Shadowing practice
Shadowing means repeating a recording as exactly as you can — the rhythm and the pitch, not just the sounds. Play the model, say the word yourself, and this page draws both pitch curves on top of each other and scores how closely they match.
The model is our own recording of the word, so the comparison already includes what a tone chart cannot show: tone sandhi (你好 is read ní hǎo), neutral tones, and the pace a real speaker uses. The score is not our opinion — it is the distance between your pitch curve and the recording's.
There is a Sentences group too: 4,225 example sentences, scored on rhythm (where you pause, how fast you read) as well as intonation, because a sentence is not just a long word.
- Words with a model
- 4,231
- Example sentences
- 4,225
- Characters
- 1,050
- HSK 1
- 506
- HSK 2
- 750
- HSK 3
- 953
- HSK 4
- 972
How the score works
1. Your melody
Both recordings are reduced to a pitch curve — 32 points across the voiced part of the word, each measured in semitones above or below that speaker's own middle pitch. Comparing in semitones rather than in a stretched 0-100 shape is what keeps a small natural wobble from being read as a wrong tone.
2. How far you move
A falling tone read flat and the same tone read with half the fall are two different mistakes, so the pitch range of the two curves is scored separately. This is the part that tells you to push the changes further.
3. Pace
Timing is scored on a log scale, so twice as fast and twice as slow cost the same. Rushing a word is one of the most common reasons a tone does not land.
Multi-syllable words also get a per-syllable breakdown, and that part is only an approximation: it splits both curves into equal parts instead of detecting syllable boundaries, so read it as “which syllable drifts the most”, not as a verdict on which tone you said. For single syllables with a known tone, the tone trainer compares against the standard tone values instead.
Sentences are scored differently — on purpose
Comparing a sentence point by point sounds reasonable and is wrong. Measured on the real recordings: take one sentence and warp its timing by 8% — the same melody, just slightly different pacing — and a point-by-point comparison drops from 100 to 57. It is measuring rhythm while pretending to measure pitch.
So sentences use features that do not care about exact timing: where your pitch sits (the distribution across the sentence), the coarse shape of the melody, and how the sentence ends — statements fall, questions rise. Rhythm is then scored separately and explicitly: did you pause where the model pauses (nobody pauses mid-word), and how close your pace is. A sentence read straight through without its commas loses about 20 points even if every tone is right.
What this page does not do
It does not judge your vowels or consonants — only pitch over time. A word can score 90 while one consonant is still off, and vowel quality is often the first thing a teacher corrects. Use the score for tone and rhythm, and keep listening to the model for everything else.
Also useful: pinyin chart · tone trainer · listening practice · your progress