Voicing Assimilation and Final Devoicing

Czech spelling is famously phonemic — you mostly say what you see. But there is one big, systematic exception, and it trips up nearly every English speaker: voicing. A consonant written as voiced (like d, b, z) is often pronounced voiceless, and a voiceless one (like s, k, t) is often pronounced voiced, depending on what's next to it. The spelling stays etymological — you keep writing led "ice" — but you say it with a final t-sound. There are exactly two rules behind this, and once you have them, a whole layer of "why does it sound like that?" disappears. This page explains final devoicing and regressive assimilation, the special behaviour of ř, and how the little prepositions v, k, s, z assimilate to the word after them.

Throughout, phonetic values are shown in square brackets in the explanations. (Note: the example sentences are real, fully spelled Czech — the bracketed pronunciations live only in the prose, never in the spelling.)

First, what "voicing" means

Consonants come in voiced/voiceless pairs that differ only in whether the vocal cords vibrate. The Czech obstruent pairs are:

Both rules below work by swapping a consonant for its partner in this table. Sonorants (m, n, r, l, j) and ř are special cases handled at the end. Note the pair h / ch: the voiced partner of ch is h, which feels counterintuitive but matters for assimilation.

Rule 1: Final devoicing

A voiced obstruent at the end of a word is pronounced as its voiceless partner. The vocal cords switch off before the silence. This is the same process German has (Hund sounds like "Hunt"), and it is completely regular in Czech.

So led → final d devoices to [t]; dub → final b devoices to [p]; nůž → final ž devoices to [š]; hrad "castle" → [hrat]; muž "man" → [muš]; vůz "cart/car" → [vús]. You never write any of this — the spelling stays led, dub, nůž, hrad — but you pronounce the voiceless partner.

Rule 2: Regressive assimilation in clusters

When obstruents pile up in a cluster, they must all agree in voicing — and it's the last consonant of the group that decides for everyone. The voicing spreads backwards (regressively), which is why a learner who pronounces each letter left-to-right gets it wrong.

If the cluster ends in a voiced obstruent, earlier ones become voiced:

In sbírat, the s sits before voiced b, so it assimilates to In kdy, the k sits before voiced d, so it becomes Same with kde "where" → [gde].

If the cluster ends in a voiceless obstruent, earlier voiced ones devoice:

Wait — prosba actually goes the other way: the s precedes voiced b, so it voices to That's the voiced-trigger version again. For the devoicing direction, take těžký: the ž sits before voiceless k, so it devoices to Likewise vchod "entrance": the v precedes voiceless ch, so it devoices to (and the final d devoices too).

The special case of v

The letter v is a half-exception. It undergoes assimilation (it devoices to [f] before a voiceless consonant, as in vchod [fchot] and všechno [fšechno]), but it does not trigger voicing in a preceding consonant the way other voiced obstruents do. So in a sequence like tvůj "your," the t before v stays voiceless [tvůj], it does not become [dvůj]. V is "transparent" as a trigger but obeys the rule as a target.

The ř and its voicing

The famous Czech letter ř comes in two flavours: a voiced [ř] and a voiceless one, and which you get follows the same assimilation logic. Next to a voiceless consonant, or at the end of a word, ř devoices.

In při, tři, přijít the ř follows a voiceless p or t and is itself voiceless; in řeka, moře, dveře it is voiced. You write the same ř in every case — the spelling never marks the difference — but your tongue produces the voiceless version after voiceless consonants and at word end. For how to actually make the sound, see the ř háček page.

Voicing across word boundaries

The rules don't stop at the edge of a word. In connected speech, the last consonant of one word affects the previous one across the gap, and a word-final obstruent assimilates to the start of the next word. This is why careful spelling-by-spelling pronunciation always sounds robotic.

In vlak má, the final k of vlak comes before voiced m, so it voices: [vlag má]. Across a word boundary, a final voiced obstruent before a following voiceless one (or before a pause) devoices, and a final voiceless one before a following voiced obstruent voices.

The prepositions v, k, s, z

The one-letter (and short) prepositions are the most visible place this happens, because they fuse phonetically with the following word. They take their voicing from the first sound of the next word.

So v Praze is [f praze] (the v devoices before voiceless p), but v domě is [v domje] (the v stays voiced before voiced d). The preposition k "to" voices to [g] before voiced sounds (k doktorovi [g doktorovi]); s "with" voices to [z] before voiced sounds; z "from" devoices to [s] before voiceless sounds. These prepositions also have separate vocalized forms (ve, ke, se, ze) that appear before clusters — a different topic, covered on the vocalized prepositions page.

Common mistakes

The mistake here is purely phonetic, so the example shows the correct spelling and the note tells you the trap. English speakers make these errors:

• Pronouncing final voiced letters literally. Saying led with a clear [d], dub with [b], muž with [ž]. Final devoicing means [let], [dup], [muš].
• Reading clusters left-to-right. Saying sbírat with a voiceless [s] instead of assimilated [z], or kde with [k] instead of [g]. The cluster's last consonant decides.
• Keeping the preposition's "dictionary" voicing. Saying v Praze with a voiced [v] instead of [f], or k doktorovi with [k] instead of [g].
• Over-correcting the spelling. Writing what you hear (let, zbírat, gde) instead of the etymological led, sbírat, kde. The sound changes; the spelling never does.
• Voicing v as a trigger. Pronouncing tvůj as [dvůj]. V devoices itself but does not voice a neighbour.

Key takeaways

• Spelling stays etymological (led, sbírat, kde, vchod); pronunciation follows the rules.
• Final devoicing: word-final voiced obstruents go voiceless — led [let], dub [dup], nůž [nuš], hrad [hrat].
• Regressive assimilation: in a cluster, the last obstruent sets the voicing for all — sbírat [zbírat], kde [gde], těžký [těškí], vchod [fchot].
• ř has voiced and voiceless variants chosen by the same logic; v assimilates but doesn't trigger.
• The rules cross word boundaries, and the prepositions v, k, s, z take their voicing from the next word: v Praze [f praze], k doktorovi [g doktorovi].

For the wider system, see the phonemic reading rules, the sibilants s / š / z / ž, and the errors English speakers make.