Facts, studies and product details checked in November 2026.
Give learners the accent they will hear outside the classroom. If they will live in Mexico, work with a team in Texas or travel across Latin America, Spain's Spanish is the wrong default: about nine in ten native Spanish speakers live outside Spain, and Mexico alone has 130 million of them. Start with one clear accent of that place. As learners progress, add others, because that is what real listening and the big exams demand: IELTS and TOEFL both play British, Australian, New Zealand and North American speakers. And remember that an accent is not a defect. Research has shown for thirty years that a strong accent can still be fully understood, and that listeners get used to a new one within a minute.
In text to speech, the accent comes with the voice. A voice built for one country speaks with that country's accent. A voice that speaks many languages is convenient, but it may read all of them with one accent. In Kukarella you choose the language and the country first, and the voices built for that country come first. But you still have to listen: the sample is a fixed recording, not your text, and hearing your text in four accents uses credits four times. Each section below says what holds for any tool, then what to do in Kukarella.
In this guide:
- Does an accent make a recording harder to understand?
- Which accent should learners hear first?
- Hear it: one English sentence in four accents.
- Spanish: Spain, Mexico or Argentina?
- French and Portuguese.
- One accent or several?
- Can text to speech speak English with a foreign accent?
- Which voices keep the accent you picked?
- How to get an accent in Kukarella.
- Frequently asked questions.
Does an accent make a recording harder to understand?
Less than most people think. In 1995, Murray Munro and Tracey Derwing separated three things we usually lump together: how strong an accent sounds, how easy it feels to follow, and how much the listener actually understands. The three are related, but only partly. Looking back on their work in 2020, they put it in one line: speech "can be heavily accented but highly intelligible".
An unfamiliar accent does cost something at first. When Constance Clarke and Merrill Garrett played Spanish- and Chinese-accented English to native English-speaking listeners in 2004, the listeners were slower at first, and the delay shrank within one minute, after as few as two to four sentences. A 2018 study repeated it online and found that the listeners also understood a new speaker with the same accent better.
That first minute can still be awkward. In the clip above, a reporter from Yorkshire asks the tennis player Alexander Zverev a question in English, and Zverev, who gives his interviews in English, cannot make out the question.
Learners feel that awkwardness more than their scores show. In a 2024 study, 79 Japanese adults took a listening test with passages in American, British and Japanese-accented English. Their scores did not differ by accent, but they rated the British and Japanese accents harder to understand than the American. In a Swedish study of 160 secondary-school students, what students understood did not match which accents they said they liked.
A patient on a psychiatrist's couch: 'I don't mind the voices themselves, doctor. It's the Jersey accents that are driving me nuts.'So an accent learners have not met is tiring before it is a barrier. That is a reason to let them meet it in class, not a reason to avoid it.
Which accent should learners hear first?
The accent of the people they will actually talk to. Decide where the learners will use the language, and with whom, and start there. The numbers show why this matters more for some languages than others.
| Language | Where its speakers are | Source |
| Spanish | 520 million native speakers; Mexico 130 million, Spain 47 million | Instituto Cervantes, 2025 |
| French | 396 million speakers; 65% of them in Africa | OIF, 2026 |
| English | about three in four speakers learned it as a second language | Ethnologue figures |
Three things follow.
- "The original" is rarely the right default. About one native Spanish speaker in ten lives in Spain. Castilian Spanish is right for learners going to Spain; for most others, a Latin American accent is closer to what they will hear.
- A second-language majority changes the target. Most people who speak English learned it as a second language, and many of them use it mostly with others who did too. The Council of Europe's 2020 update of its language framework (the CEFR) judges pronunciation by how easy it is to understand, not by how close it is to a native accent.
- Where you live is not where they will work. A business-English teacher I spoke to this autumn picks a very clear accent for her beginners. When her students have a negotiation with someone from India coming up, she gives them Indian English, so that they can tune their ear before the meeting.
Many of our users tell me the same thing in other words: they came to Kukarella for one voice in one accent. Not the voices I rate highest, but a Canadian French voice, a Mexican Spanish voice, a Brazilian Portuguese one, because that is how their listeners speak.
Hear it: one English sentence in four accents
Here is one sentence, written to show the differences: "Can you park the car near the water tower? I'll be there after the dance class." Four voices read it, each built for its country. Listen for the r in park, car and tower, the t in water, and the vowel in dance and class.
