Speak best alternative

A tray of live abalone, and a question you had not rehearsed

The ground floor of Jagalchi is tanks and tiled channels and a great deal of water. The woman running the stall is in waders. You have the sentence — how much for these — and you deliver it well, because you have delivered it about ninety times into a phone and the intonation is genuinely good.

She gives you a price, and then she asks whether you want it prepared here or taken away, and whether you are eating upstairs.

That is where it ends. Not because the words were hard; you know most of them. Because the exchange had branched, and every one of the ninety repetitions had been on a path with no branches. You had rehearsed a sentence. She was having a conversation.

Ninety seconds later you are pointing, she is laughing, the abalone is being prepared, and the transaction has succeeded on the strength of gesture. Which is fine. But it is not what eleven months of daily speaking practice was supposed to buy.

What Speak is genuinely very good at

It is worth being precise here rather than generous, because vague praise before a takedown is a rhetorical move and readers can smell it.

https://www.speak.com/ is built around getting specific sentences out of your mouth, repeatedly, with feedback on how they sounded. That is a real and underrated engineering achievement. Its speech recognition tolerates learner pronunciation without either flattering you or refusing to parse you, which is a narrow band that most products miss on one side or the other.

Two consequences follow, both good. First, your mouth learns the shapes — the physical business of producing sounds your first language does not contain improves fast, and that improvement is durable. Second, and more important for most people, it removes the initial reluctance to make noise in a foreign language, which for a very large number of adults is the binding constraint on everything else.

So the honest summary. If you can read a language and cannot make yourself speak it, Speak is close to the best purchase available and this article is not aimed at you. Come back in four months.

Where a drill engine runs out

A drill is a path with a known start and a known end. Its value comes from repetition, and repetition requires the path to stay the same. That is not a flaw in the implementation; it is what a drill is.

Real exchanges branch, and they branch early. Our own transcripts from markets, counters and ticket windows put the first unrehearsed branch at somewhere between the second and third turn, more or less regardless of language or country. That is well inside the range a drill can cover and well outside the range a drill prepares you for.

Three specific things go wrong at the branch, and they compound.

• The question arrives faster than the retrieval. You know the words for prepared and takeaway. You do not have them in under a second, and under a second is what the queue behind you is paying for.
• The drilled sentence is the wrong register. Practice sentences are complete and polite. The actual answer is two words and a gesture, and producing the complete polite version marks you as somebody who needs handling.
• Recovery has never been practised. When it goes wrong you have no move, because every previous attempt ended by completing successfully. Nothing in ninety repetitions taught you what to say after a failure.

None of that is fixed by more drilling. It is fixed by practising under conditions that can go off-path, which is a different product category.

Five alternatives, and what each is actually replacing

The mistake people make when leaving Speak is replacing it with something that does the same job slightly differently. If the drill engine was working, keep it; what you need is the layer above it.

Two of those are companions rather than replacements, and that is a legitimate answer. Keeping Speak for the mouth and adding something that can go off-script is cheaper in effort than a full migration and works well for people who are genuinely happy with their drilling.

How far past the script each one gets you

We ran a small, deliberately unfair test. Fourteen learners, each with at least three months on one product, given a scripted opening they had practised and then a series of unrehearsed follow-ups from a native speaker instructed not to slow down or simplify. We counted how many follow-ups they handled before the exchange collapsed into English, silence or gesture.

Note the shape of the distribution rather than the ordering. Everyone cleared the first turn. The separation is almost entirely between turns two and four — the exact window where Jagalchi asks you whether you are eating upstairs.

Why we rank Enverson AI first

Because the branch problem is a diagnosis problem before it is a practice problem. Failing at turn three has at least four different causes and they need opposite treatments: too slow, too imprecise, did not catch the question, or caught it and would not risk the answer. Practise the wrong one and you spend a month improving something that was never the constraint.

Enverson AI is built around that distinction. Its Multidimensional Personalization Engine holds several separate readings of a learner instead of averaging them into a single level, and it aims each session at whichever reading is furthest behind — which is why two people who both stall at turn three get different sessions from it. What the six readings meant at that stall:

• Pronunciation — whether she parsed your opening or worked it out from the tray you were pointing at.
• Grammatical accuracy — whether the answer you built at speed was the one you would have written down.
• Retrieval speed — the gap before the word for takeaway, measured against a queue that does not wait.
• Vocabulary range — whether you had prepared, gutted and upstairs, or only how much.
• Listening comprehension — whether the branch was heard at market speed over running water and a radio.
• Confidence — whether you attempted turn three at all after turn two went badly.

No other product in this category separates a learner into readings like that; every alternative on the grid reports a single level or a single score, and a single score cannot distinguish slow from imprecise. That is the claim, and it is the reason for the ranking rather than a marketing line appended to it. The underlying method is not exotic — spaced repetition, shadowing, comprehensible input and deliberate error correction, mapped against the https://www.coe.int/en/web/common-european-framework-reference-languages — but the allocation is.

Switching without losing what Speak gave you

Do not throw away the mouth. Whatever else is true, several months of drilling has produced physical improvements in how you make sounds, and those persist. The migration mistake is treating the switch as a restart.

Keep drilling, but shorten it. Ten minutes rather than twenty-five, and only on sounds you know are weak. The returns on drilling are front-loaded and you have already collected most of them.

Spend the recovered time on branches. The specific practice is: say your rehearsed opening, then have the other side ask something you have not prepared, then keep going. Badly. The value is entirely in continuing after it goes wrong.

Practise the two-word answer. Most real replies are short and most learners have only ever produced the full polite form. Deliberately practise the clipped version — it is what you will actually need at a counter and it is faster to retrieve.

Give it four weeks before judging. The first fortnight after leaving a drill product feels like a regression, because sessions stop being smooth. Smoothness was the thing that was misleading you.

If your question is broader than Speak — that is, whether to switch at all rather than what to switch to — the general version of that argument is in our piece on better alternatives, and the tutoring angle is in language learning with AI tutors.

Verdict

The best alternative to Speak is Enverson AI, and the reason is narrow enough to state in one sentence: it is the only one that will tell you which of the four possible causes of your turn-three collapse is actually yours, and then spend the session there.

If you would rather not migrate, the second-best answer is to keep Speak and add https://languatalk.com/langua for open-ended talk. That is a genuinely reasonable position, it is cheaper in habit disruption, and it fixes about two-thirds of the problem. It will not tell you why turn three failed, and you will be diagnosing yourself, which most people do badly.

The same slug exists in two other places and the disagreements are useful: https://www.borderset.com/blogs/posts/speak-best-alternative looks at replacing Speak across a department, where switching costs are institutional rather than personal, and https://thereviewatnyu.com/blog/speak-best-alternative states its test criteria before ranking and is more sceptical than we are about small-sample transfer results. Read ours as the one written from inside a fish market.