Devoiced Vowels in Japanese: Why Desu Sounds Like Des

What devoicing actually is

Every vowel you normally say is voiced: the vocal cords vibrate and you can feel a buzz if you touch your throat. A devoiced vowel is the same vowel with the buzz switched off. The tongue and lips still move into the i or u position, air still flows for the same length of time, but the result is a whisper. In the phonetic alphabet the whispered u of desu is written with a small ring under it, [ɯ̥], and that ring is the whole story.

This is why desu sounds like des to a beginner's ear. The s is followed by a whispered u that blends into the hiss, so all you consciously hear is a slightly long s. The same thing happens to the u in masu and to the i in shita, where the sh stretches into a breathy hiss before the t. Nothing has been deleted. The vowel is present, silent in voice, and full in time.

The two conditions that trigger it

Devoicing is not random. It has two triggers, and both need a voiceless consonant. The first trigger is a high vowel sandwiched between two voiceless consonants: the u in suki sits between s and k, the i in shita between sh and t, the u in kusa between k and s. The second trigger is a high vowel at the end of a word after a voiceless consonant, when a pause follows, which is what happens to desu and masu at the end of a sentence.

The consonants that count as voiceless are the ones you can whisper without any buzz: k, s, sh, t, ch, ts, h, f, and p. The voiced ones, such as g, z, d, b, m, n, r, w, and y, block devoicing. That is why the u in desu ne usually stays voiced, because n is voiced, while the u in desu ka is whispered, because k is not.

• Only i and u devoice; a, e, and o do not in ordinary speech.
• Between two voiceless consonants: s(u)ki, sh(i)ta, k(u)sa, ts(u)kue, h(i)to, k(i)ta.
• At the end before a pause, after a voiceless consonant: des(u), mas(u), ikimas(u).
• Before a voiced consonant or a vowel the high vowel stays voiced: sugu, kudasai, eki de.
• When two neighbouring moras both qualify, speakers usually whisper only one of them, so kiku is not whispered twice.
• Long vowels such as the ū in kūki are not devoiced.

Everyday words that devoice

The table lists the words where devoicing is most consistent in standard Tokyo speech. The whispered vowel is shown in brackets. Read each romaji column aloud, then say the word again with the bracketed vowel whispered and the beat kept. If a word has two brackets they are separated by a full mora, so both can be whispered without the word collapsing.

Notice hito. Because the i is whispered, the h in front of it is heard as a soft palatal breath, so the word sounds like a breathy hto. Learners who voice the i produce a clear hee-toh that native speakers recognise instantly as foreign, even though every sound is technically present. Words like hikōki and hitotsu follow the same pattern.

Pitch accent interacts with devoicing, and textbooks note that the accent drop tends to move off a whispered mora. You do not need to manage that consciously. Copy the whole word from a native recording and the accent will come with it.

Whispered, not deleted: the beat stays

The single most important idea in this guide is that devoicing removes voice, not time. Desu is two moras and stays two moras. When you whisper the u, the s is stretched to fill the second beat, so the word is de plus a full beat of hiss. Delete the beat and you get a clipped des that ends too soon; Japanese listeners hear that as abrupt, and in fast speech it can make masu ka sound like mas ka with the question particle crushed against it.

The rows below contrast endings that devoice with endings that do not. In desu ka and masu ka, the u sits between s and k, so it is whispered. In desu ne, desu yo, and masu yo, the next consonant is voiced, so the u is usually voiced too. Shadow each pair and feel the second beat stay the same length whether or not it carries voice.

Gakusei desu has two whispered vowels in one line, the u of gaku and the u of desu, and both keep their beats. Yoroshiku has a whispered i between sh and k, but the final u stays voiced because a vowel follows it in onegai. Shitsurei has a whispered i before ts, and shimasu ends with the familiar whispered su.

Naturalness, not correctness, and it varies by region

Devoicing is not a rule you can break. Every vowel fully voiced is perfectly intelligible Japanese, which is why teachers reading slowly, announcers speaking with special care, and singers holding notes all voice vowels that would be whispered in conversation. What devoicing gives you is naturalness: the difference between reading a sentence and saying one. It also makes your own listening easier, because you stop waiting for a u that native speakers never voice.

