Three numbers
How long the airflow lasts
Horizontal axis in milliseconds. The thick bar is each language's range, the vertical line its median.
12 words, every one measured
Press the button to hear the span itself: airflow plus the vowel that follows. If you cannot hear it, hold the back of your hand in front of your mouth — Spanish should give you nothing.
| Word | IPA | Aspiration | |||
|---|---|---|---|---|---|
| peso | /peso/ | /p/ | 5 ms | ||
| pata | /pata/ | /p/ | 5 ms | ||
| poco | /poko/ | /p/ | 10 ms | ||
| papel | /papel/ | /p/ | 10 ms | ||
| tiempo | /tʝempo/ | /t/ | 20 ms | ||
| tapa | /tapa/ | /t/ | 10 ms | ||
| todo | /todo/ | /t/ | 20 ms | ||
| tomar | /tomaɾ/ | /t/ | 10 ms | ||
| casa | /kasa/ | /k/ | 15 ms | ||
| cosa | /kosa/ | /k/ | 20 ms | ||
| cama | /kama/ | /k/ | 20 ms | ||
| cocina | /koθina/ | /k/ | 20 ms |
Three ways to fix it
From easiest to feel to hardest to fix — do not start with the third.
Hand in front of your mouth: feel that there is nothing
Say pata; your hand should feel no cool air. Say 「怕」 and it will. No tool needed and it will not lie to you — first let your body know the target.
Borrow the shape of b
Spanish b and p differ only in voicing (see the voicing page); neither is aspirated. Say bata, then switch the voice off and keep the closure — the airflow disappears by itself. Working backwards from a puff is much harder, because your hand has already moved.
Start with the stop in the middle of a word
Mandarin speakers aspirate most on the first stop of a word (pata); the medial one (papel, tomar) often comes out right, because intervocalic stops are shorter and looser. Fix the medial one first, then move it to the front.
Where these numbers come from
Corpus: 12 Spanish sentences whose target word starts with p / t / k (four each), with the target never first or last in the sentence. All from one Spanish female voice.
Measurement: audio cut into 5 ms frames with per-frame energy and periodicity; find the first frame of the vowel (enough energy, periodic, three frames in a row), then walk back to silence — that stretch is the aspiration.
Reference: the same method on 6 English and 6 Mandarin words, using other voices of the same engine. What separates them is not voice colour but the aspirated/unaspirated category.
Limits: these are synthetic recordings; human read speech usually shortens the span. The page reports this batch only. Judge whether there is air by presence, not by exact milliseconds.
Why not the word VOT: the textbook definition needs an exact burst point, and burst location is unreliable on this corpus (the first version scored 0 on 14 of 20 — and 0 is exactly what unaspirated should look like, so the bug nearly became the conclusion). Measuring back from the vowel onset is stable.