Mix the shape of the line before the words, set the sibilance control wider than an English preset asks, and test the result on listeners who do not speak the language. I composed, produced, mixed and mastered "Stefania" for Kalush Orchestra, the song that won the 2022 Eurovision Song Contest performed entirely in Ukrainian, and the Qobuz release metadata from Sony lists all four of those roles. The language shaped half of the engineering decisions on that record, and the ones that follow apply to any vocal in a language the audience will feel before they parse it.
What does Ukrainian hand a mix that English does not?
Two things, and they pull in opposite directions. The vowels are stable: Ukrainian has six vowel phonemes, and unstressed vowels keep their identity instead of collapsing toward a neutral schwa the way English unstressed syllables do. Sung on a short note, an unstressed Ukrainian vowel still reads as itself, which keeps melodic lines open and makes long held notes flattering. That is the friendly half.
The consonants are the demanding half. The language runs long clusters, palatalizes most consonants before front vowels, and carries affricates that English writes with two letters or does not have at all. The letter shch stands for a fricative followed by an affricate, two separate high-frequency events inside what a singer feels as one sound. A line can chain s, sh, ch and shch within a single bar. On a bright pop vocal chain, that chain of events is where the mix wins or loses.
Why does a stock de-esser miss Ukrainian sibilants?
Because most presets were tuned on English material, where the offenders cluster in a fairly narrow band and arrive one at a time. A Ukrainian phrase spreads sibilant energy across roughly 4 to 9 kHz and stacks events back to back, so a narrow detector catches the s, lets the shch through, and the vocal spits twice per line anyway.
What I do instead. Split-band de-essing with the detection band opened wide enough to see the whole range, threshold set so the average event loses 2 to 3 dB and the worst one never loses more than 4. Past that point the consonants start to lisp, and a lisping consonant in a language the listener does not know reads as a mix flaw, since there is no lexical knowledge to fill the gap. When the material is dense I would rather run two gentle de-essers in series than one working hard, the same logic I use for serial vocal compression, and the second one sits after the compressor because compression pushes sibilance back up. The fuller walkthrough of that trade lives in how I de-ess without dulling the vocal.
What keeps fast consonant clusters readable?
Attack time and clip gain, in that order. A vocal compressor with an attack under 2 ms flattens the consonant transients that make clustered syllables legible, so on this material my attack sits at 5 to 10 ms and the ratio stays moderate while the input is ridden by hand. Clip gain per phrase does what a fast compressor would otherwise be asked to do, and it does it without chewing the consonants. The click of a palatalized t or d is information; remove it and a listener who does not speak the language hears mush where a native speaker would still reconstruct the word.
Doubles get the same respect. Ukrainian backing stacks tend to smear precisely because every double repeats the same dense cluster a few milliseconds apart. Tightening doubles to the lead by transient, in the 10 to 20 ms window, cleans more perceived sibilance than any de-esser, because half of what reads as harsh s is actually two s sounds arriving separately.
Where do the words live in the spectrum, and what moves aside?
Diction sits mostly between 2 and 4 kHz, and Ukrainian pop arrangements love exactly that band: sopilka lines, bandura attacks, bright plucked synths. The fix belongs to the arrangement before the EQ. Where an instrumental hook and a vocal line overlap in time, one of them moves, and only after that do I dip the instrument bus 1 to 2 dB around 3 kHz for the phrases where they still touch. On "Stefania" the folk instrument states its melody when the voice is silent, and the song's page documents how central that call-and-answer became to its identity. A carve can rescue a bar or two. It cannot rescue an arrangement that never leaves the vocal any room.
How do you test a mix on someone who does not speak the language?
At low level, on a small speaker, with a singback. I play the chorus at conversation volume for someone with no Ukrainian and ask for two things: tap the rhythm of the words, then sing the hook back on any syllables at all. Rhythm coming back means the consonants survived. The hook coming back means the vowels carried the melody. "Stefania" travels partly because its hook is a name, and names cross borders without translation.
The result suggests the method holds. The song took 631 points in Turin, and its 439 televote points were the highest total in the contest's history at that point, from a public that overwhelmingly does not speak the language. It then spent two weeks on the UK chart, peaking at number 38 according to the Official Charts Company, and went platinum in Poland. Two years later I composed "Teresa & Maria" for alyona alyona and Jerry Heil, Ukrainian lyric again, and it placed third at Eurovision 2024 with 453 points. None of those songs traded their language for reach.
What to do with this
Mixing a vocal in a language you do not speak? Ask the artist to mark the lines where the words matter most, open your de-esser's detection wider than the preset, and run the singback test before you print. My verified credits, including all four roles on "Stefania," are on the credits page, and the longer story is on the about page.