What Is Bit Depth in Audio?

Every audio sample is a number representing the amplitude — the instantaneous level — of the sound at that moment. Bit depth is how many bits encode that number, which sets how many distinct amplitude values are available. 16-bit gives 65,536 possible values; 24-bit gives 16,777,216; and 32-bit float uses a floating-point scheme that spans an enormous range with fine precision throughout. The more values available, the smaller the gap between adjacent levels, so the digital representation lands closer to the true analog amplitude and the rounding error shrinks. Bit depth is the amplitude axis of digital audio, entirely separate from sample rate, which is the time axis (how often amplitude is measured). A useful mental model: sample rate is how many photographs per second you take of the waveform, and bit depth is how many brightness levels each photograph can record. Both together determine how faithfully the continuous analog wave is captured as discrete numbers.

Because bit depth offers only a finite set of amplitude values, each sample must be rounded to the nearest available level — a process called quantization. The small rounding error between the true amplitude and the stored value is quantization error, and across a whole signal it manifests as quantization noise, a low-level hiss-or-grunge that sets the noise floor of the recording. The more bits, the smaller each rounding step and the lower that noise. At 16-bit the steps are already tiny relative to full scale; at 24-bit they are 256 times finer still. Quantization noise is most audible on very quiet passages and fades, where the signal itself is near the size of the quantization steps — which is exactly where low bit depth reveals itself as grainy distortion rather than clean silence. Dithering (covered below) is the technique used to convert that harsh, signal-dependent quantization distortion into benign, steady background noise when bit depth must be reduced.

Bit depth translates directly into dynamic range — the span between the quietest sound a system can represent above its noise floor and the loudest before clipping. The rule of thumb is that each bit adds about 6 dB of dynamic range (more precisely 6.02 dB). So 16-bit provides roughly 96 dB of dynamic range, which comfortably covers the range of a finished piece of music and is why CDs use it. 24-bit provides about 144 dB in theory — more than human hearing's roughly 120 dB span and more than any real-world converter achieves (analog circuitry limits practical converters to around 120 dB), which is precisely the point: the extra range is headroom, not audible detail. 32-bit float offers a theoretical dynamic range so vast (over 1,500 dB) that clipping and noise floor essentially cease to be practical concerns during processing. The dynamic-range headroom of 24-bit — not audibly finer sound at normal levels — is the real reason professionals record at that depth.

The noise floor is the level of the quietest ever-present noise in a signal, below which real audio disappears into hiss. In digital audio the noise floor is set by quantization noise (from bit depth) plus any analog noise from microphones, preamps, and converters. Bit depth matters here because it determines how far below full scale you can place quiet detail before it is swallowed by quantization noise. With 16-bit's ~96 dB floor, a signal recorded conservatively — say peaking at -18 dBFS to leave headroom — still has plenty of clean range, but aggressive gain later can start to expose the floor. With 24-bit's ~144 dB theoretical floor, you can record at very safe, low levels and still keep the noise floor far below anything audible, which removes the pressure to push levels close to clipping during tracking. In practice the recording's real noise floor is usually set by the analog gear, not the bits, once you are at 24-bit — which is another reason 24-bit is the sensible recording standard and going beyond it rarely helps.

Record at 24-bit — essentially always. Every modern audio interface supports it, and the payoff is headroom: because 24-bit puts the noise floor far below the audible range, you can record with conservative levels (peaks well below 0 dBFS) and never worry about either clipping the loud parts or exposing quantization noise on the quiet parts. With 16-bit, recording too quietly pushes low-level detail toward the quantization floor and can add audible grain; 24-bit removes that tension entirely. For mixing, work at 32-bit float, which every major DAW uses internally by default. Floating point lets signal levels inside the mix exceed 0 dBFS momentarily without permanent damage, so a channel or bus that overshoots can be pulled back down later with no harm — a safety net that fixed-point 16- or 24-bit integer mixing does not provide. The workflow is therefore: capture at 24-bit for clean headroom, mix at 32-bit float for processing safety, and reduce bit depth only at the final export.

32-bit floating point behaves fundamentally differently from 16- and 24-bit integer formats, and understanding why explains its growing use. Integer formats map amplitude onto a fixed scale with a hard ceiling at 0 dBFS — go above it and the signal clips permanently. Floating point instead stores each sample as a mantissa and an exponent, like scientific notation, which lets it represent values far above 0 dBFS and maintain fine precision across a colossal range. The practical consequences are twofold. In a DAW, 32-bit float internal processing means intermediate signals can exceed 0 dBFS during heavy EQ or summing and be recovered intact by simply lowering the level afterward — clipping becomes non-destructive until the final fixed-point export. In field recording, newer 32-bit float recorders exploit the huge range to effectively remove the need to set input gain: even a wildly loud or quiet source is captured within range and normalized later. The catch is size and necessity — 32-bit float files are larger, and for delivery you still down-convert to 24- or 16-bit integer, since playback systems and lossy encoders expect fixed point.

