Vie's Japanese Pronunciation Study Log

Hello and welcome to my Japanese pronunciation study log. I will keep track of my progress and share my findings related to Japanese pronunciation here.

My starting point is a native Belgian French pronunciation and my end goal is a mix of 東京弁 and 博多弁. Realistically speaking, nearly all resources and content are in 標準語 so unless I decide to move to 福岡, that 博多弁 pronunciation will most likely never materialize. At least, that gives me more leeway, like affording to say typically 平板 verbs in a 中高 fashion because that’s what happens in 博多弁 (screw long 平板 verbs, I don’t like them).

For the longest time, I was deaf to pitch accent, scoring a miserable 50%-60% on minimal pair tests but a few months ago, I unlocked a new sense, and now I can finally reliably hit 95%+ accuracy (it only took two years of trying…). So I figured, now would be a good time to start my pronunciation training arc.

However, drilling phonetics can often be dull, repetitive, and demotivating, so I’ve decided to start a public study log to help keep myself accountable. There’s also a noticeable lack of resources on improving Japanese phonetics and prosody specifically from the perspective of a native French speaker—most comparisons tend to focus on English, Chinese, or Korean. In the spirit of “being the change you want to see in the world,” I hope this log can also contribute to the broader pool of knowledge on the topic.

Sorry, what? Are you saying a French person may not think about searching an English forum for tips on improving his/her accent in Japanese? Pff, don’t be ridiculous.

The current plan of action is to regularly chorus short segments from “Shadowing 日本語を話そう 初~中級編” in Audacity, go through the exercises in Aomi Speaking, and book corrected reading sessions with teachers on iTalki.

Off the top of my head, here are the main pain points I’d like to address:

  1. Pronunciation of Japanese words that I knew prior to learning Japanese (e.g. arigatō, konnichiwa, samurai, sushi, manga, …): these borrowed words have made it into French and I’ve been saying them in a naturalized way for so long that I can’t seem to shake off this fossilized pronunciation. In particular, I hate the fact that I’ve never managed to output a convincing ありがとう(ございます).
  2. ふ sounds: due to my background, I have to say words such as フランス語 and ソフトウェア very frequently but I do not have [ɸ] in my native language so it always comes out wrong.
  3. Stop raising the pitch and lengthening the vowel sounds at the end of utterances: this bad habit can probably be attributed to the intonation pattern of French. Honestly, I am not sure how I can fix this except occasionally reminding myself of this tendency.
  4. Get better at pronouncing quick successions of た行 and か行 sounds (e.g. 暖かくない、行きたくなかった、前から戦っとったと?、他)
  5. Learn to pronounce 平板 words without dropping pitch mid-way: so many 平板 words and yet I find it tiring and weird to keep a level high tone.

To give a better picture of my current level, I will post a recording of a 朗読 session later this week.

Hopefully, I’ll be able to upload a recording of the same book passage multiple months down the line and notice an improvement. :crossed_fingers:t2:

8 Likes

This is excellent and very timely for me in particular (I’m assuming that that last part is exactly why you’re doing this, right?)

My Japanese path has lately been highly skewed towards the skills that I want/need to master right now. And it turns out that I’ve started weekly conversation practice with a tutor and while I can definitely talk and be understood, I would like to work on my accent somewhat. My goal isn’t a native accent, but a “I like listening to your accent” and/or “your accent is so cute” territory.

I’ve been playing around with the pitch accent minimal pairs test and just never feel like I get it… do you have any suggestions, or do I just need to start banging my head against the wall with it as a part of my daily practice?

I also have the shadowing book which has sat on my shelf for probably about a decade now… I was thinking I might finally be able to keep up with it and might also add that into my weekly rotation. I’m also really interested in trying chorusing techniques with it.

Anyway best of luck! I’m very much going to be stealing both ideas and a bit of motivation hearing about your process!

I mean… this log is also in English so… :blush:

2 Likes

Do not hurry through the test, take your time to ponder every option and review your mistakes.

Shadow the recording then try to repeat the same thing by humming and then vocalize and/or hum each option back-to-back, comparing with the target audio. I personally did not hesitate to play the audio multiple times before answering.

