ELSA vs Speak vs Loora for English Speaking Practice

Fast verdict by learner job

There is no defensible “best overall” here. These apps organize practice differently, and the current evidence is much stronger for what each product offers than for how reliably each AI handles the exact same difficult prompt.

• Start with ELSA if pronunciation is the bottleneck. Its current product pages put pronunciation, personalized speech feedback, Speech Analyzer, accent choices, and progress tracking near the center of the experience. That makes it the clearest pronunciation-first fit of the three. It does not prove ELSA’s scoring is universally accurate, accent-neutral, teacher-grade, or diagnostically valid.
• Start with Speak if you want the app to give you a structured speaking path. Speak documents speaking-focused lessons, AI conversation, real-time feedback, and a Premium Plus layer with personalized lessons and review built from mistakes. That is a strong fit for a learner who would rather follow a designed sequence than invent a topic every day.
• Start with Loora if you mainly want a daily English conversation routine. Loora documents a daily choice among open conversation, role-play, and Read and Talk, followed by a Practice Zone. Its free-access rule is unusually easy to understand: one full lesson before a subscription is required. The important caveat is privacy transparency: the help-center privacy link returned a 404 during this review, so first-party voice-retention and model-improvement details could not be verified from that policy.
• If desktop practice is non-negotiable, none of these is a clean fit today. ELSA says its learner app is for iOS and Android phones/tablets rather than browsers or computers; Speak says its app is not available on desktop PC; Loora says it is mobile-only and not yet available on PC, Mac, or web browsers.
• If the conversation is high-stakes, emotionally sensitive, or professionally consequential, do not confuse AI practice with a human tutor. AI can give you repetitions without social pressure; it cannot be assumed to notice every pragmatic, cultural, or interpersonal problem. See AI Speaking Practice vs a Human Tutor for that separate decision.

Evidence state: the product capabilities in this matrix are documented by current official sources; the “best-fit learner job” row is bounded editorial inference. None of the three earned a directly observed score for the five controlled spoken tasks in this review.

Disclosure and test date

Checked date: 2026-08-20. We reviewed current official product pages, help centers, privacy pages, cancellation instructions, and US app-store listings. Prices, free/trial rules, platform compatibility, subscription packaging, and AI features can change quickly; re-check the exact screen offered in your country and app store before paying.

The attempted comparison context was deliberately the same for all three products: English as the learning target, desktop web, public/signed-out access, no account, no active subscription, no trial, and no microphone permission. That made the comparison fair in one sense and limited in another: public documentation could be checked consistently, but the account-gated mobile speaking experiences could not be put through a lawful identical spoken benchmark without creating accounts, installing/using app flows, beginning a trial, or granting microphone access.

This matters because “the vendor says it has role-play” and “we watched it handle an interruption correctly” are different evidence. This article keeps them different.

• Directly observed in current test: only public-page availability and access constraints; no load-bearing spoken task result.
• Documented by current official source: product features, plan rules, platform support, cancellation paths, and privacy statements explicitly published by the vendor or app store.
• Bounded editorial inference: which learner job appears to match those documented mechanics best.
• Not observed/inaccessible: actual pronunciation-scoring quality, interruption handling, conversational follow-up quality, deliberate-error handling, and obedience to an “exactly one correction” instruction.

Test methodology

A useful AI speaking comparison needs adversarially boring consistency. So the planned benchmark used the same five tasks for every app rather than giving each vendor a flattering demo prompt.

The identical task set

1. Minimal pair/pronunciation: Say: “I said ship, not sheep.” Then tell me which vowel contrast you heard and give one retry target.
2. 60-second spontaneous answer: Speak for about 60 seconds: “Describe a difficult decision you made recently, what options you considered, and what you learned.” Do not interrupt until I finish.
3. Role-play with interruption: You are a hotel receptionist. I am a guest whose reservation is wrong. After my first full sentence, interrupt me once with a short clarification question, then let me continue the role-play.
4. Deliberate grammar error: Yesterday I go to a meeting and discuss about the budget.
5. Exactly one correction: Give me exactly one correction from what I just said. Do not add a second correction or extra language advice.

So this is not a hidden “we tested all three for a week” article. Where a section discusses quality, it distinguishes the mechanism a vendor documents from the quality we did not observe. That limitation is especially important for correction accuracy and conversational follow-ups, where a marketing sentence is weak evidence.

