Am I doing pitch accent test (コツ) wrong?

I’ve never had much trouble with native speakers understanding my speech, but I know it’s probably not perfect so I started using the minimal pairs pitch accent test to see if I could hear the pitch. I used to play piano so I know I’m not completely tone deaf, but I’ve run into some frustrating minimal pairs that make me question whether I’m even doing the test right.
For example, on this word シケー, I heard the pitch go down at the end, which is why I chose シケ\ー, but apparently the pitch drop either wasn’t large enough or early enough to be correct. Listening to the シケ\ー clip, they made the drop super obvious, but the drops are not always that obvious so I get tripped up.

I forgot to screenshot it, but there was even a word トッカ where the answer was ト\ッカ, but the graph actually went up after the glottal stop so I got it wrong.
I will say that about half of the words are clear as day for me, and the few Youtube videos I’ve seen on pitch accent have all used really clear examples so they haven’t really been helpful for gauging if I’m mishearing or not.
So as a complete pitch accent noob, am I doing these tests wrong or am I just bad right now?

Side note

My coworkers actually got several questions wrong, but I live in rural Hokkaido and they probably have preconceived notions of which words are being said which influenced their answers.


Not really sure what changed tbh since it’s only been 2 hours and my score went up 20 points.
Maybe I just get tilted if I get more than 2 wrong in a row :thinking:

1 Like

Is there any way for you to download the audio and attach them here (or otherwise link to them)? It’s kind of hard to comment without hearing the audio.

1 Like

Unfortunately, I’ve already closed the tab so I don’t think I can get it back. I think my main question isn’t about that one specific example, but just as a whole whether I’m approaching the test incorrectly. I can hear pitch going up or down but if it’s not enough to constitute a “drop” then I often get the question wrong.
The website I’ve been using is コツ if that helps though :sweat_smile:

Can’t access the site without an account unfortunately (maybe I’ll make an account later, but no time right now). My general comment would be that with enough listening practice (and clean audio) single-word pitch accent is usually pretty obvious. So it’s possible you need more practice. It’s also possible that the audio on the site is unclear or wrong. Does the audio sound like a real person or computer generated? Because in my experience computer generated audio can be a bit wonky.

Also, does it ever show you the kanji for words? I ask because theしけい example is kind of weird. From a quick dictionary search, I think 死刑 is probably the most common word with that reading, and its pitch accent is シケ\ー. Now, I suppose the site could just be giving two pitch accents for the reading without any consideration for real words, but that approach would at least make me question whether the site is that good or reliable.

All that is of course without me trying the site, so take any speculation or criticism with a grain of salt.


It doesn’t even ask for an email so as far as I remember the account creation takes like 5 seconds. The audio sounds like real people and is generally pretty clear. I don’t doubt that I need more practice, since I’ve only just started. But for now “fake drops” are a bit frustrating.

It never shows the kanji for the words, which is a bit annoying but understandable since the point is that you can correctly hear the pitch and not just remember which words have what pattern. When I think of common words I know well, I’m pretty sure I know the pitch accent, so maybe when they use a homonym with a different pitch accent my brain sort of doesn’t want to listen to the unfamiliar word. It’s also so regional that memorizing specific pitch accents doesn’t seem too productive to me.

As far as whether the site is reliable, I actually couldn’t find any other site with pitch accent practice tests so it’s kinda all I got atm haha

1 Like

I can’t vouch for it, but someone posted this recently:


Japanese people will get plenty wrong when they start because they’ve never thought about the idea of an accent being on a mora probably. My girlfriend did pretty bad when I showed it to her, but she was quickly able to calibrate her brain to thinking about where the drop was. Japanese people’s typical mode of thinking about pitch seems to just be 上がる or 下がる, which isn’t even technically correct since 平板 is not only “flat”, but actually slightly goes down over the course of the words typically. But, it’s easy to see where they’re coming from.

Pitch fluctuates in speech for a number of reasons. Noticing what is pitch accent and what is something else is part of being able to “hear” pitch accent. One guy I know who is really good at pitch said something along the lines of “anyone can hear pitch, but you might not realize it’s the accent of the word and just be attributing the pitch to something else like the way the person says the word, their intonation, or something else”.

All this to say, I don’t think you’re doing the test wrong at all. It’s just a natural part of sucking. The more natural speech is, the harder it will be to hear out the pitch accent apart from the other parts of their speech. But being able to separate them is a skill you can and will need to hone.

EDIT: I got しけい in the test just now and the flat one was distinctly heiban to me. Keep practicing and it will get easier.


I just tried it and got 88% over 43 words. Some mistakes were definitely of the “tricked myself” variety. Others I can’t really hear why my answer was wrong even after listening several times. I wonder where they get their audio from. One was definitely computer generated (I know the specific TTS voice from Kitsun) and a couple sounded like they could be from forvo, but overall they were pretty clear.

I’ll probably try this again later. I didn’t realize there was a timer at first, so next time I think I’ll try a speed version and only listen to each audio once.


It’s worth noting I got one, ゆきたけ where it was pronounced [3] but the answer was [2]. When you click to hear the audio they were both the same exact clip.

