320 kbps vs 192 kbps: what you actually get

They are staring at a bitrate dropdown and do not want to pick wrong. Every converter site nudges them to the biggest number because it sounds generous. They want to know whether it matters before they re-encode a hundred files.

Pick 192 unless you have a specific reason not to. A 320 kbps MP3 is about 66% larger than the same recording at 192, and the clearest thing those extra bytes buy is a slice of treble above roughly 18.6 kHz — a band most adults stopped hearing years ago. In a properly blinded comparison the two are hard to separate on music, and nobody has shown it on speech. Where 320 does earn its bytes is narrow and worth knowing, so it is set out below.

The one real difference: the lowpass filter

An MP3 encoder is not given a quality dial. It is given a byte budget, and it decides what to throw away to hit it. The first thing every encoder throws away is the top of the frequency range, because that is where the fewest people notice. LAME — the encoder this site and most others use — does this with a lowpass filter whose cutoff moves with the bitrate you ask for.

LAME carries the cutoffs in its own source, in a bitrate-to-frequency table called freq_map. Older versions differ by a few hundred hertz, and the filter rolls off rather than stopping dead, so treat these as the middle of a slope:

So the treble 320 buys you over 192 lives between about 18.6 kHz and 20.5 kHz — under two kilohertz, at the very top. Human hearing is usually described as topping out at 20 kHz, and that figure comes from healthy children. Hearing at the top of the range declines with age for almost everyone: by the mid-twenties many people are done above 17 kHz, and by the forties 14 to 15 kHz is common. If you cannot hear a 19 kHz tone at all, the extra band is bytes spent on silence you will never hear.

The second difference is subtler. With more budget the encoder makes fewer compromises lower down too: less pre-echo on sharp transients, fewer artefacts on dense material. That is real, and it is what the listening tests below are actually chasing.

Why blind tests keep coming up empty

Sighted listening is worthless here. If you know which file is which, you will hear a difference, because expectation is louder than 19 kHz. The standard method is ABX: you get A, you get B, and you get X, which is secretly one of them, and you have to say which. Repeat it and count.

The arithmetic is unforgiving. Guessing wins half the trials, so a run of 16 needs 12 correct before it means anything. Twelve or better by luck alone comes up 3.8% of the time, just inside the 5% threshold people conventionally use. Ten out of sixteen feels like a result and is not one: pure guessing reaches it 23% of the time.

This is why the well-known public listening tests are run at 64, 96 and 128 kbps. Those are the bitrates where codecs can still be told apart. Push the test up to 192 and the results collapse into noise: listeners score near chance, and the few who beat it are usually working with one specific "killer sample" — applause, massed cymbals, a harpsichord, a solo triangle — rather than with music in general.

What 320 costs you in storage and in time

Bitrate arithmetic is the one bit of audio maths worth memorising: kilobits per second divided by eight is kilobytes per second. From that:

Going from 192 to 320 across a hundred tracks is not a rounding error; it is roughly another 366 MB per hundred tracks, and on a phone that is the difference between the library fitting and not. If size is the thing pressing on you, the file size tool lets you name a target in megabytes and works the bitrate out backwards.

Encoding time barely moves between the two. The encoder runs the same psychoacoustic analysis either way and only spends the resulting bits differently, so a 320 kbps encode is not meaningfully slower than a 192 kbps one. What changes is everything afterwards — the copy, the sync, the transfer to a podcast host, the download over a bad connection.

Test it on your own file

Do not take anyone's word for it, including ours. The test takes five minutes:

1. Start from a lossless source — WAV, FLAC, ALAC or a CD rip. Testing with an MP3 tells you nothing, because both outputs inherit its damage.
2. Run it through the converter twice: once at 192, once at 320. Nothing leaves your browser, so you can do this with material you would not put on a stranger's server.
3. Rename them a.mp3 and b.mp3, then have someone else shuffle which is which — or use a player set to shuffle and cover the screen.
4. Listen on the equipment you actually use. If that is a phone speaker or Bluetooth earbuds, the answer is more or less settled already: Bluetooth re-encodes everything it sends — SBC, AAC, aptX or LDAC, all of them lossy — so your two files are compressed a second time on the way to your ears, and a difference this fine is unlikely to survive the trip.
5. Score at least sixteen goes. Fewer than twelve right means you did not hear it.

Compare each MP3 against the lossless original as well, not just against each other. "Can I tell 192 from the source" is the question that decides whether the conversion cost you anything; "can I tell 192 from 320" is a much finer distinction and almost nobody wins it.

When 320 is right, and when it is waste

Reasons to use 320 that survive scrutiny:

• This MP3 is your only copy. If the lossless master is going away, the extra headroom is cheap insurance against a decision you cannot revisit.
• Something downstream will re-encode it. Podcast platforms, video editors and messaging apps often compress again. Every generation costs, so starting higher leaves more to lose.
• A spec demands it. DJ pools, some broadcast deliverables and some client contracts name 320 explicitly. Arguing is not worth the fee.
• The material is a known killer. Applause, harpsichord, close-miked cymbals, some electronic music with a lot of energy above 15 kHz. If you can ABX it on your own recording, believe your own ears.

Reasons it is waste:

• The source is speech — a voice memo, an interview, a lecture.
• The source is already lossy at a lower bitrate.
• You are emailing it, or sending it somewhere that has a size cap.
• You listen on Bluetooth, in a car, on a commute, or through a laptop speaker.
• You are converting an M4A into an MP3 only because something refuses to open M4A. The goal there is compatibility, and 192 gets you compatibility at 60% of the size.

Speech is a different question entirely

Everything above is about music. A human voice occupies a much narrower band, has none of the high-frequency detail bitrate is spent on, and is often mono to start with. Encoding a recorded interview at 320 kbps is like posting a postcard in a shipping crate.

For spoken word, 64 kbps mono is genuinely fine and 96 kbps mono is roomy. That is where a long recording stops being a problem: two hours at 64 kbps mono is about 55 MB, where the same interview at 192 kbps stereo is around 165 MB of mostly duplicated channel.

Switching stereo to mono halves the file on its own, and for a single voice recorded on a phone it removes nothing, because both channels carry near-identical audio. That is usually a bigger win than any bitrate change.

VBR, and why the ladder stops at 320

Constant bitrate spends the same bytes on a drum solo and on a fade-out. Variable bitrate does not: you set a quality target and the encoder spends where the music is difficult. Encoding the same three-second stereo fixture on this build produced 122,297 bytes at constant 320 and 40,775 bytes at VBR V0. That fixture was simple material, which is exactly where constant bitrate wastes the most; on dense music the gap narrows. VBR is what most people who have tested this properly end up recommending. Worth knowing — but the converter here works from the bitrate ladder rather than a VBR quality dial, so every figure on this page is a constant-bitrate figure.

As for the ceiling: 320 kbps is simply the highest bitrate the MPEG-1 Layer III specification defines. It is the top of the format, not a grade of quality, and it is nowhere near uncompressed — a CD carries 1,411 kbps, so even 320 is throwing away more than three-quarters of the data. If you want the whole thing, you want a lossless format, and the WAV versus MP3 comparison covers what that actually gets you and what it costs.

The ladder here is 64, 96, 128, 192, 256 and 320, with 192 selected to begin with. The app's ladder starts at 128; the two rungs below that are web-only, because speech does not need more and a long interview is the file that actually causes trouble. For OGG the number is translated rather than obeyed, because Vorbis works on a quality scale: 192 maps to quality 6 and 256 to quality 8, and the file averages roughly the bitrate you asked for instead of hitting it exactly.