MP3, AAC, FLAC or Opus: which audio format for which job

Audio formats split cleanly in two, and almost every confusion comes from mixing the halves up. Lossy formats — MP3, AAC, Opus — permanently discard sound to save space. Lossless formats — FLAC, ALAC, WAV — keep every sample exactly. Choosing well is mostly a matter of knowing which half your job belongs in.

What lossy compression actually throws away

Not "the quiet bits" and not "the high frequencies", which are the two common guesses. Lossy audio codecs exploit psychoacoustic masking — measured facts about what the human ear cannot hear.

The main one is simultaneous masking: a loud sound makes quieter sounds at nearby frequencies inaudible while it is playing. A cymbal crash hides a great deal of what is happening around it. There is also temporal masking — for a few milliseconds after a loud transient, the ear is effectively deaf to quiet sounds.

So the encoder models your hearing, works out what you could not have heard anyway, and spends no bits on it. At a decent bitrate the discarded material is genuinely inaudible, which is why a well-encoded 256 kbps AAC is indistinguishable from the master for almost everyone in almost every listening situation.

The catch: those decisions are permanent, and they are made on the assumption that this is the final encode. Re-encoding a lossy file breaks the assumption, which is why the second half of this article matters more than the first.

MP3

Finalised in 1993 and still the universal answer. Every device, every car stereo, every piece of software, every website. It is measurably less efficient than everything below it — you need roughly 30% more bitrate for the same quality as AAC — and it does not matter when compatibility is what you are buying.

Practical bitrates: 320 kbps is transparent for essentially everyone. 192–256 kbps is transparent for essentially everyone in real listening conditions. 128 kbps is where artefacts become audible on cymbals and applause if you listen for them. Below 128, MP3 falls apart quickly.

Use it for anything going to a device you have not seen, and for anything going into a system that predates about 2010.

AAC

The successor, and the default for Apple, YouTube and most streaming. Roughly 30% more efficient than MP3 — a 192 kbps AAC is about as good as a 256 kbps MP3 — and notably better at low bitrates, where MP3 becomes unpleasant and AAC merely becomes soft.

Support is near-universal on anything modern: every phone, every browser, every recent car. The gap is old hardware and a few stubborn corners of the Linux world.

AAC is the right default for new audio in 2026 unless you specifically need MP3's reach.

Opus

The best lossy codec here by a clear margin, and royalty-free. It is transparent at around 128 kbps for stereo music, and it remains genuinely usable at bitrates where MP3 is a joke — 32 kbps mono speech is clear and natural. Every voice call you make over the internet is almost certainly Opus.

Its weakness is entirely outside the browser. Browsers all support it; car stereos, older TVs, hardware players and a lot of editing software do not. It lives in .opus or .ogg files that plenty of consumer devices simply refuse.

Use Opus for audio on your own website, for voice, and for podcasts delivered through a player you control. Do not email an .opus file to your father.

FLAC (and WAV, and ALAC)

Lossless. FLAC compresses to roughly 50–60% of the original size with every sample bit-identical on decompression — the audio equivalent of a ZIP, not of a JPG.

The important property is not audiophile quality. It is that a lossless file can be re-encoded any number of times without loss. It is the correct format for anything you will edit, master, archive or convert again later.

WAV is uncompressed — same audio as FLAC, roughly twice the size, universally supported by editing software. ALAC is Apple's lossless format and is, for practical purposes, FLAC with a different logo.

Keep masters and archives in FLAC. Deliver in something lossy. Never distribute WAV to listeners; you are sending twice the bytes for zero audible difference.

The mistake that costs real quality

Converting one lossy format to another. MP3 to AAC, AAC to MP3, either to Opus — this is called transcoding, and it compounds damage.

The reason is specific. Each codec has its own model of what is inaudible. The first encoder removed everything it thought you could not hear. The second encoder then looks at what survived — a signal already full of the first codec's artefacts — and removes what it thinks you cannot hear, including material the first one deliberately kept. The two models disagree, and the disagreement is audible.

So: always convert from the most original file you have. If the only copy is a 128 kbps MP3, converting it to 320 kbps AAC produces a bigger file that sounds slightly worse. Bitrate is a ceiling, never a floor.

Our audio converter and audio compressor both run FFmpeg in the browser, so the conversion is real and the file never leaves your machine — but they cannot repeal this rule, and neither can anything else.

The short version

• Sending to someone, unknown device → MP3, 192 kbps or better
• General listening, modern devices → AAC, 192 kbps
• Audio on your own website → Opus, with an AAC fallback
• Speech, podcasts, voice notes → Opus at 48–64 kbps, or AAC at 96 kbps for compatibility
• Masters, archives, anything to be edited → FLAC
• Into an editor → WAV or FLAC, never a lossy file

And one honest note on bitrate: above roughly 256 kbps AAC or 320 kbps MP3, in a normal room on normal headphones, nobody reliably hears a difference in blind testing. Storage is cheap enough that erring high costs nothing — but if you are choosing 320 over 256 because you can hear it, the evidence says you probably cannot.