Somebody asks for a WAV and sends back a 400 MB file. An upload form only takes MP3. A recording that was crisp in the app sounds thin after conversion. All of it comes down to two ideas, and neither is complicated.
The one number that decides the size
Bitrate — how much data per second of sound. Multiply it by the length and you have the file, near enough:
| Bitrate | Per minute | An hour |
|---|---|---|
| 64 kbps | about 0.5 MB | about 29 MB |
| 128 kbps | about 1 MB | about 58 MB |
| 192 kbps | about 1.4 MB | about 86 MB |
| 320 kbps | about 2.4 MB | about 144 MB |
| 1411 kbps (CD-quality WAV) | about 10 MB | about 600 MB |
That last row is not a typo, and it is where most surprise comes from. A WAV is not compressed at all — 44,100 samples a second, 16 bits each, two channels. Multiply it out and you get 1,411 kilobits per second, whether the recording is an orchestra or thirty seconds of somebody saying hello.
“128 kbps is about a megabyte a minute” is worth remembering. Everything else is that number scaled.
The two kinds of format
Lossy formats throw sound away — the parts of the signal that are masked by louder sounds nearby, which the ear was not going to separate anyway. MP3, M4A and OGG all work this way. This is where the enormous savings come from.
Lossless formats keep everything. WAV keeps it uncompressed; FLAC compresses it the way a zip file does, so it is about half the size of the same WAV and identical when decoded.
| Format | Kind | Use it when |
|---|---|---|
| MP3 | Lossy | You are sending it to somebody, or a form is fussy. It plays on everything ever made |
| M4A (AAC) | Lossy | You want better sound at the same size. Apple’s default, and what most video uses |
| OGG | Lossy | An open format, expected by some games and Android apps |
| WAV | Lossless | An editor, a studio or a phone system asked for it. Nothing else |
| FLAC | Lossless | Archiving music you want kept perfectly, at half the size of WAV |
For nearly everybody, nearly always: MP3 to send, M4A to keep, WAV only when asked.
What bitrate should you actually pick?
It depends entirely on what the sound is, and the difference between speech and music is larger than most people expect.
- Speech — 64 to 96 kbps, in mono. A lecture, an interview, a voice memo, a podcast. Higher is spent on detail a human voice does not contain.
- Music you are sending casually — 128 kbps. Fine on phone speakers and earbuds on a train.
- Music you care about — 192 to 256 kbps. The point at which most listeners on most equipment stop hearing a difference.
- 320 kbps. The top of MP3. Worth it for a master copy, and largely inaudible against 256 for the rest of us.
Mono is the saving nobody takes
A single voice recorded in stereo is two nearly identical channels, and it costs twice as much to store as one. For a lecture, an interview or a voice memo, mono halves the file and removes nothing that was there. Music is a different matter — stereo is the point of it.
And the sample rate?
44.1 kHz is CD, 48 kHz is what video uses, and both are far beyond what anybody can hear the top of. Higher rates exist for production, where audio is stretched and processed and the extra headroom matters. For listening, they are larger files that sound the same.
The conversion that helps nobody
Taking a 128 kbps MP3 and converting it to 320 kbps produces a file two and a half times the size that sounds — at very best — exactly the same, and slightly worse in practice.
The detail thrown away at 128 is gone. Nothing re-creates it. All the higher bitrate does is describe the damaged version more precisely, and the second pass of lossy compression adds a little damage of its own.
This is the same trap as converting a JPG to PNG and expecting a better picture. A bigger container does not refill itself.
The rule that follows: convert once, from the best source you have. If you have the original, work from that, never from a copy that has already been compressed.
Taking the audio out of a video
A video’s soundtrack is already a compressed audio track sitting inside the file. There are two quite different ways to get it out.
Re-encode it — decode the existing track and compress it again, as MP3 or anything else. This works everywhere and costs a second round of lossy compression.
Lift it out untouched — copy the existing track into an audio file without decoding anything. It is instant, it loses nothing at all, and the format is whatever the video already used, usually AAC. The converter here offers this as “Original”, and when it is available it is the right answer.
If all you want is the sound of a lecture or a song from a video you have, this is also the biggest saving available: hundreds of megabytes of picture, discarded, for something you were only ever going to listen to.
Why voice notes are so small
A ten-minute voice note is often under a megabyte, which seems impossible next to the numbers above.
Two reasons. It is mono, and it uses a codec built specifically for speech at very low bitrates — Opus, in most messaging apps — which is remarkably good at the narrow range a human voice occupies and would be poor at an orchestra. Specialised beats general, when you know what the sound is going to be.
When somebody asks for WAV
Usually one of three reasons, and all of them are real:
- An editor or a studio. Editing lossy audio and re-exporting compounds the loss, so professional workflows keep everything uncompressed until the very end.
- A phone system or a piece of hardware. Voicemail systems, hold music and older devices frequently accept nothing else, sometimes at a specific sample rate.
- Somebody repeating what they were told. Common, and worth a polite question — if the file is going to be listened to rather than worked on, a 192 kbps MP3 is a twentieth of the size and nobody will hear the difference.
If you do have to send a WAV and it is too large, cutting the silence and the parts nobody asked for is the only saving that costs nothing, because the format has no quality setting to lower.