SoundForgePro Official 15-day trial

Audio Fundamentals

16-bit vs 24-bit Audio: Which to Record, Mix and Deliver

Guide navigation Table of contents Sections
    Brass large-diaphragm condenser mic in a spider shock mount beside an open box of brass calibration weights and two memory cards on cream linen over a travertine table
    Record at 24-bit, then pick the delivery format at the end.

    Record at 24-bit, keep the native 24-bit master for delivery, and make a 16-bit copy when you need a CD or listening file. The extra bits lower quantization noise, which lets you leave safer recording levels without making the file format the main noise limit. For playback, the master matters more than paying for extra bits alone; the listening studies below need different interpretations.

    I'd record new takes at 24-bit and keep a 24-bit master. The extra 8 bits cost 50% more uncompressed PCM storage at the same sample rate and channel count, and the quantization noise stays well below that of typical recording hardware. The DAW's internal floating-point processing is a separate choice from the recorded file format. For a listening copy, I'd choose the master I prefer before paying for extra bits alone. The two bit-depth-only surveys below found no consistent preference for 24-bit; they don't prove that the files were indistinguishable. Keep the native 24-bit master for delivery and make the 16-bit copy when you need it.

    Quick answer: which bit depth for each job

    Use 24-bit for recorded files and the native master. Let the DAW process in floating point, and use 32-bit float for intermediate renders when the next application supports it. Reduce to 16-bit only for a destination that needs it.

    JobBit depthWhy
    Recording24-bitLeaves more room below a cautious recording level. The Scarlett input measurement below is 111.6 dB; the output measurements don't establish capture performance.
    Editing, mixing, mastering24-bit source files; floating-point processingKeep intermediate renders at 24-bit or 32-bit float. A 16-bit source does not force the DAW to calculate at 16-bit.
    CD master16-bit, 44.1 kHzThat's the CD audio standard. Make it from the 24-bit master, with dither as the last step.
    Spotify; Apple Digital MastersNative 24-bit master when availableSpotify accepts a native 16-bit master when no higher-depth master exists. The Apple Digital Masters program requires native 24-bit sources.
    Listening copies16-bitA practical playback copy when the master and levels are suitable. The two surveys below found no consistent preference for 24-bit.
    Archive24-bit, losslessKeep the master untouched. Padding a 16-bit copy back to 24-bit later can't recover anything.

    16 vs 24 bit audio: what the extra bits change

    Bit depth decides how far below the signal the noise floor sits. A 16-bit sample can take one of 65,536 values and a 24-bit sample one of 16,777,216, as the Wikipedia overview of audio bit depth lists them. Each extra bit doubles the number of values and pushes the rounding error, called quantization noise, about 6 dB further down.

    In properly dithered audio that rounding error comes out of the converter as a low, steady hiss. The stair-step drawing in many explainers suggests that a 24-bit wave is smoother or carries finer detail, and that picture misleads. iZotope's primer on sample rate and bit depth makes the same correction: more bits lower the noise floor and widen the dynamic range, and the waveform you hear above that floor stays the same.

    Levels in a file are counted in dBFS, decibels relative to digital full scale. For fixed-point PCM samples, 0 dBFS marks full scale and quieter values are negative. Floating-point files and processing can carry values above 0 dBFS; the analog output still needs a level that avoids overload. Inside the session most DAWs process in 32-bit or 64-bit floating point while interfaces convert at 24-bit, so the bit depth you pick matters at the file boundaries: the recording, the stems and the export. Bit rate, the number of bits a file uses per second, is a separate setting from bit depth.

    Ideal quantization numbers: 98 dB vs 146 dB

    The familiar 98 and 146 dB figures describe an ideal full-scale sine wave's signal-to-quantization-noise ratio, or SQNR. The formula is approximately 6.02N + 1.76 dB, where N is the bit depth; the audio bit-depth explanation gives 98.09 dB for 16-bit and 146.26 dB for 24-bit. This model treats the quantization error as uniformly distributed over one quantization step. It is not a measurement of a dithered recording, an interface or the contrast between a song's loud and quiet passages.

