What MP3 Bitrate Should I Use? Balancing Quality and File Size

The honest short answer: 192kbps MP3 for anything you care about, 128 for speech, and if you get to choose the codec, do not choose MP3.

The longer answer is worth having, because the usual advice is either “320 always” (wasteful) or “128 is fine” (it is not, for music), and neither tells you how to check whether the file in front of you actually got what you asked for.

What a bitrate buys

A lossy encoder does not compress. It discards, and the first thing it discards is the top of the frequency range: it is expensive to store, most adults cannot hear above 16kHz, and cutting it is the cheapest way to hit a target.

So the bitrate you choose sets, fairly directly, where the audio stops. Measured on one source encoded down a ladder:

The interesting part of that table is where it flattens. Between 192 and 320 you are paying 67% more storage for 1.4kHz at the very top of hearing. Between 96 and 128 you are buying 1.4kHz right in the middle of where cymbals, sibilance and presence live. The curve is steep exactly where it matters and flat exactly where it does not.

Why the codec matters more than the number

kbps is not comparable across codecs. Roughly, from the published equivalences:

So a 96kbps Opus file is not a low-quality file. Judging it by MP3-era instincts marks down the best codec in the list for being efficient — which is exactly the mistake to avoid when someone hands you a modern file and it “looks low”.

If you control the format: Opus for anything streamed or stored at scale, AAC for compatibility with everything, MP3 only when something downstream genuinely cannot read anything else.

The recommendation, by what you are storing

• Music you will keep — 192kbps MP3 / 128 AAC / 96 Opus. Above this you are buying storage, not sound.
• Music you are mastering or will re-edit — lossless. Not because you can hear the difference in one pass, but because every subsequent export compounds on what the first one discarded.
• Speech, podcasts, interviews — 96–128kbps MP3, or 64 Opus. Voice occupies a narrow band and the top of the range carries little.
• Anything going into a video — match or beat the audio you were given. Re-encoding audio for a video export is where quality quietly leaks.

The number in the header is not the number in the file

This is the part almost nobody checks. A bitrate describes the last time a file was saved, not what is inside it. A 128kbps download, re-exported at 320, declares 320kbps in every byte of metadata and still contains 128kbps of audio — the encoder faithfully reproduced a file that had nothing above 16.5kHz, using two and a half times the bits to do it.

Nothing in the file’s properties reveals this. The audio does, plainly: measure the highest frequency still carrying real signal and you have a fingerprint of the lowest-quality stage the file has ever been through, regardless of what happened after.

Measured on a real laundered file: an Excellent bitrate reported beside content stopping at 10.9kHz of a possible 22.1.

One caveat worth stating, because it is the limit of the technique: this cannot tell a codec’s cutoff from a deliberate low-pass filter. A gentle filter at 16kHz and a 128kbps MP3 look similar. So the finding is always about the content — where the audio stops — and never a claim about how it got that way.

Check a file

Upload one audio file and DiffALL reports its frequency range against what its sample rate could carry, its bitrate adjusted for its actual codec and channel count, its loudness in LUFS, whether it is clipped, and whether a stereo file is really stereo. Each with a verdict and the measurement behind it.

If you have the before and after — the master and the export — comparing them tells you what that specific encode cost, which is the number you actually want when choosing a setting.