Five vowels · where they sit, and being pure Spanish has a e i o u and every one is pure: the colour stays the same from start to finish, with no off-glide. Mandarin finals do not work that way — the written «o» is usually realised as uo. So the common problem for Mandarin speakers is not failing to produce a vowel, it is adding a tail to it.
Where the five vowels sit in the mouth
Horizontal axis is F2 (tongue front-back, further left = fronter); vertical axis is F1 (openness, higher = closer). The ellipses are interquartile ranges — the same vowel varies this much across speakers and words.
Four steps to a pure vowel
First notice the tail you are adding, then remove it.
Five, and not one more
Spanish has exactly five vowels, with no length contrast (unlike English ship / sheep) and no diphthongal finals of the Mandarin kind. ai / ei / ao / ou are one final in Mandarin but two vowels in Spanish — that difference underlies everything else on this page.
Take the tail off o
The Mandarin «o» is in practice uo: after the rounding, the tongue keeps moving. Spanish o needs rounded lips, a fixed tongue position, and no movement. Hold a long «o» without closing the lips and without sliding towards u.
Do not let e drift to «ei»
Spanish e is [e] — tongue front raised, closer and steadier than English bet. The Mandarin «ei» slides towards i. Compare: mesa should not sound like English «may-sa».
Locate yourself against this chart
If you can record yourself and read a spectrum, compare your values with the numbers here: too far right (low F2) means the tongue is too far back; too high (low F1) means the mouth is not open enough. This site does not record audio — that step is yours to do with another tool.
Which one are you
The three most common vowel habits.
o comes out as «oh-oo», e as «ay»
The classic case: a Mandarin diphthong habit that appends a glide. In Spanish it makes /o/ sound like /ou/ and /e/ like /ei/ — still understandable, but a strong accent.
Every vowel is short and tight
Mandarin syllables are even and short by nature. Spanish stressed vowels are clearly longer (this site measures about 1.3× the unstressed ones). Short vowels make whole sentences sound rushed.
a is fine, but i and u feel unclear
Mandarin has three different i sounds (zhi / zi / ji), Spanish has one. If your i carries friction it drifts towards /ʝ/. Keep it short and pure.
Corpus and acoustic measurements
Click any word in a sentence to play from that word. The pale band over the waveform marks the measured boundary of the target word.
Per-token data
Every number below is measured directly from the audio, not hand-annotated.
| Token | Accent | Sex | Target | Target word | Vowel duration | F1 | F2 |
|---|
Where these numbers come from
Human recordings, and formant measurements — not listening.
Corpus: VoxForge Spanish, 12 recordings from 15 volunteers (prompts are public-domain text). 16 kHz mono. Gender and dialect are labelled as recorded.
Formants: Praat Burg method, 25 ms window, 5 ms step, maximum formant 5000 Hz. We take the median F1 and F2 over the middle 60% of the target vowel (20% trimmed at each end) — the edges are transitions into neighbouring sounds, where the formants are moving by design.
Filtering: frames outside the plausible range are dropped (F1 outside 200–1100, F2 outside 600–3200 means the tracker failed), then a continuity check removes frames where F2 jumps more than 400 Hz from the previous frame — the signature of the tracker confusing /i/ with /u/. Fewer than 3 valid frames and the vowel is not counted.
Sample: 219 vowels across all aligned recordings (a 81, e 44, i 27, o 48, u 19). The five vowels separate cleanly in the F1/F2 plane: a 664/1570, e 442/1952, i 330/2331, o 454/1082, u 390/982.
Is this reliable: for the same speaker and the same vowel, the median standard deviation across repetitions is 66 Hz for F1 and 123 Hz for F2, while the distance between vowels is over 300 Hz (a vs i differ by 334 Hz in F1). The signal is much larger than the noise.
An attempt that failed twice, stated plainly: we first tried to measure whether the vowel moves from start to finish (the F1/F2 trajectory). But the vowels in this corpus have a median duration of 60 ms; after slicing 30 ms halves Praat could not return even two valid frames, and a regression over the whole segment had only 4–9 frames — noise, not movement. So this page reports where the vowel sits (a median needs no more than that) and not how much it moves. In other words: it can tell you where your o should be, not whether your o is moving.
The 12 clips were selected: to keep the chart readable we picked 12 vowels close to each vowel's centre (3 each for a and o, 2 each for e, i, u), spread across speakers. The ellipses and numbers on the chart use all 219, not those 12.