Why is a 24-bit file only 1.5 times a 16-bit file?

Because the sample grows from two bytes to three, not from two to four. The ratio is the ratio of the byte counts, so 24/16 is 1.5. A 32-bit float file, which is what a DAW uses internally, is four bytes and therefore twice the 16-bit size.

How large is an hour of 96 kHz 24-bit stereo?

About 2.07 GB: 96,000 frames a second, six bytes per frame, 3,600 seconds. At 48 kHz it is half that, and at 44.1 kHz 16-bit stereo it is about 635 MB.

Does FLAC halve the file?

Roughly 50 to 70 per cent of the uncompressed size, and the ratio depends on the material rather than the settings. Quiet acoustic material compresses far better than a loud distorted mix, because the predictor has less to fight.

Audio file size: three multiplications and nothing else

Published 27 October 2026 · 9 min read

Uncompressed audio has no cleverness in it. The size is the frame rate, times the frame size, times the duration — and every format that stores raw samples, from WAV to AIFF to BWF, is that number plus a header.

The arithmetic is trivial and the consequences are not, because it sets how long a recorder runs on a card, how much a session costs in storage, and why the choice between 48 and 96 kHz is a bandwidth decision before it is a quality one.

Audio file size calculator

Uncompressed PCM only. A compressed format depends on the material, not on the format.

Sample rate (Hz)

Bit depth

Channels

Duration (minutes)

Frames: —
Bytes per frame: —
Data rate: —
Total size: —
One hour: —
Recording time per gigabyte: —

Bit depth does not double the size when it goes from 16 to 24: a 24-bit sample is three bytes rather than two, so the file grows by half. That factor of 1.5 is behind most of the surprise when a session is moved from 16-bit to 24-bit capture.

Three multiplications

A frame is one sample per channel, taken at one instant. Its size in bytes is the channel count times the bytes per sample, and the byte count matters: 16-bit is two bytes, 24-bit is three, 32-bit float is four. The frame rate is the sample rate. Multiply by the duration and the file size falls out.

bytes = seconds × sampleRate × channels × (bits / 8)

Interleaving matters for the layout and not for the total: a stereo frame is left then right, three bytes each, and the file is a long sequence of those six-byte groups.

The numbers you will meet

44.1 kHz, 16-bit, stereo — 176,400 bytes a second, 10.6 MB a minute, 635 MB an hour.
48 kHz, 24-bit, stereo — 288,000 bytes a second, 17.3 MB a minute, 1.04 GB an hour.
96 kHz, 24-bit, stereo — 576,000 bytes a second, 34.6 MB a minute, 2.07 GB an hour.
48 kHz, 32-bit float, stereo — 384,000 bytes a second, 1.38 GB an hour, and the format most DAWs use on disk between sessions.

Those four are worth knowing because they answer the question that actually gets asked: not "how big" but "how long can I record". A 32 GB card holds about 30 hours at 48/24 stereo, or 15 at 96/24, before any allowance for the filesystem.

What the container adds

A WAV header is 44 bytes and a BWF extension adds a further 602. At the scale of a three-minute song that is nothing; at the scale of ten thousand one-second sound effects it is 6 MB of headers, which is the point at which a sample library's metadata starts to show up in the total.

Filesystems add their own overhead: a cluster is allocated per file, typically 4 kB, so a library of tiny files occupies far more than the sum of their sizes. That is a property of the disk rather than the audio, and it is why "size on disk" and "size" differ in the properties dialog.

Compression changes the question

Lossless compression predicts and codes the difference, so its ratio depends on how predictable the signal is rather than on the sample rate. A 24-bit file compresses better than its size suggests, because the low byte of a 24-bit sample is mostly noise from the analogue front end — the predictor leaves it alone, and it is a third of the payload.

Lossy compression is a different calculation entirely, because it discards what a psychoacoustic model says will not be heard. A 320 kbit/s file is 40 kB a second whatever the source format was, which is why bitrate rather than sample rate is the number quoted for delivery.

Disk arithmetic in practice

Drive capacities are decimal and operating systems report binary, so a 1 TB drive shows as 931 GB. At 1.04 GB an hour for 48/24 stereo, that is about 895 hours — a number worth knowing before deciding that 96 kHz is free.

The honest summary is that storage stopped being the constraint on sample rate some years ago, and bandwidth became it instead. A 96 kHz session is 2 GB an hour to store and the same to back up over a network, and that is the cost that actually bites.

Five mistakes worth avoiding

Assuming 24-bit is twice 16-bit. It is 1.5 times, and 32-bit float is twice.
Confusing bit rate with bit depth. One is bytes per second, the other is resolution per sample.
Forgetting the channel count. Mono is half the figure, and a 5.1 capture is three times it.
Budgeting a card to its label. Filesystem overhead and the recorder's own reserve take a few per cent, which matters on a long take.
Comparing lossless sizes across material. The ratio is a property of the recording, so two albums of the same length can differ by 20 per cent.

Summary

Sample rate times channels times bytes per sample times seconds. Everything else is a header. The useful derived number is hours per gigabyte, because that is what decides whether the next session fits.

Values shown are engineering aids rather than measurements; see the disclaimer and the tool index.