The United States: Aria.
English (US): Aria
The United Kingdom: Libby.
English (UK): Libby
Australia: Annette.
English (Australia): Annette
India: Neerja.
English (India): Neerja
The biggest single difference is the r. American English pronounces it after a vowel, in car and park; standard British and Australian English drop it there and keep it only before a vowel. That one rule already makes the two easier to tell apart.
Spanish: Spain, Mexico or Argentina?
Spanish shows how much an accent and a script differ. A voice can change how words sound; it cannot change which words you wrote. Here is what to listen for, and what to write.
| Feature | Spain (most of it) | Latin America |
| z, ce, ci | like English th: caza differs from casa | like s: caza sounds like casa (seseo) |
| ll and y | like English y | in Buenos Aires and Montevideo, like sh or the s in measure |
| "you" in the plural | vosotros among friends, ustedes formally | ustedes for both |
| "you" to one friend | tú | tú; in Argentina, Uruguay, Paraguay and much of Central America, vos (vos sabés) |
The last two rows are in your text, not in the voice. A Mexican voice reading vosotros sabéis will say it, and it will sound wrong to a Mexican listener. Write the script for the place, then choose the voice for the place.
Here is one sentence with z, c and ll in it: "Hace cinco años, mi hermana cocinaba en una ciudad llena de calles tranquilas y cielos azules."
Spain: Abril.
Spanish (Spain): Abril
Mexico: Candela.
Spanish (Mexico): Candela
Argentina: Elena.
Spanish (Argentina): Elena
Is one of them "correct"? No. The Spanish exam DELE accepts any educated variety of Spanish from a candidate, as long as it is used consistently. Its level descriptions follow a sensible path: at B1, a learner understands standard Spanish that is not too local; at B2, several varieties; at C1, any variety at all. It is a good order for a course, too.
Part of Kukarella's language picker with Spanish open: Mexico, Spain, the United States, Peru, Uruguay and more.The picture above is part of the list you see when you choose Spanish in Kukarella: one entry for each country. But the large accents have many native voices and the small ones only a few. For Argentina, Colombia or Peru you may hear every native voice in a minute and still not find the one you want.
French and Portuguese
French. Most French speakers live in Africa. If your learners will work in Quebec, they need Canadian French. In Quebec three people in four have French as their mother tongue, and one sound marks it at once: t and d before i and u become "ts" and "dz", so tulipe sounds like "tsulipe". Kukarella lists French by country: France, Canada, Belgium and Switzerland. But there is no separate African accent of French in the library today.
One organisation that trains its staff with us chose a single Canadian French voice for its training narration and keeps it. For them it is the organisation's voice. A language teacher needs the opposite: a different speaker, sometimes a different accent, in the next lesson.
Portuguese. There are nearly nineteen Brazilians for every Portuguese person, so for most learners the default is Brazil. For learners going to Portugal, or reading texts written there, choose Portugal. Kukarella has both, but Portugal has fewer voices than Brazil.
One accent or several?
One first, then several. The research on listeners points that way, with one limit worth knowing.
- Several speakers help with a new speaker. Listeners who heard several Chinese-accented speakers understood a new Chinese-accented speaker better than those who heard only one (Bradlow and Bent, 2008).
- Several accents help with an accent never heard. After training on speakers of five language backgrounds, listeners understood new speakers better, including speakers of accents not in the training (Baese-Berk and colleagues, 2013).
- For learners, variety in training works and lasts. A 2025 review of 79 studies of second-language learners found that training with many voices improved what they heard, and the gains remained.
The limit: the first two studies tested native English listeners, and the third trains single sounds, not whole conversations. I have not found a study that gives learners several national accents and measures whether they then understand a new one better. The exams have decided anyway. TOEFL added British, New Zealand and Australian speakers to its North American ones in 2013. IELTS uses British, Australian, New Zealand and North American speakers. Cambridge tells teachers to prepare candidates with "a range of voices, accents".
So here is the order I suggest, and it is my advice, not a study's: one clear accent at the start, a second one when learners can follow the first without effort, and at the advanced level, whatever they will meet. The tool below turns that into a plan for your learners.
If your learners are learning English, the quiz below plays one sentence in three accents and asks which is which. Two minutes, and it shows them how much they already hear.
Can text to speech speak English with a foreign accent?