It also varies by place. Standard Tokyo Japanese devoices readily. Speakers from Osaka, Kyoto, and much of western Japan tend to voice these vowels more, and the habit shifts by city, generation, and speaking speed, so do not treat a voiced desu from a Kansai speaker as wrong. For a learner aiming at standard Japanese, the safe target is the Tokyo pattern in the everyday words above, with fully voiced vowels kept for slow, emphatic, or sung speech.

The robot mistake and the clipping mistake

The first learner mistake is to voice every u and i at full strength. Each desu becomes de-soo, each shita becomes shee-ta, and the speech takes on a careful, even, slightly mechanical quality that native listeners associate with a reading machine or a very early textbook lesson. The words are right; the texture is wrong. This habit usually comes from learning through romaji, where every letter looks equally important.

The second mistake is the overcorrection. Having heard that desu sounds like des, the learner deletes the vowel and its time together, producing clipped endings that stop dead. Sentences finish too early, masu ka becomes a crushed mask, and suki loses its first beat so it sounds like a foreign word. The fix for both mistakes is the same drill: keep the beat, remove the buzz. The rows below contain the words learners most often either over-voice or clip.

In sushi the first u is whispered between s and sh, while the final i stays voiced because ga follows. In mimashita and tsukaremashita the i of shi is whispered before t, and in tsukaremashita the first u is whispered as well. Say each row at a slow, even pace with a finger tapping the moras, and let the whispered beats be exactly as long as the voiced ones.

Devoicing inside sentences

Inside a sentence the same two triggers apply, but the surrounding word can change the answer. Kita on its own has a whispered i, because a pause follows the ta; but in kimashita the i of ki is followed by m, which is voiced, so it stays voiced and only the shi before ta is whispered. Eki alone can end in a whispered i before a pause, but eki de keeps the i voiced because d is voiced.

So the habit to build is listening to the next sound rather than memorising word lists. The rows below put the common words into ordinary sentences. Before you say each one, find the i and u sounds, check what follows them, and decide which to whisper. The framing after the rows gives the answers.

The whispered vowels are: the shi of kimashita and aimashita; the u of kusa; the i of ashita and shiken and the u of arimasu; the first u of kutsu, while the final u is voiced before o; the i of chikatetsu and the u of ikimasu; the first u of sukoshi; the u of futatsu and the u of tsu before kudasai; the i of hikōki and the shi of ikimashita; the u of shukudai and the shi of wasuremashita; the u of tasukete; and the shi of shimashita. Sentaku keeps its final u voiced because o follows it.

A drill that keeps the mora

This drill trains the whisper and the beat at the same time, which is the only way to avoid swapping the robot problem for the clipping problem. It uses tapping, because a finger will keep time honestly when your ear is busy. Start with sō desu, the most common sentence in Japanese, and do not move on until the tap and the whisper land together every time.

• Tap four slow beats and say sō desu on them: sō takes two taps, de one, su one. Voice everything.
• Repeat, and on the fourth tap switch the voice off but keep the mouth in the u shape and the s hissing for the full tap.
• Add ka. Now the su is between s and k, so it stays whispered, and ka takes a fifth tap with full voice.
• Swap ka for ne. Bring the voice back on su, because n is voiced, and feel that the beat length has not changed.
• Do the same four steps with suki desu, ashita ikimasu, and hito ga imasu, whispering only the vowels the table marks.
• Finally speed up gradually while still tapping. If a whispered beat starts disappearing under the next word, slow down until it returns.

How to check yourself

Record the four rows below and play them next to a native recording of similar sentences. Listen first for length: are your desu and masu endings the same duration as the native ones, or shorter? Then listen for texture: does your ending hiss, or does it carry a clear oo sound? A phone recording is enough to hear both. Correct one word at a time, re-record, and compare again rather than trying to fix a whole sentence at once.

Then take the same sentences to your AI Sensei in Unihongo's immersive 3D classroom, where the pronunciation coaching listens to complete sentences rather than isolated words. Ask the teacher to keep asking you questions that end in desu ka, so you have to produce both the whispered su before ka and the voiced su before ne in real replies, and read the corrections in the session report afterwards.

In the second line the u of samukatta stays voiced, because m is voiced, and only the final desu is whispered. In the third line gakusei and imashita each carry one whispered vowel, and the i of toki may be whispered before the comma or voiced, since word-final i before a pause is less consistent than u. Both are natural.