You press play, hear the word, say it back, and press play again. After several repeats it feels familiar. Then someone uses it in a conversation and you miss it, or your version still sounds different and you cannot tell why.
Give each replay a specific job. Listen first, read the word's IPA transcription to identify what you heard, then say it and compare. The symbols give you something concrete to check: a vowel you have been replacing, a quiet syllable you missed, or stress landing in the wrong place.
Why IPA changes what you hear
Ordinary spelling can pull you toward the wrong sound. An unfamiliar vowel can also sound close enough to one from your own language that you copy the familiar one without noticing. Hearing the same recording again does not necessarily reveal that substitution.
The International Phonetic Alphabet, or IPA, gives sounds labels that are separate from a word's spelling. If two words look similar but their transcriptions have different vowel symbols, you have a reason to listen more closely. You can ask, "What changes in the middle?" instead of trying to imitate the whole word at once.
You do not need to memorize the alphabet before starting. Learn the symbols in the words you are practising; our beginner's guide to reading IPA explains how to interpret them, including stress marks.
Treat a dictionary transcription as a guide to the word's sounds. It is not a detailed record of everything a particular voice does. Accent, emphasis and neighbouring sounds can all affect the audio, so keep listening rather than forcing what you hear to fit the symbols.
Set up the listen, read, repeat loop
Choose one familiar word whose pronunciation you are unsure of, or a short phrase you understand. Use a transcription and voice for the same language variety: British English IPA alongside a British voice, for example. Comparing different accents can make a normal variation look like a mistake.
Work through these steps with one item:
- Listen without reading the IPA. Notice the vowel, the number of syllables and the strongest beat. Make a tentative guess about what you hear.
- Read the IPA. Find one feature that your first listen missed or left uncertain. Look up an unfamiliar symbol rather than assigning it the sound of a familiar letter.
- Listen again with that feature in mind. Follow the transcription while the audio plays. Try to locate the sound, rather than silently reading the ordinary spelling.
- Say it after the audio ends. Leave enough silence to finish your version. If a phrase is too much to remember, choose a shorter one.
- Compare and repeat. Record yourself with a separate phone or computer recorder, then play the model and your recording in turn. Change one feature on the next attempt.
Listening before looking gives you a useful first impression. Looking afterwards gives the next replay a purpose. Finish by hiding the transcription and listening once more, so the written cue does not become something you always need.
Try it with ship, sheep and about
These examples use broad British English transcriptions. The word links lead to Wiktionary's pronunciation entries; stress marks are omitted on the two words with only one syllable.
| Word | IPA | What to listen for |
|---|---|---|
| ship | /ʃɪp/ | The vowel in the middle, with the same beginning and ending as sheep. |
| sheep | /ʃiːp/ | A different vowel quality as well as a difference in duration. |
| about | /əˈbaʊt/ | A weak first vowel, followed by a stressed second syllable. |
Start with ship and sheep. Play each word once without looking, then compare the transcriptions. Both begin with /ʃ/ and end with /p/; the vowel is the target. If your versions sound identical, listen again for the vowel's quality, meaning its sound character, as well as its length. Simply stretching your version of ship is not a reliable way to make sheep.
Say each word after its audio, then record the pair. On playback, ask whether your two vowels are distinguishable. The English minimal-pairs practice guide has more contrasts when you are ready to work on another sound.
Now try about. The first symbol, /ə/, is schwa, the weak vowel at the start of this word. The mark ˈ comes before the stressed syllable: the stronger beat belongs on the second syllable. Listen specifically for that balance, then record yourself and check whether you have given the first vowel too much weight.
Those are two different listening tasks. The pair directs attention to a vowel contrast; about directs attention to weak and strong syllables. Use whichever task matches the problem you actually hear.
Compare one feature in your own recording
"Do I sound right?" is too broad to guide the next attempt. "Is my first syllable as weak as the model's?" gives you something you can change. Keep the word, voice and recording conditions reasonably consistent while you work on that question.
Play the model, then your recording, with a short gap between them. Speaking over the model makes it harder to hear your own version. Listen for your chosen feature first; leave differences in pitch or voice quality alone unless they are the focus of the exercise.
If you cannot hear whether your attempt is closer, keep both recordings and ask a teacher or a proficient speaker to compare them. Repetition is useful only when you have a reasonably clear target. A confident impression after many replays is not proof that you are making the intended sound.
Keep a small note such as "first vowel too strong" beside the word. Next session, listen with the IPA hidden before checking that note. You are testing whether you can notice the feature again, as well as whether you can say it.
Where text-to-speech can mislead you
Text-to-speech is convenient because you can request a word whenever you need it. But voices typically read isolated words in their citation form: the careful form you would hear in a dictionary. A conversation does not consist of dictionary entries played in sequence.
English to, for example, has a strong form /tuː/ and a weak form /tə/ that can occur before a consonant, as in to work. Wiktionary's entry for to lists these forms and other variants. Practising only the isolated strong form can leave you unprepared to recognize the weak one in a sentence.
Connected speech means the changes that occur when words are spoken together. Weak forms, linking across word boundaries and sounds being reduced or dropped may be missing or inconsistent in synthetic speech. A TTS sentence can contain some of these features, but a smooth voice is no guarantee that it matches casual conversation.
Synthetic voices can also mispronounce rare words, technical terms and names, or choose an unexpected stress pattern. Replaying an error makes it familiar. If a word matters for a presentation or a person's name, confirm it with a recording of a real speaker; for someone's name, their own pronunciation is the best reference when available.
Keep the IPA and audio open to correction. If they seem to disagree, check the accent, the word's meaning in context and a dictionary or human recording before deciding which version to practise.
Check real speakers, then put the text away
Use Forvo's recordings of real speakers to check an individual word. Compare more than one recording when possible, paying attention to the speaker's region. Agreement on a sound across several examples is more useful than copying an unexplained difference from a single recording.
For phrases, search YouGlish's examples in real video clips. Listen to the words before and after your target, too. Notice whether a small word becomes weaker, a boundary sounds joined, or the speaker emphasizes a different word to express a different meaning.
Repeat the same loop with that human recording. Listen once, consult the IPA for a troublesome word, listen again, then say the phrase and compare. A dictionary transcription can explain the word's basic form while leaving conversational changes unmarked; let the actual recording guide your phrase practice.
Once you can repeat a short phrase comfortably, try the shadowing routine for rhythm and connected speech with a recording and transcript. That adds the challenge of speaking along with the person. Keep the pause-and-repeat exercise for spots that need closer attention.
End the session with an audio-only check: hide the text and listen to a different speaker using the word. Can you recognize it and hear the feature you practised? That check connects your careful practice to the listening problem you started with.
Try the loop with words you already read
Choose three words from something you are reading and give each one a specific listening target. In the IPAtics desktop app for Mac and Windows, select text in any app, press a hotkey, see the IPA and play the audio. It uses natural Google voices, including US and UK English voices.
Replaying audio is unlimited on every plan. Free allows up to 20 characters per lookup; Premium allows up to 100. You can compare the plans and lookup limits if you practise longer phrases. Use a separate recorder for your own attempts, and finish by checking the word in real speech.