Answer in brief
Look the word up in a dictionary that gives a transcription and audio, read the IPA symbols and the stress mark, and keep to one model, British or American. Then train the sounds Russian lacks or handles differently, th, w, long and short vowels and voiced final consonants, and check yourself by listening, shadowing and recording.
Why English spelling hides the sound of a word
Most people who look up how to pronounce English words hit the same wall: the letters do not show the sound. Wikipedia's article on English orthography explains why. Most spelling conventions come from Middle English and generally do not reflect the sound changes since the late 15th century, the Great Vowel Shift among them. Borrowed words often kept their foreign spelling too.
The same article gives numbers. The letters ough can be pronounced in at least ten ways, as in though, tough, cough and through. The vowel /uː/ can be spelled in up to 24 ways. Some letters are silent, like the l in talk and the w in two. A transcription, which records sounds rather than letters, is the place to start.
Reading an IPA transcription and the stress marks in it
Dictionaries write the sound of a word between slashes in the International Phonetic Alphabet, one symbol per sound. Wikipedia's key for English, Help:IPA/English, pairs each symbol with ordinary words: /θ/ as in thigh, /ð/ as in father, /æ/ as in trap, /ʌ/ as in strut. Learn first the symbols that are not ordinary letters, because they carry the sounds that spelling hides. The rest can wait until you meet them.
Symbols vary between dictionaries. The key warns that some of its choices are specific to Wikipedia: many dictionaries write the vowel of let as /e/ rather than /ɛ/, and the Oxford English Dictionary uses /a/ for /æ/. The mark ː, as in /iː/, shows a long vowel, though such vowels are often shorter before voiceless consonants. Read the guide of your own dictionary once.
Stress is marked with a small vertical stroke. In IPA, ˈ for the main stress and ˌ for a secondary one stand before the stressed syllable, unlike in some dictionaries published in the United States, the key notes. Its example is intonation, /ˌɪntəˈneɪʃən/. Wiktionary transcribes record as a noun with the stress on the first syllable and as a verb with the stress on the second, and the vowels change with it.
British and American pronunciation in the same dictionary entry
Many entries carry two pronunciations. Wiktionary labels them Received Pronunciation and General American and often adds a recording to each; Wikipedia calls General American the umbrella accent used by a majority of Americans. Wikipedia's own key is diaphonemic: one transcription serves both accents and, to a large extent, Australian, Irish and Scottish English. Check which line you are copying.
The first difference learners hear is r. Wikipedia's article on rhoticity describes Received Pronunciation as non-rhotic: r after a vowel is silent unless a vowel follows, so hard and butter lose it. American English, by contrast, is now predominantly rhotic. In car owner the British r comes back; this is the linking r.
The second is t between vowels. Wikipedia's article on General American says /t/ and /d/ become a flap between a vowel and an unstressed vowel, so latter and ladder can sound alike. Wiktionary shows it on water: /ˈwɔː.tə/ in Received Pronunciation, [ˈwɔ.ɾɚ] in American speech. Neither model is more correct: pick the one you hear most and keep it.
The th sounds and the w and v pair for Russian speakers
English has two th sounds: voiceless /θ/ in think and voiced /ð/ in breathe. Wikipedia's article on the voiceless dental fricative calls it rather rare as a phoneme among the world's languages, and the consonant inventory in its article on Russian phonology has neither. The tip or blade of the tongue touches the upper teeth, often resting between upper and lower teeth. Air passes through a narrow, flat gap, without the groove of s.
Speakers who did not acquire the sound in childhood typically replace it with s, t or f, the same article says; linguists call these th-alveolarization, th-stopping and th-fronting. Each swap risks another word: think with an s sounds close to sink. Practise slowly in front of a mirror, where the tongue should be visible. Then say the word at normal speed inside a short phrase.
Russian has no /w/ either, so v is the nearest familiar sound. Wikipedia's articles on the two sounds place them differently: for /w/ the back of the tongue rises toward the soft palate while the lips round; for /v/ the lower lip touches the upper teeth. Wiktionary gives wine as /ˈwaɪ̯n/ and vine as /ˈvaɪ̯n/: one sound apart, two meanings. Start the w with rounded lips, close to a Russian у, and keep the lower lip off the teeth.
Long and short vowels and voiced consonants at the end of words
Russian tells vowels apart by quality and stress; the article on Russian phonology describes no contrast in length. English has one. Wiktionary transcribes sheep as /ˈʃiːp/ and ship as /ˈʃɪp/, and Wikipedia's article on minimal pairs lists bin and bean. Since the length mark does not always mean a longer vowel, copy each vowel from its recording, quality and length together, instead of stretching one sound.
