WAV 16-bit vs 24-bit vs 32-bit float — picking an audio bit depth
Compare audio bit depths for recording, editing, and distribution by dynamic range, file size, clipping headroom, and compatibility. Why 32-bit float matters during editing and when 16-bit is fine for distribution.
Four axes — range, size, clip resilience, compatibility
Picking a WAV bit depth is not “higher number equals better quality” — it is four trade-offs running in parallel. Dynamic range is the ratio between the quietest and loudest sample a format can encode, and bit depth sets it directly (96 dB at 16-bit, 144 dB at 24-bit). File size scales linearly with bit depth, which matters for distribution and long recording sessions. Clip resilience (headroom) decides whether a momentary level spike destroys the sample, and is the fundamental difference between integer PCM and 32-bit float. Compatibility matters on the downstream side — CD pressing, streaming platforms, video editors, and DAWs each have their own preferences.
The modern shorthand is “record in 32-bit float, distribute at 16-bit,” and the four axes explain why. On the recording side, you need a safety net against a blown level setting that you cannot reshoot; on the distribution side, the streaming platform re-encodes your file anyway, so spending bits beyond 16 is wasted.
Side-by-side comparison
| Property | 16-bit PCM | 24-bit PCM | 32-bit float |
|---|---|---|---|
| Numeric type | Signed integer | Signed integer | IEEE 754 floating point |
| Value range | -32,768 to +32,767 | About ±8,388,608 | Practically ±3.4×10^38 |
| Theoretical dynamic range | About 96 dB | About 144 dB | About 1,500 dB |
| 1 minute at 44.1 kHz stereo | About 10.1 MB | About 15.1 MB | About 20.2 MB |
| 1 hour at the same settings | About 605 MB | About 908 MB | About 1.21 GB |
| Recover from a clip | No (data destroyed) | No (data destroyed) | Yes (values past -1.0 / +1.0 preserved) |
| Typical use | CD / streaming masters | Pro recording / mid-mix | Field recording / DAW projects |
| Standard support | Every player / every OS | DAWs and most players | DAWs only, many consumer players reject it |
| Dither needed | When down-converting from higher | Usually no | Required when converting to 16-bit |
A 96 dB dynamic range comfortably spans the human hearing range (a quiet room is around 30 dB, a live concert around 110 dB), but that is only true when the recording level was set optimally. A cautious engineer who tracks peaks at -20 dBFS effectively loses 20 dB of usable range, leaving about 76 dB. Move to 24-bit and the same cautious setup still has 124 dB of usable range, and gain adjustments during the mix never push quantisation noise into earshot.
The reason 32-bit float took over field recording is the “recover from a clip” row in the table above. Modern recorders — Zoom F6, Sound Devices MixPre, Zoom F3 — adopt 32-bit float because the red light is no longer fatal. With integer PCM, the moment an analog input crosses ±full-scale the sample saturates and that information is gone forever. With 32-bit float, values past -1.0 or +1.0 are encoded directly, so dropping the gain in the DAW afterwards reconstructs the waveform exactly. An unexpected shout, applause, or siren no longer ruins a take — you keep rolling and fix it in post, which is decisive in interviews, documentaries, and live recordings where the moment will not happen again.
Recommendations by use case
Streaming to Spotify, Apple Music, YouTube: 16-bit / 44.1 kHz. The platform re-encodes to AAC or Opus on ingest, so 24-bit or 32-bit float buys you nothing past mastering. Dither down to 16-bit at the final step and ship the standard distribution format.
Field recording, interviews, documentary: 32-bit float / 48 kHz. Recoverable level mistakes are priceless when the moment cannot be repeated. Zoom F6 / F3 and Sound Devices MixPre-II and newer all support it natively.
DAW project intermediates: 32-bit float, or 24-bit. Gain stages and plugin processing accumulate rounding error, and floating point makes that error inaudible.
Video voiceover / amateur podcast recording: 24-bit / 48 kHz. A mistracked level still has 144 dB of headroom to recover from. Even without a 32-bit float recorder, almost every USB audio interface supports 24-bit.
CD mastering: 16-bit / 44.1 kHz (the spec is fixed). Bring in 24-bit or 32-bit float sources, then dither down to 16-bit as the final step.
Sample rate (44.1 kHz / 48 kHz / 96 kHz) is chosen independently of bit depth. Music distribution defaults to 44.1 kHz, video work to 48 kHz, hi-res audio to 96 kHz or 192 kHz. Staying on a single rate from recording through distribution minimises the number of sample-rate conversion stages and the artifacts that come with them.
Browser-only conversion and the dithering trap
When dropping from a higher bit depth to 16-bit, you must apply dither (a tiny pseudo-random noise added before truncation). Without dither the samples get rounded and the quantisation error correlates with the signal, leaving audible distortion (typically as harmonic artefacts around quiet passages). Going the other direction — 16-bit up to 24-bit or 32-bit float — only adds zero bits and does not need dither (dither would be a no-op there).
When you need to convert a WAV’s bit depth or container format on your own machine, audio-convert handles the conversion entirely in the browser. It supports WAV, MP3, OGG, FLAC, AAC, and M4A in both directions, and multiple files come back as a single ZIP. Unlike upload-based converters, there is no code path that ships an unreleased master or a confidential interview to a remote server. The implementation is open on GitHub, and the DevTools Network tab makes it trivial to confirm nothing leaves the browser during conversion — a meaningful difference when you are handling client deliverables under NDA.