Assimilation Put two sounds side by side and the first one gets pulled toward the second. don't you becomes doncha — the t is not dropped, it fuses with the y into a single tʃ.
Six pairs, compared by transcription
Both cells of a pair are read by the same speaker. This page reports no milliseconds — assimilation swaps one sound for another, the syllable count stays the same, so the overall duration barely moves. What changes is the transcription.
Why you cannot hear it
Every point below can be heard in this page's own audio.
You read it as two words; native speakers say one sound
You put a pause between the words. Native speakers take the t at the end of don't and the j at the start of you and release them together — what comes out is one tʃ, not two sounds.
This is the first pair on the page. When you listen, count the symbols, not the words.
The change happens at the seam, not inside a word
That is why a single word's transcription never shows the problem: don't on its own is perfectly normal, and so is you. Only when they are joined does the tʃ grow at the seam.
Which is why this page shows the transcription of the whole phrase.
d + y is a different animal
Textbooks often list did you → didja alongside don't you → doncha. They are not the same: the first is d + y → dʒ, the second is t + y → tʃ.
This page covers t+y only, because the synthesiser does not fuse d+y (measured: the j in didja keeps 71 ms of its own). We cannot produce that class, so we leave it out.
Overcorrecting when you speak
Mandarin ch is close to English tʃ but not identical: the tongue sits further forward and there is slight lip rounding. Imitate the audio here, not pinyin.
This page covers recognition; the mouth position is on the tʃ / dʒ page.
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 | Word length | target word length | F0 | Pitch range | HNR |
|---|
Where the numbers come from
This section documents how each number is computed, so it can be checked.
Corpus: 12 English sentences synthesised by a TTS engine in 4 American voices (two male, two female), arranged as 6 pairs (unassimilated form / assimilated form). Audio is 22.05 kHz mono at 64 kbps.
One speaker per pair: so the only difference between the two cells is the assimilation itself.
No duration difference is reported: assimilation replaces one sound with another and the syllable count is unchanged, so overall duration barely moves (measured: don't you 322 ms against doncha 323 ms). Printing that 1 ms would read as "no difference" — the difference is in the transcription.
Where the transcription comes from: both sides are the G2P reading of the spelling, not a transcription of the audio. That is, the tʃ on the right is what doncha is supposed to say; forced alignment separately confirms the sound is actually present (gotcha's tʃ measures 161 ms).
Target words sit mid-sentence: -cha forms degrade in forced alignment, and when they do the final word's end time is stretched to the end of the clip (measured: fishin' came out at 1252 ms). Mid-sentence, word boundaries come from the engine and are accurate.
t+y only: for didja (d+y) the engine produces d ɪ d j(71 ms) ə — the j stays separate and never fuses. That class needs conversational human speech; this page does not cover it.