Audio Bit Depth

音频位深 refers to the number of binary bits used to represent each sample value in digital audio. For linear integer PCM, n bits can represent 2ⁿ discrete values; 16-bit has 65,536 quantization levels, and 24-bit has 16,777,216 quantization levels. Bit depth describes the amplitude quantization precision; it does not determine the number of samples per second.

Quantization maps continuous amplitudes to a finite set of values, resulting in quantization error. For an ideal linear PCM system with appropriate dithering, each additional bit of bit depth reduces the quantization noise power by approximately 6 dB relative to full scale; therefore, the commonly used theoretical signal-to-noise ratio is approximated as 6.02n + 1.76 dB. This relationship describes an ideal full-amplitude sine wave and a quantization system; it is not equivalent to the actual dynamic range of recording equipment, master tapes, or listening environments. Analog front-ends, inherent noise, and level settings typically impose limitations first.

The effective bit depth may differ from the container’s bit count. A 24-bit file may originate from a converter with lower effective precision, or it may simply place 16-bit data within a 24-bit word; 32-bit floating-point audio uses a sign, exponent, and mantissa representation and cannot be directly interpreted using the quantization step formula for integer PCM. Workstations primarily use floating-point formats for computational headroom and to avoid intermediate clipping, though the final output may still be quantized to 16- or 24-bit integers.

When downsampling from a higher bit depth to a lower one, direct truncation causes errors to correlate with the signal and introduces distortion. Dithering introduces a low-level random signal before quantization, converting this correlated distortion into more stable noise; noise shaping can also shift more noise to frequency bands where the human ear is less sensitive. Dithering does not restore lost bits, nor does it increase the original information content.

The bit depth reported by media tools only indicates encoding or container parameters. Saving 16-bit audio with padding to 24-bit does not introduce new details, nor does decoding lossy audio to 32-bit floating-point restore encoding losses. Determining effective precision requires considering the production chain, noise floor, and sample data—not just file tags.