Review your mistake by comparing both options back-to-back and really try to hear the difference. Sometimes, it will sound exactly the same to the point where you may wonder if it is not the same audio clip being inadvertently duplicated. If you really do not believe it, you can use the web page inspector dev tool, select the two buttons and confirm that the button link to different audio links or base64 attachments. If you really still can’t hear the difference after 30s of comparative listening, just move on and try to grab points from easier questions.

Do not mix in too many pitch accent patterns at once. I can recommend focusing only on the 平板 vs. 頭高 distinction at once and only once you can reliably achieve a satisfying score discriminating those two, grind 中高 exclusively, before finally testing all pitch patterns mixed with each other.

This may be a controversial take, but I’ve found that testing for pitch accent on my Anki cards (i.e. I must remember the reading, the meaning and the pitch accent of the word to mark it “Correct”) helped to send a signal to my brain that this is an important factor. When discussing this, what always comes up in my mind is an instance of a Japanese learner complaining that the word-reading quiz that can be seen in some Japanese learning circles is too strict because it does not accept, for example, りょこ for 旅行, which was super annoying for him because he never paid much attention to short vs long vowels in his Anki reviews since one could think “whatever, it sounds pretty much the same and people will understand me anyway so why bother remembering such minute details?”. Doesn’t that line of thinking sound familiar? Having said that, I no longer test for pitch accent in my Anki reviews because that really kills my accuracy lol.

Listen to dialectal speech, especially those that are not 東京式アクセント like 関西弁 and prick up your ears whenever you hear a native say え、イントネーションおかしくない? in chat or in comments, about the way another native pronounces a certain word.

5 Likes

Nonsense. This log was clearly created specifically for me, a French person looking for tips on Japanese pronunciation a week after starting to learn Japanese, on an English forum.

Already a very useful resource, thanks for the link. I’ve been pronuncing ふ as a French [fu], sometimes an English [hu], but now I have more examples in Spanish and Korean.

3 Likes

Here is the promised audio recording:

This time, I decided to go back to my roots and read one of the tadoku books: 丸メガネ寮母さん | L12

Finding a book with an associated voice recording that does not exceed 3 mins and is composed of fully connected sentences turned out to be harder than I thought but in the end I did find exactly what I was looking for.

At first, I was thinking “wouldn’t this be too easy?” but I had forgotten how difficult it is to record one take in a foreign language without tripping up at least once somewhere.

Parts that tripped me up:

  1. 羽田空港 : At the beginning, I was saying はねだくうこう ̄ but that felt hard and weird and made me double-take what I was saying multiple times but after listening back the tadoku recording more intently, I realized it is actually pronounced はねだく\うこう (which I confirmed with youglish and コツ search as well). After adjusting my intonation, the word and the things that come after it became so much easier to say, wow.

  2. 空が : I should be saying そ\らが. It is not particularly difficult to pronounce but it is wild how much variance in pitch patterns I had on this word alone from one take to another.

  3. しとしとと : the direct juxtaposition of しと’s と and the “quotation” と (for lack of a better word) probably confused my brain a little.

  4. 国と : I kept hesitating on the pitch accent of the と particle which bugged me a little

  5. 来てくれていました : I kept saying 来てくれました instead of 〜ていました. That makes me realize that I’ve (pretty much) never used the 〜て form of くれる in my speech. I always say 〇〇てくれました, not 〇〇てくれていました.

  6. 真剣に心配そうな(丸メガネ) : For some reason, I struggle to say 心配 after 真剣に. I also had to fight with the urges of my mouth and tongue to say そうの instead of そうな. Why am I so tempted to say の here?

3 Likes

This got me to try that. Took me a bit before I understood the notation, and the fact that I’m not used to hearing stuff in that vocal range made it a bit trickier for me (the majority of my input is female speakers). It was interesting. Does the test ever end? It felt like it was endless, and had no progress indicator - which made it very tedious

I just have jpdb auto-play audio, when I flip the card. It definitely makes a difference vs just doing it silently in Anki. And I’ll shadow it, if it’s an unfamiliar word, or I’m having trouble. Idk if I could necessarily identify pitch accent reliably, but I can imitate/shadow it just fine. It also lets you play multiple patterns for many words. Probly wouldn’t hurt me to learn the names/patterns more formally though

1 Like

Now that I’ve been paying more attention I can definitely hear the difference between the two pairs. I just can’t tell you where the drop is. Maybe the better move is for me to learn the pitches of a few dozen words of each to really start noticing where the pitch dips and then move to the pairs testing? :thinking: I started paying attention to the pitch chart on my jpdb cards yesterday… that’s probably a good first start at least…

1 Like

I think the idea that there is always a hard drop is just a first approximation. E.g. in the word 静かに it’s more like the pitch starts high and then goes down in three steps until it’s completely down when you reach the に.

