Say this out loud, at the speed you’d use talking to a friend: Could you get me a glass of water?
Now slow down and notice what your mouth wanted to do. The careful, textbook version has eight separate words with clean edges. The version an American says has maybe three lumps: something like KUH-juh geh’-me uh GLAS-uh WAH-der. The words ran into each other. The t in water turned into a soft d. The of lost almost everything it had. Could you fused into a single kuh-juh.
If you’ve spent years on your individual sounds and people still ask you to repeat yourself, this is usually why. Your vowels are probably fine. Your consonants are probably fine. What you were never taught is what happens in the gaps between the words, where American English does most of its real work. The textbook gave you the bricks. Nobody gave you the mortar.
Connected speech is the set of changes that happen when words run together in natural speech. American English isn’t spoken faster than the version you learned; it’s spoken fused. Five mechanisms operate at the seams between words: words link (consonant slides into the next vowel), tiny glide sounds intrude between vowels, weak sounds get elided (dropped), neighboring sounds assimilate (blend into each other), and unstressed function words weaken to a schwa. Learn to hear these five and the wall of fast American speech resolves into something you can follow. Learn to produce them and you stop sounding like you’re reading the sentence off a card.
What connected speech actually is
Connected speech is what happens to words when you stop saying them one at a time.
In isolation, did is did and you is you. Put them next to each other at conversational speed and they become DIH-juh, the didja you’ve seen written in dialogue. The d and the y collided and made a new sound that was in neither word. That collision is connected speech, and English does it constantly, by rule, not by carelessness.
When learners say American speech is “too fast,” they’re rarely reacting to actual speed. A newscaster reading at a measured, formal pace is still linking, dropping, and blending sounds the whole time. The difficulty has little to do with tempo. The word boundaries you’re listening for have dissolved. You’re waiting to hear eight words and you’re getting three blurred shapes, so your ear falls behind while it tries to chop the stream back into pieces.
American English isn’t spoken faster. It’s spoken fused. Stressed syllables get the full, clear sounds, and everything around them gets compressed, linked, and reduced to keep the rhythm moving. (English is what linguists call a stress-timed language: it compresses the unstressed material between stressed beats instead of giving every syllable equal weight. Spanish, Italian, and most other syllable-timed languages don’t do this, which is exactly why the habit feels foreign.) The mortar between the bricks is where that compression lives.
The five things that happen at the seams
Almost everything that makes connected speech hard to follow comes down to five mechanisms. They overlap in real sentences, but it’s worth meeting them one at a time.
1. Linking: the end of one word slides into the start of the next
When a word ends in a consonant and the next word starts with a vowel, Americans don’t restart the airflow between them. The consonant just slides over and attaches to the vowel. An apple becomes uh-NAP-ul. Turn it off becomes tur-ni-doff. Get out becomes geh-dout. The consonant you’re listening for at the end of the first word has quietly moved to the front of the second, which is why an apple and a napple sound identical. (In get out and turn it off, the linked t also flaps to a soft d; connected-speech changes stack on each other.) This is the most common mechanism by far, and the SayWaader reference page on consonant-to-vowel linking has the full pattern.
Vowel-to-vowel linking is the same idea without a consonant to carry it, and it sets up the next mechanism.
2. Intrusion: a tiny glide appears between two vowels
When one word ends in a vowel and the next begins with one, your mouth needs something to bridge the gap, so it inserts a small glide you never spelled. After an “ee” or “ay” sound, the bridge is a faint y (/j/): I agree comes out I-yuh-GREE, the end becomes thee-YEND. After an “oo” or “oh” sound, the bridge is a faint w (/w/): go on becomes go-WAHN, do it becomes do-WIT. How automatic this feels depends on your first language: if yours already bridges vowels with a glide, you do it without thinking, but if yours separates vowels with a hard catch in the throat, as German, Dutch, and Arabic tend to, you have to consciously replace that catch with a smooth glide. The reference for vowel-to-vowel transitions is the vowel-linking page.
One honest warning, because a lot of pronunciation content gets this wrong: you may have read about an “intrusive r,” as in law and order becoming law-r-and order. That’s a feature of non-rhotic accents: British RP, Boston, parts of New York. General American is rhotic, so it isn’t your target. If you’re aiming at a standard American sound, the y and w glides are your targets and the intrusive r is not.
3. Elision: sounds that quietly disappear
Some sounds get dropped entirely when they’d be awkward to pronounce. The biggest culprit in English is the t and d caught in the middle of a consonant cluster. Next day becomes neks-day. Must be becomes muss-bee. Sandwich drops its d to SAN-wich. Friendship loses its d. The general rule: when t or d is squeezed between two other consonants, it tends to vanish. Unstressed vowels disappear too, which is how probably becomes PROB-lee and every becomes EV-ree. The unstressed h in pronouns goes the same way: tell her runs together as tell-er and get him as geh-dim, the h dropped and the words linked. The cross-word version lives on the elision reference page, the cluster-specific t-dropping has its own entry on dropped-T-in-clusters, and the pronoun pattern lives on the dropped-H page.
