Delivery Bit Depth
Delivery depth follows the receiver, and dither belongs only on the final reduction.
Related
Bit depth is not a volume setting
Save a 16-bit file as 24-bit and nothing new appears except eight zeroes. Bit depth sets headroom while recording and decides, at delivery, whether the quietest passage survives. Once the source is already 16-bit, a larger word length is just a larger container.
The receiver decides
| Destination | Depth | Dither | Note |
|---|---|---|---|
| Digital release WAV | 24-bit | No | The platform transcodes; leave 1 dB of true-peak margin |
| CD or 16-bit master | 16-bit | Once, at the end | TPDF or noise shaping, as the label asks |
| Film and broadcast post | 24-bit | No | Usually paired with 48 kHz and headroom |
| Session exchange | 32-bit float | No | Only by agreement, never for release |
| Stems | 24-bit | No | Equal length, common start, common rate |
The table is not a ranking of sound quality. If the receiver asks for 16 bits, send 16 bits; sending 24 and letting them convert pushes the reduction and the dither downstream.
Dither once, on the last reduction
Dither scatters quantization error into low-level noise, at the cost of a slightly raised floor. It belongs to the final reduction: add it once when exporting 16-bit from 32-bit float. Do not add it to intermediate stages, to 24-bit files, or to a version that still faces mastering. A dithered file should not be gained, resampled or limited afterwards, or the noise is reshaped and the earlier decision is wasted. Noise shaping pushes noise toward high frequencies, which suits CD but not necessarily streaming transcodes; when in doubt, use flat TPDF.
Five checks before export
- The depth matches the receiver's written spec, not a guess
- The 16-bit version comes from 32-bit float or 24-bit in one pass, with one dither
- True peak is measured again after reduction, since dither can raise it slightly
- 24-bit files are not pointlessly upconverted or dithered
- The filename and delivery note state both depth and sample rate
Where it breaks
A 24-bit container holding 16-bit audio changes nothing audible but suggests headroom that is not there. Dithering material that is already limited makes the noise pump with the music and dirties the quiet bars. Upconverting 16-bit to 32-bit float restores none of the lost detail and amplifies the quantization noise along with everything else. The quietest failure: a 24-bit file goes to mastering, mastering reduces to 16-bit and dithers, and the earlier version was already dithered, so the noise is stacked twice.
Listen in the library
All tracks from midicn-lib — inline badge is the license tier (C1 commercial / C2 non-commercial / C3 study only).
Along the same line
Sources
- Audio Engineering Society, AES3-2009，数字音频接口的字长与传输规定（标准）
- Ken C. Pohlmann, Principles of Digital Audio，量化、抖动与噪声整形（教材）
- BBC 交付技术规范，音频交付文件的位深与文件形态要求（机构规范）