Pronunciation feedback

If your main problem is “people do not understand the sounds coming out of my mouth,” ELSA has the clearest documented specialization. Its current site describes real-time personalized feedback, a Speech Analyzer, pronunciation guidance, accent choices, and progress tracking. The current US App Store description also centers pronunciation training and detailed speech feedback. That is more pronunciation-specific positioning than the core lesson structure we found for Speak or the conversation-first lesson loop Loora documents.

Speak still treats pronunciation as part of speaking practice. Its current product materials promise real-time feedback and speaking-focused lessons; the point is that pronunciation sits inside a broader learn-practice-apply path rather than being the sole organizing idea. If you want “teach me what to say, make me say it, then use it in a conversation,” that may be a better behavioral fit than opening an app primarily to polish sounds.

Loora documents pronunciation feedback during its Daily Conversation and Role-play, and its key-takeaway system can select a pronunciation issue and present focused practice. That makes pronunciation present and actionable, but the product’s daily rhythm is still conversation-led: choose a conversation/role-play/read-and-talk lesson, speak, review corrections, then practice.

The missing evidence is the most important part: we did not run the “ship/sheep” sample in any of the three apps. We therefore cannot responsibly say which one heard the vowel contrast most accurately, which tolerated a particular accent better, or which gave the most useful retry cue. A numerical speech score should be treated as app feedback, not as a clinical, teacher-grade, universally accurate, or accent-neutral diagnosis.

If pronunciation is your entire buying question, use the narrower Best English Pronunciation and Accent Apps comparison rather than forcing this three-app page to answer a broader pronunciation market question.

Official sources checked: ELSA product page, Speak product page, and Loora Daily Conversation help.

Open conversation and follow-ups

“Has AI conversation” is now a weak differentiator. The useful questions are whether you can choose the topic, whether the AI can stay in role, whether it asks sensible follow-ups, and whether you can recover when a conversation goes sideways.

ELSA documents AI role-plays and open-ended dialogue, and its ELSA AI help page says learners can use suggested scenarios or create a scenario. That gives it more conversational flexibility than a pronunciation-drill-only description would suggest.

Speak documents AI conversation as a core feature, while Premium Plus adds the full Speak Tutor experience and personalized lessons. This is an appealing structure for someone who wants both guided material and a place to apply it. What we do not know from this test is how reliably Speak Tutor follows an unusual instruction mid-conversation.

Loora is the most explicitly conversation-centered of the three in its current help materials. Daily Conversation allows any topic; Role-play offers prepared and custom situations; and the vendor says the AI responds during the exchange while surfacing corrections. If your success metric is “Did I actually talk for 15 minutes today?”, that design is easy to understand.

But none of that answers the hardest part of this section: follow-up behavior was not directly observed. The controlled role-play asked the AI to interrupt once, ask a short clarification, and then let the learner continue. That task was inaccessible under the no-account/no-trial/no-microphone test boundary for all three. We therefore do not award a winner for interruption handling, conversational memory, or prompt obedience.

A useful buying test, if you later try one yourself, is to deliberately derail the conversation: correct the AI, change your mind, ask it to slow down, and then return to the original role-play. A polished first turn tells you very little about repair.

Lesson structure

This is where the products separate more cleanly than they do on “AI” branding.

ELSA: personalized speech training with multiple practice modes

ELSA’s current individual offering documents a personalized learning path, role-plays, instant feedback, native-language support, Speech Analyzer, and progress tracking. For a learner who likes diagnostics, targeted retries, and visible skill areas, that is coherent. It can also be more machinery than you need if your only goal is to talk freely every day.

Speak: curriculum first, AI inside the curriculum

Speak’s product page positions speaking lessons, AI conversation, and real-time feedback as the core experience. Its paid packaging adds an important distinction: Premium Plus includes unlimited personalized lessons, custom review built from mistakes, “Made for You” lessons, and the full Speak Tutor. In other words, the higher tier is not just “more minutes”; it changes how much personalization and tutor-style practice the vendor says you get.

Loora: one daily lesson, then a practice area

Loora’s structure is simpler to explain. You choose one of three Daily Lesson types: Daily Conversation, Role-play, or Read and Talk. Current pricing/help pages say paid plans include one daily lesson and an unlimited Practice Zone after completing it. That can be attractive if elaborate course trees make you procrastinate. The trade-off is that “simple daily routine” and “carefully sequenced curriculum” are not the same learner job.

Bounded editorial judgment: Speak is the cleanest match when you want an app to sequence the learning for you; Loora is the cleanest match when the habit itself is “have an English conversation every day”; ELSA is the cleanest match when the sequence should keep pulling you back toward speech/pronunciation issues. Those are fit judgments from documented structures, not measured learning outcomes.