Its a thing in pitch accent where when the accent would fall on a devoiced mora, it slides down one to the right. The site doesn’t seem to realize that the audio it’s using for [2] underwent that shift and became [3].

I would be careful of any other instances of that on the site. Kotsu didn’t really have those iirc (maybe once?).

1 Like

I’ve noticed heiban is definitely where I get the majority of my errors, presumably because I get baited by the slight decline in pitch. Would intonation even come into play when it comes to isolated words? I always thought of intonation as how the pitch fluctuates in a sentence.
I think I do much better when I stop to think what pitch pattern the word falls into rather than just where the drop is but maybe that’s just a crutch because I’m new to paying attention to pitch.

That’s possible. Heiban was always the easiest to me, but I couldn’t hear the decline in pitch at all really. Maybe your sensitivity to pitch is giving you more information, but that also leaves you with more that you need to sort out.

Intonation can be present for single words I think, assuming I’m not messing up the definition. The questions only quiz is an example of that. They are all single words but have that rising intonation (if that’s the right word) we use for questions. It’s pretty exaggerated, but once you can separate out pitch you should be able to get those questions right too despite the fact that the intonation is rising.

Id say just keep struggling with it and keep upping the level once you get to ~100%


If it makes you feel better, I am about 62% on the kotsu minimal pairs test (having not practiced at all). I can hear the difference between the two options when listening to both at once but my ability to tell pitch pattern from hearing just one is awful.

1 Like

I know you don’t really output, but do you listen to japanese regularly?

1 Like

Not super often, no. I guess I listen to something in Japanese at least once a week but way less often than I do reading. At the moment I’m reading the Fatal Twelve VN so that’s an hour or two a week, and I have a podcast or two I listen to sometimes. It’s definitely not part of my routine though. I did a bit more when I was participating in the the “listen every day” thread.

I remember back when I was about N3 level and studying in Japan that I could often not tell minimal pairs apart even when hearing the two words back to back, so in that sense I’ve improved a bit :slight_smile:

One of the words I got wrong was like まちがえる (don’t remember the exact word). Basically the kind where technically it’s [4] but often it sounds like [3] (to me at least, and I think for some words both have become officially recognized). I heard [3] as I often do for these kinds of words but answered [4] because I knew that was the technically correct answer most of the time. But they actually were expecting [3] (hence tricked myself). When I got this word I didn’t realize you could play both versions afterward, so I’m not sure if they used different audio clips or not.

1 Like

I honestly didn’t even know this existed. I just did 20 of the minimal pairs and got one of them wrong, a heiban one to be specific. I think I understand what you’re struggling with.

In the recording of the heiban word I got wrong, the pitch actually did go down. Which I believe even native speakers do to some extent when they get lazy/casual with their speech, but I didn’t expect that in a test on pitch accent. After my miss, I paid attention to the recording of heiban words afterwards. Some stay rather flat, others did actually go down. After making that realization I could adjust and tell which ones where heiban anyway, because the ones that do go down in pitch do so only slightly. If it’s atamadaka, it will go down more noticeably. Perhaps paying attention to that could help you out.


I know this is just a graph, but it won’t let me add the sound clip.

This is supposedly 平板?
The two audio clips sound quiet different but 10 times out of 10 I would get this one wrong.

Edit: Couple of rounds later:

I know I’m just bad at this right now but the pitch drops are killing me

Very cool :upside_down_face:


I am one of aneninja devs. Thank you for the feedback, sorry for the late reply. I’ve just seen it and wanted to clarify a few things.

All audio comes from nhk 2016ント辞典, in which there are two different possible pronunciations for ゆきたけ (same as in 1998 edition). We used two different clips, but I agree it is hard to spot the difference.

Here is ゆきたけ [2] (pitch graph using praat)

And here is ゆきたけ [3]

When it comes to dealing with accented devoiced mora, it is either like you mentioned shifting the accent, but also it is possible to leave the accent as it is. Devoiced mora doesn’t have it’s own pitch, but it can derived from neighboring moras (like in やきたけ [2] following mora (た)has more drastic pitch fall). In the past accent shift was the main way of dealing with it, but over the years leaving the accent has gained on popularity. It is quite complex and interesting topic, to read more about it I recommend articles from 北原真 or 杉藤美代子.

Overall though you are right that it can be really hard to differentiate those two and also both strategies are correct, so we implemented an option to filter devoiced accents.

Btw in kotu, you can encounter them as well (I’ve seen 父, 吹く There is even an option to test only words with devoiced moras.


Are you saying these are the two audio clips that would have played on your site when I clicked the respective pitches? Because they didn’t.

Those two are blatantly different clips, but the ones on the site were the exact same when I used it. I don’t just mean that they were the same pitch, but that it was literally the same exact audio clip being played regardless of what pitch pattern I clicked as if they pointed to the same file.

Im not sure if it was some bug or something that has since been fixed, but only one of those played for me when I got ゆきたけ. It’s been awhile so I can’t say which one though

Yeah, you’re right, but I just loaded up and did some and I could hear them perfectly fine. The fact that there was some clear audio file bug with uroninja on the one problem I got wrong when I did it really just kinda seemed sus to me…