system
The System I Used to Fix My English Pronunciation and Speak More Confidently
I spent years feeling insecure about my English pronunciation. This is the step-by-step system I eventually used to learn the sounds, practice them, and become much more confident speaking English.
I'm a Spanish speaker. English at school was bad, so I learned on my own from about sixteen, mostly by watching videos and reading technical books. My comprehension got pretty good after a few years, but I couldn't say the same about the way I spoke English. I've always felt kind of insecure about speaking English because I knew I didn't know how to pronounce words correctly. I hesitated over the pronunciation of most words.
The usual answers didn't help. I tried listening and repeating for a while, and it never made that discomfort disappear. Living somewhere English-speaking is the other standard advice, and it wasn't really an option for me.
There was a moment last year when, after buying an iPad, I tried using dictation and it couldn't understand me at all. That, along with the fact that I wanted to start creating content in English, motivated me to find a way to actually solve the problem. So I started researching how to do it. That's how I came across the IPA, the International Phonetic Alphabet, which gives each sound its own symbol regardless of spelling.
Take the word "build," for example. A lot of Spanish speakers might pronounce it by mapping the spelling directly to Spanish sounds, but its IPA transcription is /bɪld/. This was quite a shock for me because it was only after learning the correct pronunciation that I could actually start hearing the /ɪ/ when listening to people say the word in videos.
The same thing happened with words like "of." I used to think of it as an "o" followed by an "f," but it's pronounced /əv/ (weak form), with a schwa followed by /v/.
This is important because when a sound doesn't exist in your native language, you may perceive it through the closest sound categories you already have. You can then repeat what you think you heard while actually reproducing a different sound.
There is quite a bit of research on how the languages we already know shape the way we perceive and produce sounds in a second language.
At first, I thought the solution would be relatively simple:
- Learn the English sounds I was missing.
- Learn how those sounds are used in actual words.
In practice, that wasn't enough to turn the idea into something I could systematically follow. There wasn't really a system or anything that could guide me through the whole process. After going through it end-to-end myself, the process eventually became four stages:
- Learn the phonemes: Define which phonemes you need to learn, then go through each of them, learning how they're represented and how to actually produce them (place and manner of articulation). This is the most technical part, and possibly the most difficult to get right.
- Practice the phonemes in individual words: Once you have an idea of how each phoneme is represented and pronounced, you can start practicing with words. Basically, this means producing each phoneme explicitly and very slowly, manually placing your jaw and tongue where they need to be.
- Learn a lot of words: Expose yourself to at least a thousand words and their pronunciations. Over time, you'll start generalizing and building a sense of how English words are pronounced.
- Practice full texts / connected speech: Take a text, such as a story, get its full IPA transcription, and read it out loud very slowly.
I went through the whole process myself, and my spoken English improved considerably. I can hear my own pronunciation much better now, and I'm able to regularly talk to AI in English using voice. What I noticed after going through the process was:
- My English feels much more fluid. Before, I tended to pronounce words more independently, with the more defined cadence I naturally use in Spanish. Now they feel more connected. Sounds flow into one another, and things like vowel reduction and the schwa make the transitions between words feel much more natural. I'm much closer to that continuous flow now than when I started.
- I can actually use dictation on my iPad now. It may sound like a small thing, but it used to annoy me that I had bought an iPad and its dictation could barely understand what I was saying.
- I feel much more confident speaking English. I wanted to start creating content in English, and this helped me a lot in that regard too.
Note: To make the process much easier and more efficient for myself, I built an app around the system, which is what I actually used, especially for the part about learning the sounds. The app isn't strictly necessary, though, so I'll also show you how to follow the entire process using free resources available online.
Learn the phonemes
The point here is to identify the English sounds that are different from the ones you already know, learn how they are represented, and learn how to physically produce them.
