If the sample rate sets how many times per second we measure the signal, bit depth sets how precisely we record each measurement. It is an often misunderstood parameter: it is not about "quality" in a generic sense, but about something very specific, namely the distance between the loudest representable sound and the background noise introduced by conversion.
What the bits represent
Each sample is an integer. With 16 bits you can represent 65,536 different levels; with 24 bits more than 16.7 million. The value measured by the converter is rounded to the nearest level: this rounding is called quantization and introduces a small error.
In most cases that error behaves like a very faint noise added to the signal: quantization noise. The more bits you use, the denser the levels, the smaller the error and the lower that noise.
Theoretical dynamic range
There is an easy rule of thumb: each added bit is worth about 6 dB of dynamic range. The more precise formula for a sine wave signal is 6.02 × N + 1.76 dB, where N is the number of bits.
| Bit depth | Levels | Theoretical dynamic range |
|---|---|---|
| 8-bit | 256 | about 50 dB |
| 16-bit | 65,536 | about 98 dB |
| 24-bit | 16,777,216 | about 146 dB |
For comparison, the difference between a quiet room and an uncomfortably loud sound is well below the theoretical 146 dB of the 24-bit format. In practice no real converter reaches that figure: the limit is the noise of the analog electronics (preamp, microphone, room), which kicks in much earlier.
Why record at 24-bit
If 16 bits are enough for the final product, why does everyone recommend recording at 24-bit? The answer is the safety margin (headroom).
When you record you don't know the exact peaks in advance. With 16 bits, to use the full dynamic range you would have to record close to digital zero, risking clipping. With 24 bits you can keep peaks at -12 or -18 dBFS without worry: quantization noise still stays far below the preamp's own noise.
A concrete example: a voice recorded with peaks at -18 dBFS in 16-bit sits about 80 dB above quantization noise, still acceptable but with no margin left. At 24-bit, in the same situation, quantization noise is completely irrelevant.
The 32-bit float format
Many DAWs work internally in floating point at 32 or 64 bits, and some portable recorders and interfaces record directly in 32-bit float. It works differently from integers:
- Each sample is made of a mantissa (24 bits of effective precision) and an exponent that shifts the scale.
- The result is a huge representable range that extends far above 0 dBFS.
- A signal that goes over digital zero inside the DAW is not clipped: simply lower the level later in the chain to recover it intact.
Watch the converter
A 32-bit float file does not rescue a signal that is clipped at the input if the analog-to-digital converter itself is overloaded. Only devices designed for this, using several converters in parallel at different gains, offer real protection against clipping while recording. On a traditional interface, recording in 32-bit float gives you a bigger file but does not protect you from a preamp set too high.
When to use what
- Recording: 24-bit integer is the safe standard. 32-bit float is useful with recorders that genuinely support it, for example for unpredictable live events.
- Editing and mixing: let the DAW work in float; export intermediate bounces in 32-bit float so you lose no information.
- Distribution: 16-bit is enough for listening; 24-bit is required by some music distributors. For podcasts the final file will be compressed anyway.
Common mistakes
- Thinking more bits mean more volume: bit depth does not change level, it changes the distance from quantization noise.
- Reducing from 24 to 16 bits without dither: plain truncation produces distortion in quiet passages such as reverb tails. You need dithering.
- Recording too low "to be safe": even at 24-bit, peaks at -40 dBFS bring the signal close to the preamp noise. An average around -18 dBFS is a good reference.
- Confusing bit depth and bitrate: bitrate is the amount of data per second and also depends on sample rate and channels, as well as on compression.
A simple check
In my work I always run a test before an important session: I record ten seconds of silence with the microphone connected and the gain set. Looking at the meter immediately shows the real background noise level, which is almost always dominated by the room or the preamp, not by the bits. It is the quickest way to see that bit depth, beyond 24 bits, is rarely the bottleneck.
In short
- Bit depth defines the precision of each sample and therefore the level of quantization noise.
- Each bit is worth about 6 dB: 16-bit gives about 98 dB, 24-bit about 146 dB in theory.
- Recording at 24-bit lets you leave plenty of headroom without increasing audible noise.
- 32-bit float protects against clipping inside the DAW, but not always at the converter input.
- Going down to 16-bit in the final file requires dithering.
