Music Tech3 min read

Audio Codecs for Musicians: What Actually Happens to Your Music

How MP3, AAC, and Opus decide what to throw away, why cymbals suffer most, and what your track sounds like after a platform re-encodes it.

Every listener hears a re-encoded version of your master. Understanding what codecs do explains which mixing decisions survive that process and which don't.

The core trick: masking

Lossy codecs exploit auditory masking — the fact that a loud sound hides quieter sounds near it in frequency and time.

If a loud snare hits at 3kHz, quieter content around 3kHz becomes inaudible for a few milliseconds. The codec knows this and simply doesn't store the masked content. You can't hear what was removed because you couldn't hear it in the original either.

That's the honest basis of lossy compression. It's not "lower quality" so much as "storing only what you can perceive." The problems come when the model is wrong.

Where codecs struggle

Cymbals and hi-hats. Broadband noise with sharp transients is the worst case — it doesn't compress efficiently, and errors are audible as swirling or splashiness.

Sibilance. Vocal "s" sounds are high-frequency noise bursts. Codecs often make them lispy or metallic.

Reverb tails. Low-level, diffuse content spread across the spectrum. Frequently the first thing sacrificed at lower bitrates.

Pre-echo. A sharp transient after silence can smear backwards, producing a faint sound before the hit. Most noticeable on solo percussion and plucked strings.

Very wide stereo. Codecs often use joint stereo, encoding mid and side signals. Extreme width is expensive and can collapse.

The codecs

CodecStrengthTypical use
MP3Universal compatibilityLegacy, references
AACBetter than MP3 at the same bitrateApple Music, YouTube
OpusBest quality at low bitratesSpotify (low tiers), voice, streaming
VorbisOpen, good qualitySpotify (some tiers), games

What this means for your mixes

Leave true-peak headroom. Encoding can push peaks above 0dBFS. Master to -1.0 dBTP so the decoded output doesn't distort.

Don't over-brighten. Excessive high end costs the codec data and is where artifacts appear. A mix that's harsh before encoding is worse after.

Be careful with extreme stereo width. Check your mix in mono; if it collapses badly, joint stereo encoding will hurt it.

Never upload lossy. Uploading an MP3 means double encoding — your artifacts, plus the platform's. Always upload WAV or FLAC.

Check your master encoded. Export a 256kbps AAC and listen. It's what your audience actually hears, and it takes two minutes.

Related reading: audio file formats explained, LUFS and loudness explained, and how to export for streaming.

Frequently asked questions

How do audio codecs decide what to remove?

They use psychoacoustic models — rules about what human hearing cannot detect. The main one is masking: a loud sound hides quieter sounds at nearby frequencies, so the codec discards the hidden content because you would not hear it anyway.

Why do cymbals sound bad in compressed audio?

Cymbals are broadband, noisy, and transient — the hardest content to represent efficiently. Codecs allocate limited data to them, so cymbals and hi-hats are usually where compression artifacts appear first.

Does my master change when Spotify encodes it?

Yes. Platforms re-encode your upload to their own lossy format for delivery. Uploading a lossless master means that happens once; uploading an MP3 means it happens twice, which is audible.

Start making music in Veena

Free, browser-based, no downloads required.

Try Veena Free