Which phonemes you need to focus on will depend on the languages you already speak. As a Spanish speaker, the ones I had to pay the most attention to were the vowels (/iː/, /ɪ/, /ɛ/, /æ/, /ɑ/, /ɔ/, /ʊ/, /uː/, /ʌ/, /ə/, /ɝ/, /ɚ/) and some consontants (/ɹ/, /ɾ/, /v/, /θ/, /ð/, /h/, /ŋ/, /z/. These may vary from language to language.
To get familiar with the IPA and the English sound inventory, Hadar Shemesh has two very good videos covering the vowels and consonants. You can also use an IPA chart as a reference:
Some of the phonemes will be hard to distinguish. I remember hearing the /ɑ/ in "not" and interpreting it sometimes as an "o" and sometimes as a "u." So this part is kind of tricky because you have to distrust your own perception and use the IPA chart and theory to understand how to actually make the sound.
I went back to this step many times while practicing the phonemes in individual words. This is not about "learn the phonemes, move on," but about identifying the sounds where you probably can't trust your first intuition, and learning how to actually produce them instead of producing what you think you hear.
One thing I found extremely useful was having a very detailed description of each phoneme. Not just what it sounds like, but how to actually produce it: where the tongue goes, how open the jaw should be, what the lips are doing, where the sound is articulated, etc.
You can find articles and videos explaining individual phonemes, or ask an AI for a detailed description. If you use AI, make sure to fact-check the output because it can get phonetic details wrong. You can use a prompt like this:
A very detailed description as if you were explaning someone how to make the sound, including how to position the tongue, the mouth, etc. You can use technical detail and go in depth.
/ɪ/ sound
By the end of this step, you should be able to look at something like:
/kæn aɪ pɑɹk ɪn hɪɹ?/
And kind know how to read it. The pronunciation may still be lacking (that will be addressed in the next step).
The important thing is that when you see a symbol like /ɪ/, you know what sound it represents and roughly what you need to do with your mouth to produce it.
Practice the phonemes in individual words
By this point I already more or less knew the phonemes, with some still not so honed-in.
This is probably the most important part of the whole process, and maybe the most tedious. Practicing the phonemes only in isolation isn't enough. You need to start producing them inside actual words.
This is in some sense listen and repeat, but not really. We are not trying to guess how to pronounce a word by listening to it. We already have its IPA transcription and know how each individual sound is supposed to be produced. The reference audio is there to compare against, not to figure out the pronunciation from scratch.
For a word like on, whose transcription is /ɑn/, I would first make the /ɑ/ sound and then the /n/, manually positioning my tongue and jaw the way I had learned in the previous step.
For consonants that were already in my native language, I usually wouldn't bother doing this and would just pronounce them normally. There is a bit of a compromise you can make here: some phonemes are drastically different or don't exist at all in your native language (/ɑ/ or /ɪ/ in my case), while others differ only slightly (like /l/ in my case). So there's a decision to be made about how far you want to go.
For a word like about, /əbaʊt/, I would manually and consciously make the effort of producing each part:
/ə//b//aʊ//t/
Yes, it's quite slow, and you have to do it for each word, and for a lot of words. But that's kind of the point: to teach yourself to produce the sounds the way you just learned instead of automatically falling back to the closest sounds from your native language.
If I tried to go faster, I would quickly go back to my default pronunciation: /ʊ/ would turn into /u/, /ɪ/ into /i/, etc.
For example, because I had never actually learned the sounds, I unconsciously mapped them to Spanish sounds. So communication (/kəmjuːnɪkeɪʃn/) sounded to me something like "comiuniqueishon." The last part, for instance, is not "shon," but /ʃən/ (or /ʃn/).
The raw process I followed was:
- Get the IPA transcription for a word.
- Review any phonemes you don't remember well. I would usually review them anyway, because the point is to produce the actual sound rather than rely on my potentially faulty memory of it.
- Listen to reference audio.
- Pronounce the word very slowly and record yourself.
- Listen to your recording and compare it with the reference.
- Adjust your pronunciation and repeat.
- Optionally, use a pronunciation assessment tool for additional feedback.
This is also where I really drilled some of the phonemes I couldn't produce well. With on, for example, I spent hours listening to the reference audio, recording myself, and experimenting with the position of my tongue and jaw.