Correction quality

This is the section where comparison articles are most tempted to bluff. All three vendors document feedback. That does not mean all feedback is correct, equally useful, or equally well timed.

ELSA documents instant personalized feedback in lessons and detailed speech analysis. Speak documents real-time feedback and, on Premium Plus, review built from learner mistakes. Loora documents highlighted corrections during Daily Conversation, pronunciation/grammar feedback, and a post-lesson key takeaway that focuses on one mistake.

Those are correction mechanisms. The current controlled test did not establish correction quality. We did not feed the three apps the same deliberate error — “Yesterday I go to a meeting and discuss about the budget” — and compare whether each fixed tense, verb pattern, naturalness, or everything at once. Nor did we test the harder instruction: “Give me exactly one correction.”

That distinction matters because a learner can be harmed by two opposite failure modes:

• Under-correction: the system lets a recurring error pass, so the learner assumes it is fine.
• Correction overload: the system turns every sentence into a red-pen explosion, so the learner stops speaking naturally.

Loora’s documented “key takeaway” is interesting because the product explicitly narrows post-lesson attention to one priority issue while still making other feedback available. ELSA and Speak also document personalized feedback/review, but this review did not directly compare which system chooses the right correction to prioritize.

If a correction looks surprising, verify it with a trusted reference or human teacher before building a new habit around it. AI can be useful for volume; high volume does not magically turn every suggestion into good language pedagogy.

Beginner support

Beginners need more than a microphone button. They need comprehensible instructions, a way to recover when they freeze, and enough structure that “talk about anything” does not become “say nothing for three minutes.”

ELSA has the clearest explicit bilingual bridge in the current sources. Its product page says beginners can start conversations in their native language, and the current individual plan lists learning in your native language. For a learner who is not ready to process every instruction in English, that is a meaningful design choice.

Speak has broad native-language support for learning English. Its current help center lists English courses for speakers of Korean, Japanese, Traditional and Simplified Chinese, Spanish, French, German, Italian, Portuguese, Polish, Vietnamese, Hindi, Thai, Turkish, and Indonesian. That does not guarantee every interface element or AI response is equally localized, but it does show that Speak’s English-learning product is deliberately organized for multiple language backgrounds.

Loora says its conversation adapts to your level and positions the service for adult English learners from beginner upward. Its lesson choices can reduce blank-page anxiety: pick a role-play or a reading rather than inventing a topic. Still, the official material we reviewed gives less emphasis to native-language scaffolding than ELSA or Speak.

One non-negotiable age detail: ELSA’s privacy notice says people providing personal data through the service represent that they are 16 or older; Loora’s help center says users must be 16 or older; Speak’s general privacy notice says its services are not intended for children under 13. App-store content ratings are not substitutes for those service-policy statements.

Motivation and progress

“Motivation” is often a polite word for “Will I open this app on Thursday when I am tired?” The three products attack that problem differently.

ELSA documents visual progress dashboards and gamified learning. That suits learners who respond to visible skill areas, scores, and targeted goals. Just remember that an in-app score is a product metric; movement in that metric does not automatically prove equal improvement in spontaneous real-world conversation.

Speak’s motivational advantage is more curricular: a speaking-focused path gives you a next step, and Premium Plus documents custom review built from mistakes plus personalized lessons. That can reduce the daily decision burden. The risk is the familiar subscription-app trap: paying for a sophisticated path does not create the habit of using it.

Loora’s habit loop is the simplest: choose one Daily Lesson, get corrections and a key takeaway, then use the Practice Zone. Its progress help page says progress is tracked through conversation rather than tests or quizzes, across vocabulary, pronunciation, grammar, and fluency. That may appeal to learners who dislike formal testing, but those measures are still the vendor’s internal model of progress, not an independent proficiency exam.

When comparing motivation features, ignore the streak fireworks for a moment and ask a boring question: Which app makes your next 10 minutes obvious? The product that reduces friction for your actual goal is often more useful than the one with the longest feature list.

Price/free tier

Price check: 2026-08-20. Subscription screens vary by country, promotion, billing channel, eligibility, and legacy offer. The US App Store can also expose multiple in-app-purchase SKUs for the same nominal billing period. The figures below are therefore examples visible in the US Apple listing on the check date, not a promise of what your account will see.

Two commercial details change the practical meaning of “free.” Speak’s trial is broad but asks for payment details up front. Loora’s free lesson is narrow but does not require payment. ELSA documents both a free individual offer and a premium-trial prompt, but the public pages we reviewed do not give a clean universal cap that would justify writing “you get exactly X minutes/lessons per day.”

