Chinese Pronunciation: A Complete Guide to Mandarin

Every syllable is an initial, a final and a tone. There are 21 initials, around 35 finals, and 5 tonal categories. Not every combination is legal, which is why the total lands at roughly 1,300 rather than the thousands the arithmetic would suggest.

Mandarin syllables are also structurally constrained in a way English syllables are not. There are no consonant clusters at all — nothing like 'street' or 'sixths'. A syllable can end only in a vowel, -n, -ng, or the retroflex -r. This means Chinese speakers learning English struggle with clusters, and English speakers learning Chinese tend to add consonants that should not be there.

Initials group into families, and the families are where the contrasts live.

The single most consequential fact about Mandarin consonants is that the b/p, d/t, g/k, j/q, zh/ch and z/c pairs are distinguished by aspiration — a puff of air — rather than by voicing.

English distinguishes them by voicing: the b in 'bin' vibrates your vocal folds, the p in 'pin' does not. English also aspirates its voiceless stops at the start of a word, which is why 'pin' has a puff and 'spin' does not.

In Mandarin, both members of each pair are voiceless. b is an unaspirated p; p is an aspirated p. Hold a sheet of paper in front of your mouth: 爸 bà should not move it, 怕 pà should.

The good news is that English speakers usually produce this correctly by accident, because the English unaspirated allophone in 'spin' is very close to Mandarin b. The problem is perception — learners hear the contrast as voicing and so mishear unfamiliar words.

j/q/x against zh/ch/sh is the first. Both are 'sh-like' to an English ear because English has only one sound in that region. The difference is tongue position: for j, q, x the tip points down behind the lower teeth and the blade arches toward the hard palate. For zh, ch, sh the tip curls up and back. Drill with 家 jiā against 抓 zhuā, 七 qī against 吃 chī, 西 xī against 是 shì.

The zi/ci/si and zhi/chi/shi/ri vowel is the second. That i is not a vowel in the usual sense — it is a syllabic continuation of the consonant's own buzz. 四 sì is a sustained s with voicing; 十 shí is a sustained retroflex sh with voicing. There is no 'ee' anywhere in either.

ü is the third. It is a high front rounded vowel, present in French and German but not in English. Say 'ee' and, without moving your tongue at all, round your lips as if for 'oo'. The temptation is to move the tongue back, which produces 'oo' instead. 女 nǚ against 努 nǔ is the minimal pair to drill.

Mandarin's finals are built from a small vowel set plus glides and the endings -n, -ng and -r. The complication is that several vowels change quality depending on what surrounds them, and pinyin does not always show it.

Learning the four tones in isolation is the easy part. Producing them in a sentence, where each syllable's pitch is affected by what surrounds it, is where the real difficulty is.

The third tone is rarely produced in full. Its complete dip-and-rise contour appears only in isolation or at the end of a phrase. Elsewhere it is a low tone with no rise, or — before another third tone — it surfaces as a second tone entirely.

Neutral tone syllables are short, light and pitched relative to what precedes them. 妈妈 māma has a neutral second syllable pitched mid-low; 爷爷 yéye has one pitched higher. Giving neutral syllables a full tone is one of the most audible learner markers.

Sentence-level intonation exists on top of tones, not instead of them. Mandarin does raise overall pitch for surprise or emphasis, but the tonal contours are preserved within that raised register. English speakers tend to let question intonation flatten the final syllable's tone, which is why 你是学生吗 so often comes out with a rising 吗.

Work on perception before production. You cannot reliably produce a contrast you cannot hear, and drilling production without perception just reinforces whatever you are already doing. Minimal-pair listening — two words differing in one sound, played back to back — is the standard technique and it works.

Use a model with synchronised text. Hearing a sentence while seeing where each word begins and ends solves a problem that audio alone cannot: Mandarin gives no acoustic cue to word boundaries, so a learner hears an undifferentiated stream. Watching the segmentation while listening builds the parsing skill directly.

Shadow rather than repeat. Speaking simultaneously with a recording forces you to match pitch contour and rhythm in real time, which is a substantially different exercise from repeating after a pause and produces faster improvement in prosody.

Record and compare. The gap between what you believe you produced and what you actually produced is where errors live, and the only way to see it is to listen back against a native model of the same text.