    You'll also see the roughly 6 dB-per-bit shorthand: 96 dB and 144 dB. iZotope's bit-depth explainer calculates -96.3 and -144.5 dBFS from the maximum half-step rounding error relative to peak full scale. That error bound is not an integrated RMS noise measurement. The sine-wave SQNR calculation uses the sine's RMS level and the RMS quantization error; dither and noise shaping change the noise result further.

    Quantity and condition16-bit24-bit
    Values per sample65,53616,777,216
    Ideal full-scale-sine SQNR, uniform quantization-error model98.09 dB146.26 dB
    Maximum half-step rounding error relative to peak full scale-96.3 dBFS-144.5 dBFS

    Those ideal numbers need context. Xiph.Org's explanation of 24/192 music downloads describes how noise-shaped dither moves noise toward frequencies where hearing is less sensitive, giving 16-bit audio a perceived dynamic range of about 120 dB. That is not a flat, unweighted broadband SQNR figure. Dither also permits signals below one quantization step to remain in the reconstructed output, so the step size is not a hard cutoff for quiet musical detail. The real converter figures below describe hardware under their own measurement conditions.

    16 vs 24 bit sound in real converters: test results

    The measured inputs and outputs answer different questions. An ADC input result is relevant to recording; a DAC output result describes playback. Keep dynamic range and signal-to-noise ratio separate too. The figures below come from three Audio Science Review interface reviews, with one sample unit per model, so they aren't four independent tests or a pooled capture benchmark.

    Interface and measurement pointMetric and conditionMeasured (dB)Maker spec (dB)
    MOTU M2, balanced outputAES17 dynamic range, 4.1 V116.3120
    Scarlett 2i2 4th gen, line outputAES17 dynamic range, volume = 0 dBFS, 4 V116.0120
    RME Babyface Pro FS, outputSignal-to-noise ratio, 6.5 V just before clipping114.5115
    Scarlett 2i2 4th gen, A/D inputAES17 dynamic range; measurement weighting not stated111.6115, A-weighted

    Each published figure here averages the left and right channels of that measurement. The Scarlett input is the only A/D result in this set. Its 111.6 dB is about 13.6 dB above the ideal 16-bit full-scale-sine signal-to-quantization-noise figure of 98.09 dB, but that calculation isn't a measured improvement from switching recording formats. The input conditions, dither, gain and source noise still matter. The three output rows can't extend that figure to recording on the other interfaces.

    Maker specs sit a little higher. The five quoted output or D/A specifications span 115 to 129 dB, under their makers' respective conditions; this is a list of specifications, not a normalized benchmark. Focusrite lists 120 dB for the Scarlett 2i2 line outputs, MOTU states a measured 120 dB on the M2 outputs, Universal Audio lists 129 dB for the Apollo Twin X Gen 2 D/A converters, Audient lists 126 dB for the iD14 D/A converter, and RME's 115 dB figure is the one quoted in the Babyface review. A fair measured-versus-specified comparison needs the same bandwidth, weighting, level and measurement point.

    The converter chips set the ceiling. ESS lists its Sabre ADC chips at 116 to 128 dB per channel, a median of 122 dB across nine models, and its Sabre audiophile DAC chips at 112 to 132 dB, a median of 124 dB across 22 current and legacy models. The highest per-channel DNR in that cited DAC set is 132 dB; combined-channel modes use different conditions. A chip figure describes the converter alone, before the preamps, power supply and output stages of a finished product.

    Read the measurement labels before comparing numbers. The RME SNR chart stays separate from the AES17 dynamic-range results, and the weighting isn't always stated. Specs and chip figures are vendor ceilings. A-weighted figures typically read about 3 to 6 dB better than unweighted ones, by the rule of thumb in NwAvGuy's explainer on noise and dynamic range, so the Scarlett's A-weighted input spec and its measured input figure aren't a like-for-like pair. None of these specifications is an audibility result, and output specifications don't establish the noise floor of a recording input.

