ELSA Speak listens to your English and scores pronunciation, intonation, fluency, grammar and vocabulary, then drills the gaps through interview, meeting and presentation role-plays on a path built from your own level and goals.
Used this tool? Rate it
Used this tool? Rate it
ELSA Speak is an AI speaking coach for English: you talk, and the app scores what you said and tells you how to fix it. As an English pronunciation app the name stands for English Language Speech Assistant, and its premise is that the bottleneck for most learners is not vocabulary but the confidence and clarity to say things out loud. Every lesson ends in instant, personalized feedback on real-world English practice rather than a multiple-choice score.
Rather than one fixed syllabus, the AI identifies your strengths and weaknesses and builds a tailored study plan around them. Two learners opening the app the same day are unlikely to see the same next lesson.
The company was founded in 2015 and raised a $15 million Series B in January 2021, led by VI Group and SIG with Google's AI fund Gradient Ventures returning; the same reporting notes a partnership with IDP, an owner of the IELTS exam, which recommends the app for test preparation. Scale figures depend on who counts: the vendor's own marketing headline of 90M+ downloads and a 4.9 star figure sits above what either app store publishes, and the store aggregates have a disclosed sample.
Two products share the name: the consumer app sold to individuals, schools and employers, and a speech-recognition API sold to platforms wanting the scoring engine inside their own product, priced and documented separately.
Feedback goes beyond whether a word was recognised: pronunciation down to individual sounds, intonation as pitch and volume variation, fluency as pace against a target band between 100 wpm and 140 wpm, grammar broken out by error type such as word order and verb form, and vocabulary distribution across CEFR levels. The same metrics surface in the API documentation, the clearest published description of what the engine measures.
Role-play is organised as named scenario coaches rather than one generic conversation partner. Workplace, Interview, Dating, Meeting and Presentation coaches each frame a different kind of conversation, which matters because the register that counts as "good" differs sharply between rehearsing a presentation and making small talk.
For learners who find English-only practice intimidating, bilingual support supplies explanations in your native language, so a beginner is not forced into English-only instruction from the first minute — onboarding friction, not translation.
Speech Analyzer is a browser tool rather than part of the phone app, aimed at longer stretches of unscripted speech. It offers solo practice, where you speak freely or follow guided prompts, and group conversations, which can run alongside a Zoom or Google Meet call for feedback on how you actually spoke in a real meeting. It needs microphone permission and a one-time voice enrolment, and comes with the premium plan or can be bought on its own.
For employers and schools the differentiator is measurement, not lessons. The business track offers proficiency mapped to CEFR, IELTS and TOEIC so results stay comparable across markets, plus exportable reports, individual progress tracking, and optional add-ons for custom branding and a Business API connecting to an LMS or HRIS.
A practice session is only as good as the audio it is scored on, and the vendor is explicit about the conditions it expects: a quiet environment, a strong stable connection, ideally wired headphones, and avoid Bluetooth audio devices, which the vendor says are prone to disruptions and disconnections. A low score over a flaky Bluetooth headset in a noisy room tells you about the microphone, not your English.
The window is short and unrestricted, so test what you would actually pay for before it runs out:
The coaches map onto a real deadline:
The clearest fit is someone whose English reading and writing already carry their work but whose speaking lags: the professional who writes fluent email then goes quiet in meetings. The role-play framing, industry vocabulary and exam mappings point at that learner, not an absolute beginner.
Employers and schools are addressed as separate buyers with their own reporting, and the scale thresholds tell you where the vendor draws the line between self-serve and sales-led: ELSA Business is quoted from 50 learners at custom pricing, and the School track from 21 learners, also at custom pricing behind a contact-sales button. The Individual track is the only one with a figure on the pricing table, and that figure is Free with an upgrade available.
The first is anyone needing desktop practice. The app cannot be used on PC, laptop, BlackBerry or Windows phones, so a learner without a suitable phone or tablet is left with only the browser-based Speech Analyzer, a narrower slice of the product.
The second is anyone with strict expectations about who controls their data at work or school. Learners enrolled by an employer or school are explicitly outside the public privacy notice, so their data questions are answered by the institution's contract rather than anything published publicly. A normal enterprise arrangement, but worth knowing before assuming the published notice applies.
The phone app and the browser tool cover different ground, and neither replaces the other. The main app runs on Android and iOS phones and tablets, and one account can be signed in across several; desktop machines and Windows phones are out of scope. Speech Analyzer runs in a desktop browser and is described as best used there.
Everything requires connectivity. The app needs a strong, stable data connection to communicate with the vendor's server, so lessons cannot be downloaded for a flight or commute. Scoring happens server-side, which rules out the offline-practice pattern some language apps support.
Purchases are platform-bound. A subscription bought inside the app is an Apple or Google transaction; one bought on the website is an ELSA transaction, and the two follow different cancellation and refund rules.
What the free tier buys you is time, not a permanent allowance. Creating an account gives seven days of unrestricted access with no payment method required, after which the first two lessons of every module remain free. That is enough to evaluate the scoring quality but not enough to complete a learning path.