4. Assimilation: neighboring sounds blend into each other
When two sounds sit next to each other and one is awkward to follow with the other, the first often shifts to match its neighbor. This is where did you becomes DIH-juh and would you becomes WUH-juh: the d and the y merge into a j sound (/dʒ/) that was in neither word. Won’t you becomes WONE-chuh by the same logic with a t. Assimilation also runs inside single words: tree comes out closer to chree and dream closer to jream, because the American r drags the t and d toward “ch” and “j” (the TR and DR shifts). Even across words, ten bucks drifts toward tem bucks as the n leans toward the b. The general reference is the assimilation page.
5. Weakening: the small words hollow out
Close to half the words in ordinary English speech are function words: of, to, and, for, the, a, you, your, that, can, was, are, would. Almost none of them are said the way they’re spelled. Their vowels collapse to a schwa and they shrink into the gaps between the content words. Of becomes uhv or just uh. And becomes un or n, so salt and pepper is salt-n-pepper. To becomes tuh. Your becomes yer. This weakening is what the schwa is for, and it’s the single biggest reason a careful sentence sounds non-native: if you give every little word its full dictionary vowel, you flatten the rhythm the whole language depends on. SayWaader has two deeper articles on this one mechanism alone: the schwa and the seventeen everyday reductions Americans lean on most.
These five aren’t separate topics you can study in any order. They’re five views of one habit: keep the stressed beats, compress everything else.
One sentence, decoded
Go back to the sentence we started with. Here’s what fires across Could you get me a glass of water?, joint by joint (the you + get and a + glass joins are left out because nothing changes there).
| The seam | What happens | Mechanism |
|---|---|---|
| Could + you | d + y merge into /dʒ/ → KUH-juh | Assimilation |
| get + me | the t closes off in the throat (a glottal stop), not released → geh’-me | stop non-release |
| me + a | a faint y glide bridges the two vowels → mee-yuh | Intrusion |
| a (alone) | weak form, vowel collapses → uh | Weakening |
| glass + of | glass runs straight into of, which flattens to a schwa → GLAS-uh | Linking |
| of (alone) | of often drops its v before a consonant → uh | Weakening / elision |
| inside water | the t sits between two vowels, so it flaps to a soft d → WAH-der | Flap-T |
Two of those rows sit outside the five mechanisms: the glottal stop in get me and the flap-T inside water are sound-level changes that ride along with connected speech, each with its own article linked below. The other five rows are the core mechanisms in action. Assimilation, intrusion, and linking each fire once; weakening fires twice, on a and on of, and on of it drops the v by elision as it goes. All five mechanisms, in eight words. Eight words, and only two of them land with real stress, the content words glass and water. What comes out is three rhythmic lumps (KUH-juh · geh’-me-uh · GLAS-uh-WAH-der) built around those two beats. That’s not sloppy speech. That’s a fluent, educated American asking for water at a dinner table.
Notice that the meaning lives almost entirely in those two stressed words. If you heard only glass and water out of the whole sentence, you’d reconstruct the request perfectly. That’s the design of the language: load the content words, throw away the edges of everything else.
How to hear it before you try to do it
You can’t produce a pattern you can’t hear, and most learners try to skip straight to production. Spend a week on the ear first.
Pick sixty seconds of unscripted American speech: a podcast, a talk show clip, a sitcom scene, not a slow ESL listening track. Play it once at normal speed and just feel the difficulty. Then play the same clip with the transcript or subtitles in front of you and read along. The gap you feel is the exact gap connected speech creates: you know all those words, and you still couldn’t catch them in the stream, because they weren’t pronounced as separate words.
Now do one targeted pass. Pick a single mechanism (say, linking) and listen only for it. Every time a word ending in a consonant runs into a word starting with a vowel, mark it. Pick it up. Turn it on. Hold on a second. Once you’ve spent ten minutes hunting one mechanism, your ear keeps flagging it on its own afterward. Then switch to a different one the next day. The point isn’t to catch everything at once. You’re retraining what your ear treats as a word boundary, and that retraining is what eventually lets fast speech slow down in your head.
How to do it
The instinct that fights you here is the careful one: the habit of giving every word a clean start and a clean finish, because that’s how you were taught to be “correct.” Connected speech asks you to do the opposite: stop resetting between words and let them run.
The most useful single change is to stop thinking in words and start thinking in chunks. Native speakers don’t plan a sentence word by word; they plan it in breath-groups, and the words inside a group fuse together. Try it with a phrase, a breath, a phrase: Could-you-get-me (breath) a-glass-of-water. Inside each chunk, refuse to stop your voice. The consonants and vowels should spill into each other the way they do when you hum a tune without separating the notes.
Start from the reductions, because they unlock the rhythm fastest. Take any sentence and first weaken every function word to its schwa: I was going to ask you for it becomes I wuz gunnu ask-yuh fer-it. Then add the links and the flaps on top. If you get the small words to shrink, the linking and flapping tend to follow on their own, because once the function words are out of the way the content words naturally lean on each other.
