},{"text":[{"text":"The distortion happened in the preamp before the converter, so it is part of the recorded signal and turning the whole thing down turns the distortion down with it."}],"correct":true,"tempting":null},{"text":[{"text":"A 6dB reduction in software is not equivalent to 6dB less preamp gain, because preamp gain is applied logarithmically to the microphone’s voltage while digital attenuation scales sample values linearly, so the two operate on different curves."}],"correct":false,"tempting":[{"text":"Both are multiplications by a constant factor; the decibel is a ratio in either domain, so this describes a difference that does not exist."}]}],"order":[0,3,2,1],"why":null},"prev":{"slug":"the-room-not-the-mic","title":"Direct sound, reverberant sound, and the distance between them","minutes":11},"next":{"slug":"sample-rate-and-bit-depth","title":"Two numbers doing two different jobs","minutes":10}}}