    For recording, I'd still choose 24-bit: it leaves room for conservative levels without making the file's quantization noise the first constraint. The Scarlett A/D result supports that choice for that measured input. It doesn't establish a 14–18 dB recording improvement across these interfaces, and it says nothing about whether a finished song sounds better in a 24-bit playback file. The converter that sets this floor lives in your interface, and audio interface vs mixer explains that side of the hardware.

    24 bit audio vs 16 bit: can you hear the difference?

    These studies answer different questions. The two Archimago surveys found no consistent preference for 24-bit, but a balanced preference result doesn't establish that listeners couldn't distinguish the files. Meyer and Moran measured discrimination through a CD-resolution loop, while the Reiss meta-analysis combined high-resolution conditions that don't isolate bit depth.

    TestWhat changedSizeResultLimit
    Meyer and Moran, JAES 2007A 16-bit, 44.1 kHz A/D/A loop inserted into SACD and DVD-Audio playback554 trials49.8% correctChanges sample rate and bit depth together; the loop wasn't dithered
    Archimago, 201424-bit files vs the same music dithered to 16-bit140 respondentsExact 50/50 on two of three tracks; a non-significant lean toward 16-bit on the thirdUncontrolled survey of the file judged better or 24-bit; 29 used ABX tools, without one common repeated-trial protocol
    Archimago, 202324-bit vs dithered 16-bit, both at 44.1 kHz121 responses38 preferred 24-bit, 40 preferred 16-bit, 43 no preferenceUncontrolled online preference test
    Reiss meta-analysis, JAES 2016High-resolution audio vs 16-bit at 44.1 or 48 kHz, across 18 studies400+ participants, 12,500+ trialsSmall but significant; about 60% correct with trainingHigh resolution bundles sample rate and bit depth

    The Meyer and Moran trial count comes from Xiph.Org's summary of the study: across 554 trials by engineers, recording students and audiophiles on several systems, listeners chose correctly 49.8% of the time. The authors' follow-up on when the 16-bit noise became audible shows where it did. With playback turned 20 dB above their normal setting on a near-silent stretch of room tone, they scored 9 of 9 and 6 of 6, and they still got 15 of 15 at only 10 dB above normal, where the disc's loudest peaks would have reached 111 dB SPL. Their normal setting already played pink noise at -16 dBFS at 85 dB SPL at the chair, and they add that most recordings carry their own noise, which swamps the noise of the CD loop.

    The Archimago tests come closest to the question on this page, because the sample rate stayed fixed. In the 140-person blind test of 24-bit vs dithered 16-bit from 2014, respondents compared 24-bit files from 24/96 sources with the same music dithered to 16-bit, and the 44 respondents with systems costing $6,000 or more showed no stronger group selection of the 24-bit versions. The 2023 repeat with 121 responses used Daft Punk's Random Access Memories, made the 16-bit version with iZotope RX 10 dither, and found that even the 38 listeners who preferred the 24-bit file rated the change in noise level at only 0.47 out of 10 on average. Both were self-administered on respondents' own equipment. Preference, identification and a controlled repeated-trial discrimination score are different outcomes; don't turn the first into the last.

    Joshua Reiss's meta-analysis of 18 studies, published in the Journal of the Audio Engineering Society in 2016, is the strongest evidence on the other side. It found a small but statistically significant ability to tell high-resolution audio from 16-bit audio at 44.1 or 48 kHz, rising to about 60% correct with extensive training. High resolution in those studies means a higher sample rate, a higher bit depth or both, so the result can't be pinned on the extra bits.

    The surveys give no consistent preference for the extra bits, while the other studies leave bit depth entangled with sample rate. That supports a practical choice, not a guarantee that no listener can hear a difference. I'd spend effort on the master and playback chain before paying for extra bits alone. Very quiet passages at unusually high gain can expose a 16-bit noise floor, as the Meyer and Moran follow-up shows.

