Why Some Words Jump Out and How to Fix It Permanently
Vocal mixing · 4 min read
Uneven vocal levels have three causes: the performance, the mic distance, or a single hot consonant. Diagnose which one you have by level-matching the loud and quiet word and comparing their tone, then fix it in three stages — clip gain into a 3 dB window, compression, and 0.5 dB automation moves per phrase. Compression alone fixes only one of the three.
Three causes
3 FIXES
Performance · mic distance · one consonant
The 60-second answer
Level-match the loudest and quietest word with clip gain and listen. Same tone means performance dynamics: clip gain, compression, automation. Duller on the quiet word means off-axis: add a high shelf. Boomy on the loud word means they moved in: add a 200 Hz dip. A spike with no loudness change is a consonant: clip it.
Why the meter cannot tell you which problem you have
All three look the same on a meter: one word 6 dB above its neighbours. They are different physical events. Performance dynamics are a level change with the tone intact. Mic distance is a level change plus a tone change, because off-axis loses highs and moving in adds lows through the proximity effect. A consonant spike is a peak change with no loudness change, because the ear averages over 200–300 ms and a 2 ms transient barely registers.
What a compressor does to a tone problem
If a line is quiet because the singer turned 40 degrees off the mic, it is also 3–4 dB darker above 5 kHz and boxier around 300 Hz. Turn it up with a compressor and you have a loud, dull line next to bright ones. The level matches, the mix still sounds wrong, and you go hunting for an EQ problem that does not exist. That is the most common reason a vocal feels uneven after heavy compression.
The three-stage fix, in order
| # | Do this |
|---|---|
| 01 |
Diagnose with the tone test Copy the loudest and quietest word to a new track, clip-gain them to the same loudness, loop them. Listen for tone, not level. 30 seconds, saves an hour |
| 02 |
Stage 1: clip gain into a 3 dB window Split at breaths, ride each phrase, then fix the outliers. Loudest and quietest words end up within 3 dB of each other. Clip gain: +/-3 dB |
| 03 |
If it was mic distance, add the matching EQ move On off-axis clips, 2–3 dB of high shelf above 6 kHz. On close, boomy clips, 2–3 dB out around 200 Hz with a wide bell. Shelf 6 kHz / bell 200 Hz |
| 04 |
If it was a consonant, use peak control A hard P or T that spikes without sounding loud is a peak problem. Clip it or de-ess the band, do not lower the threshold. Clip 1–2 dB |
| 05 |
Stage 2: compression, split in two Slow compressor for 2–3 dB, fast compressor for 2–3 dB, with different attack times. Total GR 5–6 dB |
| 06 |
Stage 3: automation after the chain 0.5–1 dB per phrase on the post-insert fader, plus a 1 dB bump on every word you cannot make out. If that is not enough, re-cut the line. Moves: 0.5–1 dB |
The trick: the tone test tells you which fix to use in thirty seconds
Copy the loudest word in the song and the quietest word onto a fresh track next to each other. Use clip gain to make them the same loudness — not the same peak, the same loudness by ear — and loop them back to back. If they sound like the same person in the same place, you have a pure level problem, and clip gain plus compression fixes it permanently. If the quiet one sounds duller and boxier, the singer turned off the microphone axis, and no compressor fixes that because you are amplifying a different tone rather than a quieter one: add 2–3 dB of high shelf on those clips. If the loud one sounds thick below 250 Hz, they moved closer and that is the proximity effect, so add a wide 2–3 dB dip at 200 Hz. And if the two sound equally loud even though the meter said 5 dB, you were looking at a consonant peak and the answer is a clipper.
Stop guessing the chain.
The Vocal Chain Bible is 88 real vocal chains from Grammy-winning engineers — every plugin, every setting, in order, with free alternatives for each one. 5+ years, 100+ interviews, one PDF.
Get the Vocal Chain Bible — $48 →Mistakes that kill it
- Lowering the threshold until it stops. Ten dB of gain reduction to catch one shouted word flattens the other two hundred words in the song.
- Deleting the loud word and pasting a quieter one. The room tone either side never matches, and the edit is audible on headphones.
- Automating before compressing. Change the compressor afterwards and every move is wrong by a different amount in every bar.
- Fixing a tone problem with a limiter. The word gets loud enough and still sounds like a different session, because it was a different mic angle.
Fifteen minutes of diagnosis and clip gain gets you further than any plugin. Pros make it look fast because they identify the cause before choosing the tool.
Keep going
FAQ
Why does one word in my vocal always jump out?
Usually an emphasised syllable, or a moment the singer leaned into the mic. Level-match it against a quiet word and compare the tone to find out which.
Can a compressor fix uneven vocal levels on its own?
Only if the unevenness is pure level. If the tone changes with the level, it is mic distance and you need clip gain plus a small EQ move.
How much level variation should a finished vocal have?
Word to word, roughly 3 dB. Verse to chorus, 1–2 dB. Zero variation sounds robotic.
Who's behind this
Ivan Meshcheriakov · IvanFromRnD
5+ years ago I moved to LA with zero connections and started building from nothing — 14-hour days, meeting people and interning at studios. It took me 5+ years to meet and interview 100+ Grammy-winning engineers — then I turned it all into the Vocal Chain Bible.
Read my full story →