You can do all of this for free. The problem is that the information and tools are scattered across the internet.
For IPA transcriptions and phoneme references:
For IPA transcriptions:
For reference audio, I personally used Azure Text-to-Speech. I don't really like Google Translate voices for this, and services like ElevenLabs are very good at natural speech but not necessarily ideal for learning individual pronunciations.
For additional feedback, Microsoft also has a pronunciation assessment tool. I wouldn't rely completely on automated pronunciation scores, but they can be useful as another reference.
You also need a corpus of words with enough variety to practice.
Since I needed all of this anyway, I figured I might as well put it into the webapp. I eventually made it into a game: learning a phoneme unlocks words, which then unlock sentences, and each phoneme has its own mission.
More importantly, this is what I actually used instead of constantly going through different sites for the transcription, phoneme information, reference audio, recording and feedback.
You can try one complete phoneme mission for free here
There you can get the IPA transcription, hear the reference audio, record and replay yourself, and get pronunciation feedback. Each phoneme is also tappable, so if I'm unsure about one, I can immediately go back and review how it's produced.
about
/əbaʊt/
[button to hear]
[recording + replay]
Note: This step is probably the longest, most time-consuming, and most exhausting. It's a lot of time spent repeating words slowly and trying to produce each sound the way it's actually supposed to be produced. I think I spent more than 30 hours here in total.
Learn a lot of words
This was the tamest step, because it was really just about exposing myself to words.
Having done the previous step, I was beginning to understand how to actually pronounce words in English. The problem was that I had maybe only practiced with a couple hundred words, and because pronunciation in English is quite irregular, that just wasn't enough.
So the next part was to simply go through a lot of words. The nice thing here was that I didn't need to do nearly as much deliberate thinking. After enough exposure, I started getting better at predicting the pronunciation of words I had never seen before.
So this was mostly just reading a lot of words, trying to guess the pronunciation, then seeing the correct answer, hearing it, and maybe saying it out loud. I aimed for at least a thousand words.
In a deliberate session, I was able to cover 1,000 words in about two hours. So a couple of hours will probably be enough, and you can do it during your commute or on your phone (people look at me kind of strangely when I was repeating english words on the bus).
You can take an Anki deck like this one and go through the cards.
There were a few transcriptions that seemed a little off, but overall the deck worked pretty well.
I didn't like all the extra information in the cards, though, so I ended up making my own simpler Anki deck with basically this format:
(front card) (back card)
|-----------| |-------------------|
| alarm | | alarm |
| | | BrE: /əˈlɑːm/ |
| | | NAmE: /əˈlɑːrm/ |
| | | [hear word] |
|-----------| |-------------------|
You can download it here.
I also made an index of the 3,000 most common words, which is useful as a reference.
Note: I'm not saying that 3,000 words are all you need to know. The point is that by going through a few thousand common words, you'll be exposed to a lot of the recurring pronunciation patterns in English.
There is one other thing I did here that worked pretty well: while walking the dog, I would open Samsung Notes and just start speaking in English about something. It didn't really matter what. The important part was to pay close attention to the pronunciation of each word.
Whenever I wanted to use a word but wasn't sure how to pronounce it, I would write it down and look up the transcription later.
This was very useful because it forced me to recall words while also exposing the ones I didn't actually know how to pronounce.
Practice reading full texts
So by this point, I know how to pronounce the sounds in English, and I probably know how to pronounce most common words, or at least make a much more accurate guess than I could have just a couple of months ago. However, if I try to speak at a normal speed, I still default to my usual way of speaking.
After practicing mostly with individual words and short sentences, what was left was to practice full texts.
This step is again a little slow, because you're back to reading slowly and manually producing the sounds, this time across thousands of words. You can feel kind of stupid doing it, but after this you can finally run free! (More or less).
So what I did was take a text and get its full IPA transcription. The text itself really doesn't matter. You can generate one with AI, take a story, or use pretty much anything you want.
