Bit depth
Bit depth determines the number of levels used to describe amplitude and therefore the quantization noise:
| Format | Theoretical dynamic range | Use |
|---|---|---|
| 16 bit | about 96 dB | CD, distribution |
| 24 bit | about 144 dB (in practice limited by the electronics, about 120 dB) | Recording, stems, delivery masters |
| 32 bit float | enormous range, no internal clipping | DAW internal processing, intermediate exports |
In floating point (32 bit float) a signal above 0 dBFS does not clip: it can be brought back down by lowering the level. In fixed point (16 and 24 bit) everything above 0 dBFS is clipped.
What dithering is
When you reduce the bit depth (from 32 float or 24 bit to 16 bit), the values are rounded. The rounding error is correlated with the signal and, on very quiet passages (reverb tails, fade-outs, classical music at pianissimo), it is heard as grainy distortion. Dithering adds very low-level white or shaped noise before rounding: the error becomes constant, uncorrelated noise, which is much less annoying, and details below the least significant bit remain audible.
Noise shaping
Noise shaping moves the dither noise energy toward the frequencies where the ear is less sensitive (above 15 kHz), reducing it in the most sensitive region (2–5 kHz). The total noise is greater, the perceived noise is lower. Common types: plain TPDF (no shaping, neutral), POW-r 1/2/3, iZotope MBIT+ (with different shaping levels), FabFilter Pro-L (shaping options), Waves IDR.
Practical rules
- Dither only once, as the very last step, and only when reducing bit depth.
- Export at 24 bit: TPDF dither is recommended, but the noise stays at about −144 dBFS, which is irrelevant. Export at 32 bit float: no dither.
- Export at 16 bit (CD, some distributors): dither is mandatory; moderate noise shaping for loud pop/rock, TPDF or light shaping for dynamic music.
- After the dither, no processor must alter the signal (no limiter, no master fader, no normalization).
- Files intended for further work (stems, mixes for mastering): 24 bit or 32 float, no dither with noise shaping.
Sample rate
- 44.1 kHz: CD and most music distribution.
- 48 kHz: video, broadcast, Dolby Atmos, many modern projects.
- 88.2 / 96 kHz: high-resolution recording and mastering, “Hi-Res” distribution.
- Conversion (SRC) must be done with a quality converter (iZotope RX Resample, SoX, the limiter's or the DAW's converter in maximum quality mode) and before the final limiter, because conversion can create new (inter-sample) peaks.
- Deliver the master at the project's sample rate when possible; the distributor converts for lower formats.
In FL Studio
- Export window: WAV format, depth 16, 24 or 32 bit float; the Dither option is active only for 16 bit export (enable it if the final limiter does not already apply dither, never twice).
- Resampling quality: set the highest value on export.
- The final limiter (Maximus or Pro-L) can apply dither: in that case, turn off FL's dither.
Common mistakes
- Dither on the limiter and again on export: double noise.
- Dither with strong noise shaping on files that will be processed further: the noise in the highs is exaggerated by the mastering.
- Normalizing or changing the volume after the dither.
- Sample rate conversion after the limiter: inter-sample peaks above the ceiling.
- Delivering 16 bit stems: pointless loss of resolution.
Starting settings
| Parameter | Value | Why |
|---|---|---|
| Export for mastering | WAV 24 bit or 32 bit float, no shaped dither | Maximum quality for later processing. |
| 24 bit master | TPDF dither optional | Noise is irrelevant. |
| 16 bit master | Dither mandatory, only once | Tails and fades without grainy distortion. |
| Noise shaping | Moderate for loud music, light or none for dynamic music | Minimum perceived noise. |
| Sample rate conversion | Before the final limiter, maximum quality | No new peaks after the limiter. |
| Typical sample rates | 44.1 kHz music, 48 kHz video and Atmos | Compatibility. |
Frequencies to watch
| Range | What to do |
|---|---|
| 2–5 kHz | Most sensitive region of the ear: noise shaping reduces the dither noise here. |
| Above 15 kHz | Where noise shaping moves the noise. |
Example chain
| Step | Tool | Settings | Why |
|---|---|---|---|
| Mix | Project at the working sample rate | 32 bit float internal | No internal clipping. |
| Mastering | EQ, dynamics, saturation | Mastering values | Final sound. |
| Sample rate conversion | RX Resample / SoX / DAW at maximum quality | If needed | Target format. |
| Limiter | Maximus / Pro-L | Ceiling −1 dBTP | Final peaks and loudness. |
| Dither | FL Studio limiter or export, only once | TPDF or noise shaping | Bit depth reduction. |
How to check the result
- Peak of the final file no higher than the limiter ceiling (−1 dBFS or lower); if it is higher, sample rate conversion or dither was applied after the limiter.
- Fade-outs and tails: listen at high volume to the last seconds of the 16 bit file, no grainy distortion.
- Noise level in the final silence of the 16 bit file between −90 and −96 dBFS (dither present, not doubled).
- Duration and sample rate of the file match those required by the distributor.