    Why record and mix at 24-bit anyway

    Record at 24-bit because it forgives a cautious level. Xiph.Org's explainer names headroom, noise floor and convenience as the reasons professionals record and produce at 24-bit, and the main one is level-setting: at 24-bit you can set an approximate level and guess low without paying for it in noise. Apple's mastering guidance adds that 24-bit recording has become the norm for the same headroom and dynamic range.

    The arithmetic is easy to remember. Split 24-bit's extra 48 dB down the middle and you can leave 24 dB more headroom than at 16-bit while the noise floor still sits 24 dB further below the signal. The measured Scarlett A/D input illustrates the useful margin below the input's own noise; the D/A measurements describe a separate playback path.

    What 24-bit doesn't add is analog headroom. The microphone, preamp and converter clip at the same input level whatever the file format, so a peak that overloads the preamp at 16-bit overloads it at 24-bit too. Set the gain below overload first, keeping the loudest moments comfortably under full scale; recording in Sound Forge Pro walks through the File > New format and the level settings.

    Intermediate files are the second reason to keep more than 16 bits. A modern DAW can process a 16-bit source in floating point; it does not accumulate 16-bit rounding error at every plugin simply because the source file is 16-bit. Repeatedly rendering the result back to 16-bit creates new quantization boundaries. iZotope's primer recommends 32-bit float for intermediate exports, while 24-bit remains a practical exchange format. In NBS-Studio's account of quiet 16-bit stems, boosting the supplied files made them unusable, and the artist returned to the mix and exported again at 24-bit. That is one engineer's example, not a measured failure rate; I'd ask for 24-bit stems or an agreed floating-point format before mixing.

    Audio 24 bit vs 16 bit for delivery: CD, Spotify, Apple

    Keep and deliver the native master. Spotify asks for 24-bit when that is the native source, while Apple Digital Masters specifically calls for native 24-bit sources. The latter is a program requirement, not proof that every Apple Music delivery must be 24-bit. Check your distributor's delivery specification before exporting.

    DestinationSendRule
    CD16-bit, 44.1 kHzThe CD audio standard: each sample is a value between -32,768 and 32,767.
    SpotifyThe native master: 24-bit if that's what you have, 44.1 kHz or higherDon't make a 16-bit version of a 24-bit master and don't deliver both.
    Apple Digital Masters program24-bit source at the highest resolution availableNo 16-bit files padded to 24-bit; Apple's encoders need no added dither.
    Listening copies16-bitA practical playback copy; the two surveys found no consistent 24-bit preference.
    Archive24-bit, losslessThe same master you'd send to streaming.

    Spotify's audio file requirements ask for the highest-quality native master: 24-bit if that's the native master, 16-bit only when no higher bit depth exists. They ask you not to create a 16-bit version from a 24-bit master or deliver both, and to send 44.1 kHz or higher. Spotify handles its own conversions: anything above 24-bit is reduced to 44.1 kHz/24-bit FLAC at most, and files below 16-bit or 44.1 kHz are converted up to 44.1 kHz/16-bit and lose lossless eligibility.

    For that program, Apple's Apple Digital Masters guidelines ask for 24-bit sources only, at the highest resolution available, and rule out "bit-padding" a 16-bit file to 24-bit. The same document says Apple's encoders use the full resolution of a 24-bit source and need no added dither.

    Delivering the 24-bit master doesn't mean every listener hears 24-bit. Spotify's lossless stream, up to 24-bit/44.1 kHz, is a Premium feature that listeners pick under Media quality, as its lossless listening help page explains, and Apple Music subscribers switch Lossless Audio on under Audio Quality in the Music settings, according to Apple's lossless audio support page.

    The CD is the one destination where 16-bit is fixed: 44.1 kHz and 16 bits per sample. iZotope's rule of thumb in its bit-depth primer lines up with all three: 16-bit at 44.1 kHz for CD, 24-bit at the session's sample rate for just about everything else, and a check of what your aggregator accepts. Distributors run their own upload forms, so check what yours asks for before you export; I wouldn't assume one distributor's rule holds for the next.