And resist over-correcting. The single most common mistake after learning these patterns is doing them everywhere, including the places Americans don’t. You still keep full clear t’s at the start of stressed syllables, you still pronounce t’s and d’s that aren’t trapped in clusters, and you don’t flap a t at the very end of a sentence. Connected speech is a default, not a law. The articles on the flap-T and the glottal stop both spend time on exactly where each pattern stops, and those boundaries matter as much as the patterns.
Practice phrases
Read each one twice. First the spelled version slowly, then the spoken version at conversational speed, letting the words run together. The fused part is marked.
- Could you get me a glass of water? Kuh-juh geh'-me uh GLAS-uh WAH-der?
- What do you want to do? Whuh-duh-yuh wanna do?
- I was going to ask you about it. I wuz gunnu ask-yuh uh-bou-dit.
- Turn it off and come on in. Tur-ni-doff un come on-in.
- Did you find out what happened? DIH-juh fine-dout what HAP-und?
- It's a matter of getting it done. Its uh MAD-er uh geh-ding-it done.
- Let me know if you need anything. Lemme know if-yuh need EN-ee-thing.
- Would you mind waiting a second? WUH-juh mind WAY-ding uh SEC-und?
If they feel like tongue-twisters at first, that’s the right kind of hard. You’re asking your mouth to give up edges it has defended for years. Give it a week of the same eight phrases before adding new ones.
Where your first language puts you
How natural connected speech feels depends a lot on the rhythm of your first language. None of these are deficiencies, just different starting points.
| Your L1 | What carries over | What to focus on |
|---|---|---|
| Spanish, Italian, Brazilian Portuguese | Vowel-to-vowel linking and the flap (single r) are already in your mouth | Vowel reduction. Syllable-timed languages keep every vowel full; the hard shift is letting unstressed vowels collapse to schwa. |
| French | Linking itself is natural (French liaison is the same instinct) | Stress and weak forms. French stresses the end of phrases evenly, so the stressed-beat-plus-weak-rest rhythm of English takes deliberate work. |
| Mandarin, Cantonese | You’re used to clear, separable syllables | Almost all of it. Connect words on purpose, weaken function words, and let final consonants link forward instead of stopping. |
| Japanese | A strict consonant-then-vowel syllable shape | Avoiding inserted vowels. Japanese tends to add a small vowel after a final consonant (and → ando), and that extra vowel is what blocks linking; the consonant needs to attach to the next word, not get its own beat. |
| Korean | Native resyllabification already links a final consonant onto a following vowel | Lean on that instinct, it transfers directly. The thing to watch is the small vowel Korean adds to break up English consonant clusters, since that inserted vowel is what cuts a link. |
| Hindi, Tamil | The flap (single r) is already native | The rhythm is syllable-timed, so English stress-timing is a new system to build rather than a carryover. Weakening function words and resisting equal stress on every syllable is the main work. |
| German, Dutch | Stress-timing and reductions transfer well | A gentler attack. The hard glottal onset before vowels (a clean restart before each vowel-initial word) is exactly what blocks linking; let the words run instead. |
FAQ
No. Connected speech is the linking, dropping, and blending of sounds at word boundaries, and it happens at every speed, including slow and formal speech. A newscaster reading deliberately still links and reduces constantly. What makes fast American speech hard to follow is the connected-speech changes, not the tempo, which is why slowing a recording down often doesn’t make it easier to parse.
Usually the opposite. Native listeners are tuned to these patterns and can find them harder to follow when every word is fully separated, because the rhythm they rely on to predict what’s coming is missing. Clear stressed syllables matter far more for intelligibility than clean word edges. The goal isn’t to mumble; it’s to put your effort into the stressed beats and let the rest compress.
No. Learn to hear one at a time. Most learners get the fastest gains by starting with function-word weakening (the schwa and reductions), because that single change unlocks the rhythm and makes the linking and flapping easier to add on top. Pick one mechanism, spend a week hearing it in real speech, then move to the next.
Neither. Linking, elision, assimilation, and reduction are standard features of educated General American speech. Professors, judges, and news anchors all use them. They belong to the spoken language, not to slang, and they should never appear in formal writing, where you always write the full words.
They share the core mechanisms (linking, elision, assimilation, weak forms) but differ in the details. The clearest difference is the intrusive r, common in non-rhotic British speech (law-r-and order) and absent from rhotic General American. American English also leans heavily on the flap-T between vowels, where standard British keeps a clearer t.
Hearing the patterns reliably takes a few weeks of focused listening. Producing them without thinking takes longer and depends on your first language, but most learners notice the rhythm starting to feel automatic in two to three months of regular practice on real phrases rather than isolated words. The companion piece on how long it takes to change an accent breaks the timeline down further.
The fluent-sounding sentence and the textbook sentence contain the exact same words. The difference is entirely in the seams: what links, what drops, what blends, and what shrinks down to a schwa. That’s the part no one teaches, and it’s the part that was making people ask you to repeat yourself. American speech was never going faster than you could handle — it was just fused at the joints you were listening for. Pick one seam this week, the linking one is easiest to catch, and listen for it until you can’t stop hearing it.
