Too loud is two different faults
When a track is described as too loud, two completely different problems get confused, and they have different causes, different symptoms and very different repair costs.
The first is clipping: samples pushed past full scale, with the tops of the waveform cut off flat. This is destroyed information and it is permanent.
The second is a true peak over: the digital samples are all within range, but the analogue waveform reconstructed between them exceeds full scale. Nothing is damaged in your file. The distortion appears later, when the platform converts your upload to a lossy format.
Telling them apart is what determines whether you have a five minute fix or an afternoon of rebuilding.
Clipping: permanent damage
Digital audio has a hard ceiling. When a signal is pushed beyond it, the values that would sit above simply cannot be stored, so the peak is flattened into a plateau. That flat top is a square-ish shape, and square shapes contain harmonics that were never in the original sound. What you hear is a crackle, a fizz on the drum transients, or a general hardness that no EQ removes.
Three things make it dangerous. It hides: a single clipped transient in a dense chorus is easy to miss on laptop speakers and obvious on good headphones. It accumulates: clipping introduced in the mix survives every subsequent export. And it is irreversible: exporting, re-importing and mastering again keeps the flat tops exactly where they were.
Concrete action: treat any count of clipped samples above zero as a fault to repair, not a number to tolerate. Also note where the clipping occurs: clustered in one section usually points at a specific element, spread through the track usually points at the master chain.
True peak: the distortion you cannot hear yet
Your DAW meter shows sample peaks: the height of each stored value. But the file is not played back as a series of dots. On playback, a converter reconstructs a continuous waveform through those points, and that curve can rise above the highest sample it passes through. Those excursions are intersample peaks, and true peak metering estimates them.
This is why a master can look clean in your project and distort on a streaming service. Lossy encoding to AAC or Ogg Vorbis is not a bit-perfect process: it reconstructs an approximation of your waveform, and that approximation overshoots. A file sitting at 0 dBTP has no room for the overshoot, so it clips inside the platform's decoder, on a version you never heard.
Concrete action: set the final limiter ceiling to -1 dBTP. That headroom absorbs the overshoot introduced by encoding. Never go above -0.3 dBTP, and set this value last, because any subsequent change to EQ, compression or gain invalidates it.
How to spot both on a finished file
Both faults are properties of the rendered file, so measure the render rather than the session. The session tells you what you intended; the file tells you what you are about to publish.
- Clipping: count the samples pinned at full scale. Anything above zero is a genuine fault. A timeline of where they occur is far more useful than the total.
- True peak: read true peak in dBTP, not sample peak in dBFS. If your meter only offers the second, it cannot tell you what you need to know.
- The listening check: one full pass on headphones at high volume, listening specifically for clicks, fizz and hardness on transients rather than judging the music.
The two faults can also coexist, and they usually do when a limiter has been pushed hard: the clipping comes from the gain going in, the true peak overs from the ceiling being set too close to zero.
How to fix each one
The repair cost is where the two diverge sharply.
True peak overs are cheap. Lower the limiter ceiling to -1 dBTP and re-render. Nothing else in the master changes meaningfully, and the loudness difference is a fraction of a decibel.
Clipping is expensive, and the instinct most people follow makes it worse. Reducing the master fader after the fact does not undo anything: the flat tops are already in the audio, and you have simply made the damaged signal quieter. The fix has to happen upstream.
- Find which element is driving the level past full scale, usually drums, bass or a sub layer.
- Correct the gain staging at that source so nothing in the chain runs hot.
- Rebuild the processing that followed it, because everything downstream was reacting to a signal that has now changed.
- Re-render and re-measure. Only then set loudness, and only after that the true peak ceiling.
This ordering is not pedantry. Fixing loudness before clipping just produces a louder damaged file, and setting the ceiling early means every later change invalidates it.
Preventing it next time
Almost all clipping comes from gain staging that drifted during the mix, one element at a time, until the sum no longer fits.
- Leave headroom in the mix. Aim to have the mix bus peaking around -6 dBFS before mastering starts. There is nothing to gain from arriving at the master with no room.
- Watch the sum, not the parts. Individual tracks can be well behaved while their combination is not.
- Use the limiter for control, not for loudness. If it is pulling more than a couple of decibels consistently, the problem is upstream.
- Check after every render, not once at the end. Discovering clipping the evening before a deadline is how faults reach publication.
These are measured values, not judgement calls: clipped sample count and true peak in dBTP are computed from the file, and the same file always produces the same numbers. It is one of the few parts of finishing a record where you never have to rely on your ears being fresh.