There is also the difficulty that pitch can change for different forms of e.g. an adjective, like 高い vs 高く, or 嬉しい vs 嬉しく. Like for these adjectives pitch goes down before the い, while for the adverbs it starts to go down one syllable before: たか\い うれし\い → た\かく うれ\しく.

I am not sure how to learn something like the latter. There are probably some rules one can read about, but getting it by hearing many people say a word is another, more intuitive way, where one might sooner or later find the rules by oneself.

2 Likes

I’m definitely of the mind that if children can learn it, so can adults. A lot of the idea of losing the neuroplasticity for languages has been debunked. (I have a pet theory that a lot of accent acquisition comes from both adults gently correcting pronunciation, as well as peers and sometimes adults as well, making fun of pronunciation. Since no one is going to do this for an adult learner, this is part of what is missing that you will need to find a way to replace). That said, you can’t learn a thing that you can’t notice, so the first thing is to learn how to notice. I think that’s the stage that I’m in right now…

4 Likes

Wait until you see the wacky inconsistent notation of Japanese natives… They tend to rather use up and down arrows to describe the absolute pitch level of the mora preceding the arrow (e.g. 確かに=た\しかに=高低低低=HLLL=た↑し↓か↓に↓) but most of the time they can’t be bothered to notate every mora so they omit arrows and it quickly becomes confusing. Especially when they add more symbols into the mix like 〜, :arrow_heading_up:, and →.

For example: ダッ↓クス↑フンド→ in

Even before the video starts, the thumbnail alone already makes me go “what?”

It is indeed endless, although there is of course a finite amount of audio samples. Personally, I always reset my run after 100 and I was repeating until I could get a satisfying score on a full run. Resetting early when I notice there would be absolutely no way to reach my desired score just like video game speed runners lol.

I’ve always had auto-play audio on all my Anki cards but that did not prevent me from struggling to recall the pitch accent pattern of the words, unfortunately.

Hrm… not so sure about that. There is a real chance that falling into this rabbit-hole will hurt your mind and your self-confidence. Tread with caution.

3 Likes

Since this thread spawned so much interest in pitch accent minimal pairs test, feel free to track your progress in this shared Google Sheet. It was put together by Yuri (@ganqqwerty on Discord), who’s gathering all the data. I would have personally loved to join the data collection effort but Yuri shared this shortly after I finally hit 95%+ so I come in too late.

3 Likes

What are they using the data for?

To quote his own words: “I want to gather a dataset of several hundreds of people and then calculate some stats. For example, the average time needed to learn to hear pitch.”

It is also interesting to see what difficulties others encounter and compare with your own experience. Personally, I also think that correlating the data with the background of the person could be interesting: does musical training help? does knowing another pitch-accent language (e.g. Swedish) help? does knowing a tone language (e.g. Cantonese) help? do Japanese beginners take more or less time than people who’ve already learnt a fair deal of Japanese?

In passing, I’ve got to say that a free-form collaborative Google sheet where everyone fills in the data in their own eccentric format, conventions and colors is probably not scalable to “several hundreds of people” unless you really like data cleaning, but fine, I don’t want to disparage the initiative.

2 Likes

I don’t really see the need for that dichotomy/comparison. There’s examples of adults learning it - thus we know adults can.

There are teachers who will do this btw - sometimes as their main focus, even

Thank you for telling me. That is really aggravating UX. It would be so simple to have a “how many samples?” field, or at least write that it’s endless somewhere (sry, pet peeve)

Also minor complaint, but I wish it would (optionally) show you the name of the pattern after you get it wrong. You can of course work it out from the stats, but I think it would be helpful reinforcement to have it explicitly shown

Ok I swear the first time I did the test, it gave me a single speaker, but this time it gave me like 3 different ones, in bass, tenor, and alto(?) ranges. Much nicer

:cold_sweat: :rofl:

Also isn’t pitch accent why V-tubers sounds kinda weird in the first place?