A second habit comes from Russian, where voiced consonants are devoiced at the end of a word. The article on Russian phonology gives рассказ pronounced with a final s, нож with sh and Иванов with f. Carried into English, the rule merges pairs such as had and hat. The listener then has to guess the word from context.
English gives one more clue. The IPA key notes that vowels before voiceless consonants are often shorter, so the vowel in hat is shorter than the one in had. Keep that vowel a little longer and let the voice run on into a final d, g, v or z. Then record had and hat in a row: they should still sound like two words.
Checking a word with dictionary audio and browser speech recognition
Listen to a new word before you learn it. Wiktionary attaches recordings to many entries, often a British and an American one, as on water, sheep and wine. Browsers can help too. The W3C draft of the Web Speech API, a Community Group report rather than a finished standard, defines speech synthesis, which reads text aloud, and speech recognition, which turns speech into ranked text guesses, each with a confidence between 0 and 1.
Know the limits. The draft lets a browser listen only with explicit, informed consent and, unless a page asks for local processing, the browser may handle speech on the device or remotely; MDN's documentation says recognition uses a service provided by the user's platform by default. Recognition is built to produce text, not to grade sounds, so a sentence may come out right from context despite a wrong vowel. Test single words and minimal pairs instead.
Shadowing a speaker and recording your own pronunciation
Shadowing means repeating speech as you hear it, slightly behind the speaker. Wikipedia's article on speech shadowing says the method was created for research by the Leningrad group of Ludmilla Chistovich and Valerij Kozhevnikov in the late 1950s; close shadowing runs at a delay as short as 250 milliseconds. In language learning it is used to practise speech, with attention to accuracy in pronunciation and pitch. It needs only a recording and its transcript.
Take a minute of audio with a transcript, listen once while reading, then speak along with the voice, one step behind, several times in a row. Slow the playback at first if your player allows it. Shadowing makes you copy the melody and stress of whole phrases, while single sounds such as th still need slow work. Keep the same clip for several days.
Minimal pair drills are older. Wikipedia's article on minimal pairs says they were central to pronunciation teaching in the 1940s and 1950s, were later called artificial and remain common; it treats the claim that learners cannot hear contrasts absent from their own language with caution, since evidence for it is hard to find. Record sheep and ship, wine and vine, had and hat, and listen the next day as a stranger would. A phone recorder is enough.
Practising English pronunciation alone and what a course adds
A week alone is enough to start. Learn the symbols of one dictionary, pick British or American, choose ten words with th, w, long vowels and voiced endings, then listen, record, compare and shadow for a few minutes a day. That is how to pronounce English words you meet for the first time, at no cost. Self-study lacks a schedule for bringing words back and a regular check, so a corrected sound can slide back.
VITON13 SCHOOL's course English in 113 Days is written for readers of Russian: every explanation in it is in Russian. Each of its 113 levels brings 13 new words with sound, a transcription and an example, and the words return for review through FSRS spaced repetition. The browser voices the words through the Web Speech API and, where it can recognise speech, the course checks your pronunciation. The first three levels are free, even without signing in, and a free placement test shows where to begin.
Practical checklist
- Read the pronunciation key of your dictionary once and mark the symbols that are not ordinary letters.
- Choose British or American pronunciation and take every new word in that same version.
- List ten words with th, w, long vowels and voiced endings, and listen to the recording of each.
- Record yourself reading pairs such as sheep and ship, then compare the take with the dictionary audio.
- Shadow a short recording with its transcript for a few minutes a day, one step behind the speaker.
Questions and answers
Is it better to learn British or American pronunciation of English?
Either works as long as you keep to one model. Choose the accent you hear most in films, at work or from teachers, and take every new word in that version. Dictionaries such as Wiktionary give both, so check which line you are reading before you copy it.
Do I need the whole phonetic alphabet to read English transcriptions?
No. English uses only part of the IPA, and Wikipedia's English key lists those symbols with example words on one page. Learn the ones that differ from ordinary letters, such as θ, ð, æ, ʌ and the length mark, and look up the others when they appear.
Why does th come out as s, t or f in a Russian accent?
Russian has no dental fricatives, so speakers reach for the nearest sound they already have. The replacements even have names in linguistics: th-alveolarization for s, th-stopping for t and th-fronting for f. Slow practice with the tongue touching the upper teeth, then faster practice inside phrases, builds the new movement.
Can a browser's speech recognition judge my English pronunciation?
Only roughly. It turns speech into text and gives each guess a confidence between 0 and 1, but it is built to understand you, not to mark individual sounds, and it leans on context. Check single words and minimal pairs, and confirm with your own recording against the dictionary audio.
How often should I practise English pronunciation on my own?
Short daily sessions are easier to keep than one long session at the weekend, and ten minutes of listening, repeating and recording fits into almost any day. Spread your words over the week and return to the ones that still sound wrong on your recording.