Bit Depth Converter

How the Bit Depth Converter works

Bit depth converter: getting a good result

Bit depth is how finely each sample's level is measured. Sixteen bits gives 65,536 steps and a noise floor about 96 dB below full scale; twenty-four bits gives sixteen million steps and a floor around 144 dB, well below anything a room or a converter can reproduce. Recording and mixing at 24 bits keeps headroom for processing; delivery is nearly always 16 bits because CD, most streaming ingest and countless devices expect it.

Why dither matters. Simply rounding 24-bit samples to 16 bits produces quantisation error that is correlated with the signal. On loud material it is inaudible, but on fades, reverb tails and quiet acoustic passages it turns into a harsh, gritty distortion. Dither adds a tiny amount of random noise before rounding, which breaks that correlation and turns the distortion into a steady, benign hiss far below the music. The page uses TPDF dither — triangular probability density, the sum of two uniform random values — which is the standard choice because it removes both the distortion and any modulation of the noise floor.

Noise shaping. The dither noise can be pushed towards high frequencies where hearing is least sensitive. The page's noise-shaping option uses a first-order error-feedback filter, which lowers the audible noise in the midrange by several decibels at the cost of a little more energy above 10 kHz. It is a good default for music that will be listened to at normal levels; choose plain TPDF if the file will be processed again afterwards, because shaped noise can accumulate through further stages.

Choosing the target. The export format selector sets the output depth. Pick 16-bit to reduce and 24-bit to expand. Expanding a 16-bit file does not need dither and the page applies none; the extra bits are simply zero and the file becomes a container with more headroom for whatever you do next.

Gain. A gain control is included because bit depth reduction is usually the last step and a final trim is often needed. Positive gain that would clip is clamped and the stats panel warns you; if that happens, lower the gain or run the file through the normaliser first.

The source depth. Browsers decode everything to 32-bit float, so the page estimates the original bit depth from the file size for WAV and AIFF and reports it. For compressed sources such as MP3 or FLAC the source depth is shown as decoded float.

Bit-exact output. When dither is off and gain is zero the conversion is a pure rounding and every sample is written exactly, so a 16-bit file passed through at 16-bit comes out identical.

Related tools