Absolutely the way to go :100: :joy:

Recall = identify & remember the name, when testing yourself? Or reproduce/recognize the word accurately? Ex: I have no clue what the patterns or names for 結構 and 決行 are, but when I heard the latter in an anime for the first time, I could immediately tell (by sound) that it wasn’t the former. (Turns out 結構 has multiple patterns, while 決行 only has け↗️っこう). Not claiming I can always do this reliably, or anything

Anyway your point is taken. How did you get the audio in Anki btw?

Very low in my case (I’m the right combo of stubborn & dismissive), but thx for the warning. I really just mean learning the basic names anyway

2 Likes

I feel like there has to be more to it but at the same time I don’t want to force myself to listen to their annoying voice to dig deeper.

Given a word on the front of a card, recall what mora has the accent, or in other words, recall where the downstep occurs.

I get word audio from the local Audio Server for Yomichan add-on and context/sentence audio is automatically added to my cards by Memento. In case I mine something outside of Memento, I use Anki Media Extractor Script to quickly record an audio clip and attach it to my latest created card.

1 Like

Practice time report

Time spent on Aomi from 7月11日 to 7月19日 : 205min (3h25min)
Time spent on Audacity from 7月11日 to 7月19日 : 144min (1h22min)


Tangentially, it is funny how I came across an unexpected mention of shadowing in an anime I was watching. It is following me everywhere I go.

Unfortunately, unlike him, I am not at the ぺらぺら stage yet though :cry:

2 Likes

音読タイム!

While I am here, allow me to share the process I used to produce the color-coded transcript you saw earlier.

  1. (Optional) Manually transcribe text from https://www.youtube.com/watch?v=fr9D5GSvYjs

  2. Paste the text into コツ

  3. Generate annotated text with underline pitch

  4. Open web developer tools

  5. Find the div/span container of the generated annotated text (something like <div class="px-3 overflow-auto visual-type-showPitchAccent ruby-type-common">)

  6. Copy the HTML code into a file

  7. Pass the file as input to this Python script

  8. Copy and paste the output of the script at the bottom of this XeLaTeX template

  9. Fix issues (コツ isn’t perfect) and eventually add furigana with \ruby{kanji}{furigana}. For example: \kihukuuline{\ruby{思}{おも}い\ruby{返}{か\textbackslash え}せば}
    ihukuuline{uby{思}{おも}いuby{返}{かextbackslash え}せば}

  10. Compile the transcript with XeLaTex

  11. ???

  12. Enjoy

2 Likes

Practice time report

Time spent on Aomi from 7月20日 to 7月26日 : 138min (2h18min)
Time spent on Audacity from 7月20日 to 7月26日 : 29min


Less than I would have liked… Although I have to say that I’ve now started reading manga, novels, etc. out loud now but that’s annoying to track so I can’t say how long exactly I’ve practiced.

A recurring pain point that’s been occurring lately when I practice on Aomi are 頭高 words that start with a voiced consonant. 頭高 is the pitch pattern I am the most likely to recognize in speech but it is ironically one that I struggle to reproduce in my own speech.

I’d often get stuck on one card with scores below 80% but as soon as I replace the 頭高 word by a placeholder あ\ああ…, my score suddenly shoots up. For instance, I really struggled with ご\くごく and も\っと. However, も\う has never been a problem :thinking:

Isn’t it strange that I find it less difficult to pronounce 頭高 words with unvoiced consonants like す\やすや? I’m maybe not used to forcing my vocal chords to vibrate this much right from the start of a word.

3 Likes

I recently started listening to the audiobook of 後悔しない時間の使い方 | L30 and it is wild how incredibly robotic the voice of the narrator 吉岡 琳吾 is. It is quite troubling because it legitimately made me wonder whether this was Text-to-Speech and yet it is not grating to the ears like speech engines and vocaloids can be. All his phonemes sound consistently the same, it is impressive.

Audible preview

3 Likes

I had an audiobook the other day that I was also suspicious about it being text to speech or not… it sounded a lot like that preview :joy:

1 Like