There is a version of this that almost every learner has lived through. Your English is fine in class, fine on a quiet call, fine when you practise at home. Then you walk into a restaurant with your colleagues and suddenly you are repeating yourself three times a sentence, and the confidence you built over months evaporates in twenty minutes.
Your pronunciation did not get worse. The room removed the parts of it your listener needed, and it removed them selectively. Once you know which parts go, you can stop trying harder at the wrong things and start changing the things that actually survive.
It Is Not You, It Is the Channel
Speech carries meaning across a wide range of frequencies, but not evenly. Vowels are loud, long and mostly low-frequency. Consonants, especially the fricatives and the little bursts at the release of /p/, /t/ and /k/, are short, quiet and mostly high-frequency. A vowel can be thirty times more powerful than the /θ/ next to it.
Background noise is the mirror image. Traffic, air conditioning, an engine, and above all the murmur of a crowd sit low and loud, exactly where your vowels are, and they are strong enough to bury anything quiet nearby. So the room does not degrade your speech evenly. It leaves the vowels and the rhythm mostly intact, and it deletes the consonants, which are where most of the word identity lives.
What Goes First
- Fricatives. /f θ s ʃ v ð z/ carry their identity in energy above 4 kHz, and that energy is tiny. In noise they stop being distinguishable from one another and eventually from silence.
- Place of articulation. Within the stops, you usually still hear that a stop happened; what you lose is where it was made. So /p/, /t/ and /k/ collapse together, and so do /b/, /d/ and /ɡ/.
- Word endings. Final consonants are shorter, quieter and often unreleased, so they disappear well before initial ones. Plural /s/, past tense /t/ and /d/, and the third-person /s/ are the first grammar to go.
- What survives: vowel quality, vowel length, syllable count, word stress and sentence rhythm. Everything that lives in duration and pitch keeps working.
The Phone and the Video Call Are a Different Problem
A traditional phone line carries roughly 300 Hz to 3,400 Hz. That is not noise; it is a wall. Most of the energy that distinguishes /s/ from /f/ sits above it and is simply not transmitted. This is not a limitation you can out-articulate, and it is the reason military and aviation radio invented the NATO alphabet rather than asking people to speak more clearly.
Modern voice-over-IP is wider than that, but video calls add two new problems. Packet loss deletes short fragments of audio, and the shortest fragments in speech are exactly the stop bursts and final consonants. Aggressive noise suppression, the feature that removes your keyboard and your fan, also treats quiet high-frequency sounds as noise, and takes some of your fricatives with it.
And you lose the face. Watching a speaker's mouth is worth a real amount of intelligibility in noise, because /p b m/, /f v/ and /θ/ are easy to see and hard to hear. On an audio call, or with the camera off, that help is gone and the acoustic signal has to do all the work.
Which Contrasts Break, and What to Do
| Contrast | What happens in noise | What to do instead |
|---|---|---|
| /f/ vs /θ/ vs /s/ (free, three, see) | Collapses first. All three live above 4 kHz at very low energy. | Do not repeat the word louder. Swap it, spell it, or add a defining word: 'three, the number'. |
| /p/ vs /t/ vs /k/ (pat, tat, cat) | Place of articulation blurs; you hear that a stop happened, not which one. | Aspirate hard at the start of a stressed syllable. The puff of air survives when the burst does not. |
| final /t/ vs /d/ (bet, bed) | Usually gone. Final stops are short, quiet and often unreleased. | Lengthen the vowel before a voiced ending, and actually release the final consonant. |
| /m/ vs /n/ (may, nay) | Nearly identical in noise; nasals carry almost no high-frequency detail. | On a phone, never rely on it: say 'M for Mike', 'N for November'. |
| plural and third-person /s/, /z/ (file, files) | Inaudible. A single /s/ after a consonant is the quietest thing in the sentence. | Carry the number somewhere else: 'two files', 'both of them', 'he does it every day'. |
| thirteen vs thirty, fifteen vs fifty | Misheard constantly. The contrast is stress plus a final /n/, and noise removes both. | Say the digits: 'one three' and 'three zero'. Then confirm by reading back. |
| /iː/ vs /ɪ/ (sheep, ship) | Survives better than any consonant. Vowels are loud, long and low-frequency. | Lean on it. Keep the length and quality difference clear and it will get through. |
| stress and rhythm (REC-ord, re-CORD) | Survives almost everything. It is carried by duration and pitch, not by high frequencies. | Make this your main tool: stress the key word instead of raising your volume. |
What Native Speakers Do Automatically
When a native speaker moves into a noisy room, their speech changes in a measurable, well-documented way that researchers call clear speech. Almost none of it is volume. They open the jaw wider, which spreads the vowels further apart and makes them easier to tell apart. They lengthen vowels. They pause more, and longer, at phrase boundaries. They widen their pitch range. They release final consonants that they would normally swallow. And they drop some of the reductions and blending that make casual speech efficient.