Sometimes, and you have to listen to find out. People search for "English with a Russian accent" or "with an Indian accent". Indian English is easy: it is an accent of English, and voices are built for it. A Russian, French or Spanish speaker's English is different. No voice is labelled "English with a Spanish accent". The only way to get one is to give English text to a voice built for another language.
I tried it. I asked fourteen voices recorded in six other languages to read the same English paragraph. Some read it with no foreign accent at all. Most of the rest had an accent so strong that I could hardly follow them, and one I could not follow at all. Only two sounded like a real person speaking English as a second language: one with a slight Chinese accent, one with a strong French accent that was still easy to understand.
A trainer who prepares people for jobs abroad asked me for exactly this: English spoken by a Portuguese speaker, with the accent kept. Today there is no filter for it. The only way to know is to listen to each voice, and I told her so.
A 2025 study of two popular voice services came to the same place from another side: they did not perform equally well across five English accents. Whatever the tool, test the accent before you give it to a class.
Which voices keep the accent you picked?
Voices built for one country keep it. Voices built to speak many languages often do not. In text to speech these are two kinds of voice, and they answer the accent question differently.
- A native voice speaks one language, with the accent of one country. You choose the accent by choosing the voice. It will not read another language well.
- A multilingual voice reads many languages. It is handy for a text that switches language. But it usually has one base accent. In my own listening, it sounds fine at first, and after about fifteen seconds you hear a slip in stress or pronunciation.
There are exceptions. I played one multilingual voice, Ada M., under English (UK) and English (US), and it said tomato, potato and garage differently; in Spanish it changed between Spain and Mexico. But in Spanish it is offered under Spain, Mexico and the United States only, and most multilingual voices change nothing.
When a voice starts speaking with the wrong accent and stresses the wrong syllables, listeners know at once that it is AI. For a lesson on accents, that is the one mistake to avoid.
How to get an accent in Kukarella
A teacher told me the hard part was not finding accents but finding only that accent: native and multilingual voices were mixed together, and she lost time listening to all of them. So we rebuilt the Voice Library around the accent.
- Open the Voice Library (in Text to Voice, Add voice) and click the language name at the top, for example English (US).
- Choose the language, then the country. Or type a country in Search languages and countries: Mexico, Ireland.
- Pick from the Native voices, which come first. Further down, where Multilingual voices begins, the list says it plainly: "They speak English too, but may not have this accent."
- Press ▶ to hear a voice's sample, then tick two or three voices and play your own text. A voice with a padlock plays its sample, but your plan does not open it.
Part of Kukarella's language picker with English open: Australia, Canada, India, Ireland, New Zealand and more.Three "buts" before you build a course on it.
- Each voice reads the whole text. Text to Voice has one voice per reading; there is no accent per paragraph. To compare four accents, tick four voices: you get four recordings, and you use credits for the text four times.
- Different accents in one dialogue need DialoguesAI. There each speaker gets a voice, so a British narrator can talk to an American student. But the assistant that casts the speakers does not choose an accent: open each speaker's card and pick a voice from the accent you want.
- Small accents have few native voices. The US and the UK have many; Ireland, Kenya or Uruguay have a handful. Listen to all of them before you decide.
Frequently asked questions
Is there a "neutral" accent in text to speech?
No. Every voice has an accent: a native voice has its country's, a multilingual voice its own base accent. What people call neutral is usually a standard accent they hear often, such as General American, or simply their own.
Which English accent is best for IELTS or TOEFL?
Both tests use several. IELTS plays British, Australian, New Zealand and North American speakers; TOEFL plays North American, British, New Zealand and Australian ones. Start with the one your course uses, and add the others before the exam.
Is Castilian Spanish more correct than Latin American Spanish?
No. Both are standard Spanish, and the DELE exam accepts any educated variety a candidate uses consistently. Choose the one your learners will hear.
Do learners understand their own language's accent best?
Sometimes. In some studies, learners understood English spoken with their own first language's accent better; in others they did not, and one study found it helped one group and, on balance, not another. Do not count on it.
Can I change a voice's accent with a setting?
No. In Kukarella the accent comes with the voice. Pitch and speed change how a voice sounds, not its accent. To change the accent, choose a voice built for another country.
Can one recording have two accents?
In Text to Voice, no: one voice reads the whole text. In DialoguesAI, yes: each speaker has a voice of their own.