Phonemes
All 41 English phonemes have a page of their own. The table sorts them by where the sound is made — from the lips back to the glottis, with voiceless and voiced sharing a row. The table itself is dictionary knowledge, not a measurement. Click any cell to hear real speech, see the articulation, and check every number.
Syllables
The three things that decide whether you are understood — and none of them is a phoneme. Final consonants, consonant clusters and the -ed / -s endings are all about syllable structure: which sounds may close a syllable, and how many consonants may sit together. That is why they have no row in the phoneme table above.
English tells words apart by their ending (bed / bet / beg). If your language does not let a consonant close a syllable, you either drop it or add a vowel after it.
good → goo / good-uhTwo or more consonants with no vowel between them. English allows three at the start and four at the end; most languages allow none — so a vowel gets inserted, or a consonant gets deleted.
street → si-treet / texts → texNot a pronunciation problem — a grammar problem. Lose the ending and He walked becomes He walk. Each has three readings, decided by whether the sound before it is voiced.
walked /t/ · played /d/ · wanted /ɪd/Connected speech
Linking is hard because it does not exist in the dictionary. A dictionary gives every word its own transcription, but in real speech there is no gap between words. Each of the five sections below is about what happens when two sounds meet. Where we have published a section, the examples and every number come from the same recordings. Where we have not, we say which step we are stuck on.
Overview: the criteria, the audio source, and the one sub-class still missing →Linking
measuredHow deep the energy dip is at the word boundary. 0 dB means the seam is as loud as either side — the two words have fused. Higher means it still sounds like two pieces.
In this corpus: 921 word boundaries → 236 of the two linking types → 13 with a genuinely flat seam → 4 in sentences short enough to read at a glance (3 after dropping rare words). We show the ones that survived all of it.
All 921 measured cases →Weak forms
measuredHow long this word actually lasted, against the median word in the same sentence. The smaller the ratio, the flatter it was squeezed. The ratio is a measurement; whether that counts as 'reduced' is not something duration can settle.
In this corpus: 921 measured words → 455 function words → 169 at or below half the sentence median → 43 in short sentences. We show three different words.
All 455 measured cases →Stress
measuredThe dictionary marks one syllable with ˈ. The question is whether that syllable really came out longer and higher in this recording. A stress mark is a rule about the word, not a fact about this speaker.
On 180 multi-syllable words: the stressed vowel is 1.26× the unstressed vowels of its own word, +4.3 dB louder, and +0.9 semitones higher. Loudness moved the right way in 83% of them; pitch in 70%.
All 180 measured cases →Elision
measuredThe last sound of -ing goes from ŋ to n. Almost nothing to the ear — both are nasals — but real in spelling and in how people actually talk. This page reports no milliseconds: something 383 ms against somethin' 381 ms.
The 6 pairs, side by side →Reductions
measuredTwo ways of saying the same thing: how long the full form takes, how long the spoken form takes, and the difference. The figure is the whole phrase, not one sound.
The 6 pairs, side by side →Assimilation
measuredTwo sounds next to each other, and the first gets pulled toward the second. This page covers t + y → tʃ only: don't you sounds like doncha. The evidence is the transcription, not the duration — the two differ by about 1 ms, because the syllable count does not change.
The 6 pairs, side by side →Intonation
measuredAt the end of a sentence, does the pitch fall or rise? Measured as the F0 slope over the final syllable, not judged by ear.
In 60 sentences, 55 end lower than they started. Median slope over the final third of a second: -4.5 semitones per second; the opening sits 8.3 semitones above the ending.
All 60 measured cases →