Beyond that the published picture thins out. Individual plans sell through the app and checkout, and the pricing page itself publishes no figures and no plan comparison table — only positioning copy and a trial call to action, so the site does not publish consumer prices. One structural detail is documented: lifetime memberships are distributed only through the vendor's own site and official distribution partners, not in-app, so the in-app price list is not the full set of options.
The API is the one part of the product with fully published prices, and it is billed per every 15 seconds of audio.
| API tier | Scripted | Unscripted | Max audio per request | Assessment scope |
|---|---|---|---|---|
| Standard | $0.008 | $0.02 | 20s scripted, 20s unscripted | Word, sentence and multiple-choice pronunciation; unscripted mode also adds spontaneous speech and grammar |
| Premium | $0.01 | $0.025 | 45s scripted, 150s unscripted | Adds fluency and intonation; unscripted mode also adds IELTS, transcript and vocabulary |
The tier difference is not only per-unit price. Standard caps a request at 20 seconds in both modes, workable for reading a sentence but not for a paragraph of free speech, and the assessment matrix splits by mode as much as by tier: spontaneous speech assessment and grammar scoring are marked available in Standard's unscripted mode, while IELTS assessment, the spoken transcript and vocabulary scoring are marked only for Premium unscripted. That gap also decides whether a platform can build IELTS speaking practice on top of the engine, since scoring open-ended answers against IELTS lands on Premium regardless of the headline rate.
For the consumer app, software directories list it beside AI conversation tutors, pronunciation reference libraries and speech-assessment vendors. A conversation tutor optimises for keeping you talking, a pronunciation reference for one correct rendering; ELSA sits closer to assessment, scoring what you produced against a model. For the API the comparison set is other speech-assessment vendors, where published per-second pricing and per-request audio caps matter more than lesson counts.
The billing terms are the part of this product most likely to produce an unpleasant surprise, and they are worth reading before the trial, not after.
One documented inconsistency is worth flagging rather than resolving: the section headed for prepaid renewals repeats the month-to-month instructions verbatim, so the terms state no distinct prepaid cancellation procedure at all.
Two operational limits round this out: practice requires constant connectivity, and scoring quality depends on the recording conditions above, making the product sensitive to a learner's environment in a way a reading app is not.
On the independent side the aggregate ratings are strong — 4.8 across roughly 110,000 iOS ratings and 4.6 across roughly one million Android reviews — while a business software review platform records 4.5 across 174 reviews. The negative end of that smaller sample is more informative than the average, though the visible reviews are too few to describe a distribution: among the ones on display, one reviewer reports paying £80 and being refused a refund after two days, another says the paid product was close to wasted money, and a third complains of constant bugs. Billing and stability both show up; the speech scoring itself does not.
Speaking practice means handing over recordings of your voice, and the licence attached to those recordings is broad. Submitting content grants the vendor a worldwide, royalty-free, irrevocable and transferable licence with the right to sublicense, covering voice recordings together with the pronunciation, intonation, volume, pace, grammar and vocabulary analytics derived from them. The privacy notice separately lists what is collected from individual users: prompts, text input, pictures, audio and video recordings, feedback and scenarios.
Where that data lives is disclosed. Processing spans the United States, Europe, Japan, India, Vietnam and Singapore, with AWS storage in Ireland, the US, Japan and Singapore.
Three further points affect specific users. When recording a conversation, the user, not the vendor, carries the duty to notify and obtain consent from anyone else on a recording — relevant if you run Speech Analyzer during a real meeting. The service is not intended for children under 13, and under-18s are told to use it only with a parent or guardian involved. And deleting an account is an email request to support rather than a button in the app, which is worth knowing if you want a clean exit.
There is a free tier, but it narrows sharply after the first week. A new account gets a full week of unrestricted access without entering a card, and once that lapses only the opening two lessons of each module stay open. Enough to judge the scoring, not enough to finish a path.
Not under the published terms. A prepaid term rolls into another term of the same length and bills the renewal fee unless the renewal is switched off, and the lifetime option is listed among those prepaid terms. Treat it as a long prepaid plan with auto-renewal and check the renewal setting after buying.
It depends entirely on where you bought it. A website purchase carries a seven-day refund window handled by the vendor. An in-app purchase does not: the vendor cannot refund it, so the request goes to Apple or Google and is decided under the store's own policy.
Send notice at least one working day ahead of the renewal, by email or through the store where you subscribed. Two traps deserve naming: removing the app from your phone stops no charges, and cancelling is not the same as asking for money back — you may need both.
No. Lessons need a live connection because the speech scoring runs on the vendor's servers rather than on the phone, and there is no downloadable lesson mode to fall back on.
Only partly. The main app is Android and iOS only, with desktop and Windows-phone use ruled out. Speech Analyzer, by contrast, is designed for a desktop browser, so a computer gets you the free-speech analysis but not the lesson curriculum.
They are licensed broadly. The content licence in the terms is worldwide, royalty-free, irrevocable and sublicensable, and it reaches both the recordings and the analytics derived from them. The privacy notice names the collected categories and places storage on AWS in four regions across Europe, North America and Asia.
Yes, through the separate speech API, which meters usage in 15-second units of audio. Check the tier boundary carefully: the lower tier limits a single request to 20 seconds and withholds the open-ended assessments, so scoring free speech means the higher tier.