    Archive the 24-bit master in a lossless format, and the comparison of WAV vs FLAC vs MP3 for editing, mastering and archive covers which one to keep. Bit depth is also only half of a streaming delivery; the streaming loudness standards cover the level side.

    16 bit audio vs 24 bit: how much bigger is the file?

    At the same sample rate, a 24-bit file is exactly 50% larger than a 16-bit file, because each sample carries 24 bits instead of 16. The figures below are uncompressed stereo PCM in decimal megabytes, without the file header.

    FormatData ratePer minutePer hour
    44.1 kHz, 16-bit1,411.2 kbps10.58 MB635 MB
    44.1 kHz, 24-bit2,116.8 kbps15.88 MB953 MB
    48 kHz, 16-bit1,536 kbps11.52 MB691 MB
    48 kHz, 24-bit2,304 kbps17.28 MB1,037 MB

    An hour of 24-bit stereo at 44.1 kHz stays under 1 GB, a small price for a take you can't repeat. For other lengths, channel counts or compressed formats, the audio file size calculator works it out.

    Going from 24-bit to 16-bit: dither, once, at the end

    When a 24-bit master has to become a 16-bit CD file, add dither as the very last step. Dither trades quantization distortion for a little noise: plain truncation can leave quantization distortion in quiet fades and tails, and dither replaces it with a low, steady noise, as Apple's mastering document describes. At 16-bit that added noise sits about 84 dB below full scale on ITU-R 468 weighting, according to the Wikipedia bit-depth article, which puts it in the range of microphone and room noise.

    The order is simple. If the CD copy also needs a different rate, change the sample rate first, then reduce bit depth with dither in one final stage; that guide covers the Sound Forge settings and the dither shapes.

    Dither matters most on the way down to 16-bit. Whether a heavily processed master written out at 24-bit needs a dither stage of its own is a separate, smaller decision, and the dither guide covers that case too. Before the 16-bit file goes out, run the audio export QA checklist to confirm it's the file you meant to send.

    FAQ

    Is 24-bit audio better than 16-bit?

    For recording and production, yes: it keeps the file's noise floor far below the converter's. For playback, a properly made 16-bit copy is a practical choice. The two surveys found no consistent preference for 24-bit, which isn't proof of indistinguishability; very quiet material at high gain is a separate case.

    Does Spotify accept 24-bit audio?

    Yes. Spotify asks for the native master, which means 24-bit if that's what you have, and asks you not to make a 16-bit copy of it or send both versions.

    Does converting 16-bit to 24-bit improve quality?

    No. A padded file holds exactly the information of the 16-bit original in a bigger container, which is why Apple rules out bit-padded files for Apple Digital Masters. Working on 16-bit material inside a 24-bit session is fine for processing, but it won't bring back precision the source never had.

    Is a 24-bit download worth buying?

    I'd pay for a better master before paying for extra bits alone. These surveys didn't establish a consistent 24-bit preference, and a high-resolution release can differ from a CD because it uses a different master. Choose the master you prefer, and keep its native resolution if storage isn't a concern.

    Can a final master be 16-bit?

    Yes, for a CD and for any outlet that asks for 16-bit. iZotope calls 16-bit viable for a final master unless you need more than 96 dB of effective dynamic range. For Spotify, send the native 24-bit master when you have one. Apple Digital Masters requires native 24-bit sources; other deliveries follow the distributor's specification.

    Verdict: 24-bit to work, 16-bit to listen

    My verdict: record at 24-bit, process in floating point, keep the native master, and make a dithered 16-bit copy when the destination needs one. The measurements explain the first half. The Scarlett input measurement and the headroom argument support 24-bit capture; output measurements don't quantify that capture benefit. For playback, the two bit-depth-only surveys found no consistent preference for 24-bit. They don't prove that every listener would be unable to distinguish the files. For Spotify, send the native 24-bit master when available; Apple Digital Masters specifically requires native 24-bit sources. Mastering a track in Sound Forge Pro ends with that 24-bit master, and the audio fundamentals guides pick up the next format decisions from here.