Pitch accent: the beat is the unit

Pitch in Japanese is not intonation — it belongs to the word. The unit is the mora: each mora is H or L, and the H/L string is the accent pattern. Eight words measured mora by mora.

Eight words

Click a cell, then look at the curve.

or use the + on each cell

In a sentence

How the word sounds in a sentence. Tap the line to hear it all, tap the highlighted word to hear just that part.

Pitch curve, mora by mora

The vertical axis is relative height inside the word (not absolute Hz); letters are the measured H/L. Absolute values are in the table below.

ねこnekoatamadaka (mora 1 high)
HねLこ
cat · HL · median 223 Hz
えきekiatamadaka (mora 1 high)
HえLき
station · HL · median 224 Hz
やまyamaheiban (no accent)
LやHま
mountain · LH · median 187 Hz
かわkawaheiban (no accent)
LかHわ
river · LH · median 248 Hz
はなhanaheiban (no accent)
LはHな
flower · LH · median 186 Hz
さくらsakuraheiban (no accent)
LさHくHら
cherry blossom · LHH · median 247 Hz
おかしokashiheiban (no accent)
LおHかHし
sweets · LHH · median 247 Hz
あなたanatanakadaka (accent inside)
LあHなLた
you · LHL · median 200 Hz
WordRomajiAccent typeMoraeH-LF0 per mora (Hz)
ねこnekoatamadaka (mora 1 high)2HL246 / 202
えきekiatamadaka (mora 1 high)2HL281 / 217
やまyamaheiban (no accent)2LH171 / 214
かわkawaheiban (no accent)2LH230 / 252
はなhanaheiban (no accent)2LH175 / 220
さくらsakuraheiban (no accent)3LHH164 / 252 / 249
おかしokashiheiban (no accent)3LHH167 / 265 / 248
あなたanatanakadaka (accent inside)3LHL174 / 295 / 217

How to practise

Where these numbers come from

F0 per mora = the voiced span split into N equal parts (N = mora count), median of each. H/L = each mora against the word mean. Mora boundaries are not annotated, so the split is an approximation — which is why the page only reports high/low, not precise Hz per mora. Note also that devoiced morae (か, さ, し, く …) have no periodicity of their own — Praat interpolates along the pitch track, so those points show direction only.

Three negative results, all measured: (1) Homophones cannot be demonstrated: TTS gives one reading per string, so two syntheses of はし have identical F0 (238.3 / 217.1), as do 雨 and 飴 (261.9 / 264.1). No fake pairs here. (2) Odaka cannot be measured: in isolation it has the same H/L string as heiban, and this engine lowers every following particle. (3) Six candidates were dropped: いぬ and かお are rendered flat, ともだち and せんせい drop on the last mora, こころ and たまご do not match the expected pattern.