Then the process is:
- Get the IPA transcription for the text.
- Read the IPA transcription very slowly, consciously producing each sound. At this point you're still training yourself to produce the sounds correctly, so going slowly is important.
- If a word or sound doesn't feel right, go back to its transcription or the individual phoneme and review it.
- Read the actual English text, trying to pronounce it correctly without looking at the IPA transcription.
You would end up with a collection of texts like:
I think about how we use time, and I know that people want a way to make things better, but we do not always see how.
There was a year when I had to work on my own, and I did not know what to do, but I said to myself that I would make it, and I would not give up.
I had some ideas, and I wanted to see if they could work, so I took them and made a plan, and then I went into it with all I had.
...
/aɪ θɪŋk əˈbaʊt haʊ wi juz taɪm, ænd aɪ noʊ ðæt ˈpipəl wɑnt ə weɪ tə meɪk θɪŋz ˈbɛtɚ, bʌt wi du nɑt ˈɔlweɪz si haʊ./
/ðɛr wʌz ə jɪr wɛn aɪ hæd tə wɝk ɑn maɪ oʊn, ænd aɪ dɪd nɑt noʊ wʌt tə du, bʌt aɪ sɛd tə maɪˈsɛlf ðæt aɪ wʊd meɪk ɪt, ænd aɪ wʊd nɑt ɡɪv ʌp./
/aɪ hæd sʌm aɪˈdiəz, ænd aɪ ˈwɑntɪd tə si ɪf ðeɪ kʊd wɝk, soʊ aɪ tʊk ðɛm ænd meɪd ə plæn, ænd ðɛn aɪ wɛnt ˈɪntu ɪt wɪð ɔl aɪ hæd./
...
You definitely don't need my app or any other app for this. You can use ChatGPT or any other AI to generate the texts, and tophonetics to get the IPA transcription. It also has a nice PDF download feature that I may eventually add to my webapp.
You can also my webapp to generate the transcription, but since it doesn't have the PDF feature yet, honestly, toPhonetics is better for this particular step.
Considerations
On phonemes and phones
IPA goes a lot further than what I'm using here, and you can go very deep into the exact way sounds are produced. The same phoneme can have different realizations (allophones) depending on the phonetic context.
Going too deep into this would make the system much harder to learn, so I mostly stayed at the phoneme level. There were a few cases where I went more specific, like the alveolar flap [ɾ] that often replaces /t/ in American English.
I wrote a separate article explaining how IPA works in more detail, but I haven't published it yet.
You have to actually go slow
When I say slowly, I really mean it. If it gets to the point where people hear you and think you have some kind of problem, that's good. I tried several ways to speed up, but I would immediately fall back into my default pronunciation and switch to my "Spanish reading mode." Only after spending a significant amount of time doing it slowly did speeding up become possible.
This is actually one of the reasons why I wouldn't recommend practicing with songs. Singing along can be fun, but you'll probably follow the speed of the lyrics, which makes it much more likely that you'll fall back into your usual way of speaking.
Tip for difficult vowels
For vowels that don't exist in your native language, it can help to start from one you already know how to pronounce and modify it. For /ɑ/, what helped me was starting with /u/ as in "food" and then experimenting with lowering my jaw, changing my tongue position, and relaxing the lip rounding until I got closer to the sound.
But, at least from what I could tell, there is no real hack. It's mostly a matter of trying different things and going back and forth between explanations, the IPA chart, and recordings, both your own and from a trusted source.
I also had to develop some control over my tongue and its position. One thing that wasn't initially obvious to me is that when vowel charts talk about tongue position, they're referring to the position of the body of the tongue, not the tip.
And so that's it.
It's a process that requires some dedication, and it's not something passive that you can do without thinking too much about it.
In the end, the essence of it is pretty simple: learn the phonemes, then practice them. But just saying that leaves out a lot of considerations that are quite important in practice.
I'm not saying that there aren't other methods, but in my case, this is the only one that worked and actually helped me improve my pronunciation. There are videos of me speaking English online if you want to judge for yourself.