Inkuntri
Japanese Research, tools & pedagogy

Japanese Pronunciation Self-Diagnosis With Recording and Pitch Models

The reader can self-diagnose Japanese pronunciation using recordings, pitch-accent models, mora timing, vowel devoicing, intonation, and targeted correction.

Published March 27, 2026 Japanese

Core examples: ピッチアクセント, モーラ, 長音, 促音, 撥音, 母音無声化, イントネーション, 頭高型, 平板型, シャドーイング, 録音, 発音.

“My accent is bad” is not a diagnosis

A learner records themselves and thinks:

I sound foreign.

That may be true, but it is not useful. What is the actual issue?

  • pitch accent,
  • mora timing,
  • long vowels,
  • geminate consonants,
  • nasal sound,
  • vowel devoicing,
  • sentence intonation,
  • rhythm,
  • over-English stress,
  • unclear consonants,
  • unnatural pauses?

Pronunciation improves when errors have names.

The key principle is:

Diagnose one pronunciation feature at a time.

録音

録音

recording.

Recording is non-negotiable for self-diagnosis. You cannot reliably judge your pronunciation only while speaking.

Record:

  1. isolated words,
  2. minimal pairs,
  3. short sentences,
  4. shadowing,
  5. spontaneous speech,
  6. the same text after correction.

Learner action: keep recordings dated so progress is visible.

発音

発音

pronunciation.

Pronunciation includes segment sounds, rhythm, pitch, and intonation. It is not only “making the right vowels and consonants.”

Learner action: divide 発音 into components.

モーラ

モーラ

mora, Japanese timing unit.

Examples:

か one mora

かあ two morae

かった three morae: か・っ・た

Japanese rhythm is mora-based, not English stress-timed.

Learner action: clap or tap morae to diagnose timing.

長音

長音

long vowel.

Examples:

おばさん aunt/middle-aged woman

おばあさん grandmother/elderly woman

ここ here

高校 こうこう, high school

Long vowels change word identity.

Learner action: record minimal pairs with long vowels.

促音

促音

geminate consonant, represented by small っ.

Examples:

きて come

きって stamp/cut, depending kanji/context

さか slope

さっか writer

促音 creates a hold/closure before the consonant.

Learner action: do not pronounce small っ as a separate vowel.

撥音

撥音

the nasal ん.

It changes sound depending following sound:

さんぽ nasal before p/b/m

さんか nasal before k/g

さん final nasal

Learner action: record ん before different consonants.

ピッチアクセント

ピッチアクセント

pitch accent.

Japanese pitch accent uses high/low pitch patterns and downstep. It is not English stress.

Examples:

雨 rain

飴 candy

Pitch may distinguish words, but even when it does not, it affects naturalness.

Learner action: learn pitch for high-frequency words and phrases you say often.

頭高型 and 平板型

頭高型

initial-high pattern.

平板型

flat/unaccented pattern.

These are pitch accent categories in standard Japanese descriptions.

Example idea:

  • 頭高型: pitch drops after first mora.
  • 平板型: no downstep within word, and particle may remain high depending pattern.

Learner action: use a pitch model or dictionary, not guesswork.

母音無声化

母音無声化

vowel devoicing.

In standard Japanese, /i/ and /u/ can become devoiced between voiceless consonants or in certain positions.

Examples:

です often sounds close to desu with weakened/devoiced u

した the i may be devoiced in many contexts

Learner action: do not force every vowel equally if aiming for natural speech.

イントネーション

イントネーション

intonation.

Sentence intonation conveys:

  • question,
  • continuation,
  • surprise,
  • politeness,
  • emphasis,
  • uncertainty,
  • emotion.

Pitch accent is word-level; intonation is phrase/sentence-level.

Learner action: practice whole phrases, not only words.

シャドーイング

シャドーイング

shadowing.

Useful for:

  • rhythm,
  • pauses,
  • intonation,
  • speed,
  • linking,
  • breathing,
  • formulaic phrases.

But shadowing without feedback can copy poorly or hide errors.

Learner action: compare your recording to model audio, not just repeat.

Diagnostic protocol

Use a fixed routine:

  1. Word list: long vowels, っ, ん, pitch pairs.
  2. Sentence list: common phrases.
  3. Model audio: native speaker or reliable source.
  4. Record yourself.
  5. Listen immediately.
  6. Listen next day.
  7. Compare one feature only.
  8. Re-record after correction.

Error categories

CategorySymptom
long voweltoo short or missing
geminateno hold for っ
mora rhythmEnglish stress rhythm
pitchwrong downstep
intonationunnatural sentence melody
vowel devoicingover-pronounced vowels
consonantsr, ts, f, h issues
nasalん too uniform or unclear
speedtoo fast before accurate
phrasingunnatural pauses

Minimal practice sets

Long vowel:

おばさん / おばあさん ここ / 高校 びる / ビール

促音:

きて / きって さか / 作家 いて / 行って

Pitch:

雨 / 飴 橋 / 箸 花 / 鼻

Mora timing:

きょう きよう きょ

Example bank walkthrough

ピッチアクセント

Pitch accent.

Learner action: high/low pattern.

モーラ

Mora.

Learner action: timing unit.

長音

Long vowel.

Learner action: length contrast.

促音

Geminate consonant.

Learner action: small っ hold.

撥音

Nasal ん.

Learner action: context-sensitive nasal.

母音無声化

Vowel devoicing.

Learner action: natural weakening.

イントネーション

Intonation.

Learner action: sentence melody.

頭高型

Initial-high pitch pattern.

Learner action: downstep after first mora.

平板型

Flat/unaccented pattern.

Learner action: no word-internal downstep.

シャドーイング

Shadowing.

Learner action: rhythm imitation.

録音

Recording.

Learner action: self-diagnosis evidence.

発音

Pronunciation.

Learner action: full sound system.

Pronunciation self-diagnosis workflow

  1. Choose one target feature.
  2. Find model audio.
  3. Record 5–10 words.
  4. Record 2–3 short sentences.
  5. Compare only that feature.
  6. Mark error category.
  7. Practice slowly.
  8. Re-record.
  9. Compare with old version.
  10. Move to next feature only after improvement.

Pronunciation target table

Name the problem before fixing it.

TargetDiagnostic question
モーラare timing units even?
長音are long vowels long enough?
促音is small っ held?
撥音does ん adapt naturally?
ピッチアクセントis downstep correct?
母音無声化are i/u overpronounced?
イントネーションdoes sentence melody fit stance?
consonantsare r, ts, f, h clear?
phrasingare pauses natural?
speedis speed hiding inaccuracy?

“Accent” is too vague to train.

Recording protocol

Use three recordings:

  1. model audio,
  2. your first attempt,
  3. corrected attempt.

Keep the same word/sentence set for several weeks. Improvement is easiest to hear when the material is constant.

Pitch model caution

Pitch dictionaries and models are tools, not identity rules for every speaker, region, or style. Learn standard patterns for clarity, but do not use pitch study to shame regional variation or natural speaker diversity.

A strong tool for this article would track recordings and corrections.

Suggested functions:

  1. Recording slots.
  2. Mora timing boxes.
  3. Pitch sketch field.
  4. Minimal-pair practice.
  5. Error category tags.
  6. Before/after comparison.
  7. Weekly target tracker.

Final rule

Japanese pronunciation improves when it becomes specific.

ピッチアクセント is not モーラ. 長音 is not 促音. 撥音 is not one fixed English n. 母音無声化 affects naturalness. イントネーション carries stance. シャドーイング helps only if 録音 and comparison guide it.

Stop saying “my accent is bad.” Name the feature. Record it. Fix it.

Related reading