That is why this page refuses to calculate a fake “cost per speaking minute.” Usage behavior, local offers, and plan gates vary too much. Compare the offer you can actually buy, set a calendar reminder before any auto-renewing trial ends, and keep the confirmation screen or receipt.

Official pricing sources: ELSA subscriptions, ELSA US App Store, Speak product page, Speak trial help, Speak US App Store, Loora pricing help, and Loora US App Store.

Privacy/voice data

A speaking app cannot do much without processing speech, so “does it use audio?” is the wrong starting question. Better questions are: what does the current policy say audio is used for, who may receive it, how long data may be retained, how account deletion works, and can you actually read the relevant policy before subscribing?

Apple’s App Privacy panels are useful but limited: Apple explicitly says the disclosures are supplied by the developer and have not been verified by Apple. They also summarize categories rather than replacing the vendor’s full privacy notice.

The Loora policy-link failure deserves weight because privacy visibility is part of the buying decision. It does not prove that Loora has no policy, that no data is retained, or that the policy is broken for every user. It means exactly this: on 2026-08-20, the privacy link exposed in the official help-center footer resolved to a URL that returned 404 to our verification request, so load-bearing first-party voice-retention details remained unverified.

If you plan to discuss confidential work, health, legal, financial, or client material in any AI speaking app, do not treat a casual practice conversation as a private room by default. Read the current policy for your region, avoid unnecessary sensitive details, and use device microphone controls when you are finished.

Which app to choose

Use your bottleneck, not the biggest feature count.

Choose ELSA when the sentence is understandable in your head but not in your mouth

ELSA is the strongest documented match for a learner who wants to zoom in on speech production: pronunciation, word stress, speech analysis, targeted retry, and a personalized path. Its AI role-plays mean you are not limited to isolated sounds, but pronunciation remains the clearest reason to start there.

Do not choose it because you saw a precise score and assumed the number was objective truth. This review did not independently validate ELSA’s acoustic scoring against teachers or across accents.

Choose Speak when you want a course that keeps making you speak

Speak makes the most sense when you want the app to decide what comes next: lesson, spoken practice, feedback, and conversation. Premium Plus’s documented personalized lessons and mistake-based review strengthen that fit for someone willing to pay for deeper personalization.

Do not choose it because “AI Tutor” sounds equivalent to a human tutor. We did not test nuanced follow-up, interruption handling, or the quality of one-correction prioritization, and those are exactly the behaviors that become important in messy real conversation.

Choose Loora when your problem is simply not talking enough

Loora’s current structure is easy to turn into a habit: one daily conversation/role-play/read-and-talk lesson, then Practice Zone. It is a strong documented fit for an adult English learner who wants conversational reps with corrections rather than a large course tree.

Do not ignore the privacy documentation gap. The broken first-party privacy link we encountered does not automatically make the product unsafe, but it lowers the evidence strength for a privacy-sensitive buyer until the current voice-data terms can be read and verified.

Choose none of the three when the constraint points elsewhere

• You need a full desktop/browser speaking workflow today.
• You need a human to judge politeness, persuasion, interpersonal dynamics, or professional stakes.
• You need independently validated high-stakes scoring rather than app feedback.
• You are under the service’s applicable age threshold.
• You cannot verify the subscription, cancellation, or privacy terms shown for your country.

If you are still choosing at the category level, go to Best AI Apps for English Speaking Practice. If your shortlist includes Duolingo rather than ELSA or Loora, the separate Speak vs Duolingo for AI Learning comparison owns that decision.

Where FunFluen fits and does not

FunFluen should not be smuggled into a competitor comparison as a surprise “winner.” Its speaking route solves a different decision: choosing a speaking-practice workflow around your goal, including routes that connect practice to language you encounter in media and other contexts.

On 2026-08-20, the FunFluen speaking chooser loaded publicly in a signed-out browser and presented speaking-practice routes. The page also exposes sign-in/get-started controls. From that public check alone, we do not claim that every downstream exercise is no-signup, free, unlimited, AI-scored, or available in every context. We also do not claim FunFluen is a human tutor, an official pronunciation test, or a replacement for ELSA’s pronunciation specialization, Speak’s structured curriculum, or Loora’s dedicated daily AI conversation routine.

If your problem is not “Which subscription app is best?” but “What kind of speaking practice should I do next?”, you can try the speaking workflow that fits your goal. You do not need that link to use the comparison above.

For the larger market view, see the parent comparison below. This page intentionally stays narrow: ELSA vs Speak vs Loora, current commercial facts, and the evidence boundary around what was and was not tested.