Getting louder without those changes actually makes things worse. Shouting raises your pitch, compresses your vowel space and distorts your voice, so you end up with a signal that is bigger but less distinct. This is why the person shouting at the next table is still not being understood.
Seven Changes That Work
- Open your jaw on the vowels. Vowels are what gets through, so give them room. This single change does more than anything else on this list.
- Release your final consonants. In quiet, relaxed speech you can leave the /t/ in "what" unreleased and nobody minds. In noise, that /t/ is the difference between "what" and "wha", so let it go.
- Pause at phrase boundaries, not between words. Silence between thought groups gives your listener time to process what they half-heard. Pausing between every word destroys the rhythm cue, which is one of the few cues still working.
- Widen your pitch range instead of your volume. Pitch movement survives noise; extra loudness mostly does not.
- Drop some reductions. This is the one situation where connected speech works against you. Say "and" rather than /ən/, do not blend "want to" all the way to "wanna", and keep function words a little fuller than usual.
- Never leave the key word until last. Utterance-final words are quiet and often trail off. Put the important word where you can stress it, and if it must go at the end, add something after it: "at three, the number three".
- Choose words with fewer competitors. "One five" instead of "fifteen". "Confirm" instead of "check". Longer words with more syllables are easier to recognise from partial information than short ones, because there is more left over when half of it is destroyed.
Numbers, Letters and Names: the Worst Cases
Some material is fragile no matter how well you speak it, and knowing which is a practical skill of its own.
- The rhyming letters. B, C, D, E, G, P, T, V and Z all share the same vowel and differ only in a brief consonant. In noise they become one sound. Use "B for Bravo" or any common word; you do not need the official NATO list, you just need a word.
- M and N. Nasals carry almost no high-frequency information, so this pair is nearly impossible on a phone. Never spell anything important without a word attached.
- The teen and ty numbers. Thirteen and thirty differ by stress and a final /n/, and noise erases both. Say the digits, then confirm by having the other person read it back.
- Names and email addresses. Assume they will fail. Spell with words, spell slowly, and send the written version afterwards if it matters.
Hear the Fragile Pairs
These six words are three of the most fragile contrasts in English. Listen closely in a quiet room first, then try them again with a fan or some background noise playing.
A Drill You Can Run at Home
You can simulate the whole problem in fifteen minutes, and it is far more useful than practising in silence:
- Add noise. Play café or crowd noise from a video at a moderate volume, loud enough that you have to lift your voice slightly.
- Record something fragile. Read a list of numbers, names, email addresses and street addresses out loud while the noise plays.
- Transcribe yourself tomorrow. Wait a day so you have forgotten the content, then listen and write down what you hear. Anything you cannot decode is something your listener could not decode either.
- Re-record with the seven changes. Same list, this time with an open jaw, released finals, phrase pauses and fewer reductions. Compare the two transcriptions rather than the two feelings.
A quicker version: record a voice memo, then play it back through your phone's speaker held at arm's length. That is roughly what your listener gets on a call.
The Bottom Line
Noise does not attack your accent, it attacks your consonants, your word endings and your high frequencies, and it leaves your vowels, your stress and your rhythm standing. So stop pushing on volume and start using what survives: open the vowels, release the endings, pause between phrases, stress the key word and spell anything that matters with whole words. Train the fragile contrasts with minimal pairs, then combine this with the conversational side in the five repair moves and speed and clarity.