Skip to content

Bit depth in audio: 16-bit, 24-bit and 32-bit float

Send feedback

Last updated 2 min readTags:Getting startedExport

How bit depth affects noise, file size and peaks above full scale, and which format to export.

Use 24-bit PCM for most mix exports, including a Sonalix upload. Choose 32-bit float for an intermediate WAV when you want to preserve sample values above 0 dBFS. Use 16-bit when the destination requires it, such as an audio CD.

What bit depth changes

In integer PCM, bit depth determines how finely the file represents a sample's amplitude. Each extra bit adds roughly 6 dB of theoretical dynamic range, giving about 96 dB at 16-bit and 144 dB at 24-bit. Microphones, converters and the room usually set the practical noise floor well before a 24-bit file does. Sample rate determines the upper frequency limit.

Format Precision or range Can store samples above 0 dBFS? Stereo WAV per minute
16-bit PCM About 96 dB theoretical range No 10.6 MB at 44.1 kHz
24-bit PCM About 144 dB theoretical range No 15.9 MB at 44.1 kHz
32-bit float 24 significant bits for normal values, with a varying scale Yes 23.0 MB at 48 kHz

At the same sample rate, 24-bit PCM takes 50% more space than 16-bit PCM. The extra range is useful in a working file. For final 16-bit delivery, apply dither at the point of reduction, after sample-rate conversion and other processing.

What a float export preserves

Fixed-point PCM clips sample values above 0 dBFS when it writes the file. A float WAV can preserve them. Pull down the gain before a subsequent integer export. This works only if the previous stage preserved the peaks. Lowering the gain on an already clipped file leaves the clipped waveform intact. See headroom.

What Sonalix reads

Sonalix decodes uploads to 32-bit float. It can represent 16-bit and 24-bit PCM samples exactly. Values from 32-bit integer or 64-bit float sources round to float precision, with relative error on the order of 150 dB below the sample for normal values.

In one Sonalix test, 16-bit and 24-bit exports gave the same displayed readings. A float WAV with sample peaks of +3.5 dBFS measured +3.52 dBTP in that test. The two numbers describe different peaks. A positive true peak may come from the reconstructed waveform even if every stored sample is below full scale. Likewise, a displayed 0.0 dBTP on an integer file calls for a closer look at sample peaks and the waveform before concluding that it clipped. See true peak and the export settings for analysis.

Was this page helpful?

    Type to search titles, headings and summaries.