Whenever you reduce bit depth — a 24-bit master down to 16-bit for CD or general distribution — dithering is essential. Truncating bits without dithering turns quantization error into signal-dependent distortion that is especially audible on quiet passages and fade-outs, where it sounds like gritty, granular breakup rather than clean decay. Dithering deliberately adds a tiny, carefully-shaped amount of random noise before truncation, which decorrelates the quantization error from the signal and converts that harsh distortion into a steady, benign, barely-perceptible noise floor. The audio stays clean and natural; you trade an almost inaudible rise in background noise for the elimination of ugly artifacts. Noise-shaped dither goes further, pushing that added noise into frequency ranges where the ear is least sensitive, making it effectively inaudible. Most mastering and export tools apply dither automatically when you export to 16-bit. Common algorithms include TPDF (triangular probability, the safe default), POW-r, and various noise-shaped options. Rule: dither only once, at the final bit-depth reduction, never repeatedly.

These three are constantly confused. Bit depth is how precisely each sample's amplitude is measured — it sets dynamic range and the noise floor, measured in bits (16-bit, 24-bit). Sample rate is how many samples are taken per second — it sets the frequency range, measured in kHz (44.1, 48, 96). Together, bit depth and sample rate define uncompressed PCM audio, and their product with channel count gives the raw data rate: 16-bit × 44,100 × 2 channels = 1,411 kbps for CD-quality WAV. Bitrate, by contrast, is a property of compressed audio — the kilobits per second a lossy encoder like MP3 or AAC produces (128 kbps, 320 kbps) — and it is the outcome of the encoder's decisions, not a direct measurement of the samples. In short: bit depth = how precisely each point is measured, sample rate = how often points are measured, bitrate = how much data the compressed file uses per second. Raising bit depth improves dynamic range; raising sample rate extends frequency range; raising bitrate reduces lossy compression artifacts.

A common misconception is that you should encode MP3 or AAC from a high bit-depth source to get a 'higher bit depth MP3.' Lossy formats do not have a user-facing bit depth in the PCM sense. Encoders like MP3, AAC, and Opus transform the audio into the frequency domain and represent it with their own internal precision governed by the bitrate, not by whether the source was 16- or 24-bit. When you export a 24-bit master to a 256 kbps AAC, the encoder is not preserving 24 bits of amplitude resolution; it is spending 256 kbps of data according to its psychoacoustic model. That said, encoding from a clean 24-bit master is still good practice, because you feed the encoder the highest-quality source before its one lossy step — but the resulting file's quality is set by the bitrate, not the source bit depth. This is why bit depth is a recording, mixing, and lossless-archival concern, and effectively irrelevant once you commit to a lossy delivery format. Choose bit depth for your masters; choose bitrate for your lossy exports.

Beyond the headline numbers, two nuances matter. First, integer versus float: 16- and 24-bit are integer formats used for delivery and most recording, while 32-bit (and 64-bit) float are used for internal mixing and, increasingly, gain-free field recording. Integer is the right choice for final files that playback systems and encoders expect; float is a processing and capture convenience. Second, how many bits actually help: 24-bit's advantage over 16-bit is real for recording and mixing because of headroom and noise-floor safety, but for finished, normalized listening material 16-bit already exceeds the effective dynamic range of most playback environments — which is why CD-quality 16-bit remains perfectly transparent for delivery. There is little practical point in 32-bit integer capture on typical interfaces, because the analog electronics themselves do not resolve anywhere near 32 bits of precision, so those bits capture noise rather than detail. The honest summary: 24-bit for capture and archival, 32-bit float for mixing, 16-bit for delivery — everything beyond that is headroom for machines, not audible resolution for ears.

Match bit depth to the stage of work. Recording: 24-bit, essentially without exception — it gives clean headroom and removes the pressure to chase levels. Internal mixing: 32-bit float, which modern DAWs use automatically and which makes in-the-box clipping non-destructive. Final master: 24-bit for high-resolution distribution and archival, dithered down to 16-bit for CD and general distribution where 16-bit is standard and transparent. Distribution as MP3, AAC, or Opus: bit depth is irrelevant — pick a bitrate instead, and let the encoder handle the rest. Archival: keep 24-bit masters (in WAV or FLAC) to preserve all the headroom and detail from the session. Avoid 32-bit integer capture on consumer and prosumer interfaces, whose electronics cannot achieve that precision. The simple rule to remember: capture and archive at 24-bit, mix at 32-bit float, deliver at 16-bit (or a lossy bitrate), and dither once whenever you reduce depth.