Knowing the language is the easy part
Being understood takes more than correct sentences. It is the pause you hold instead of saying um, the word you lift so the point lands, the shape you give a story so someone remembers it. SpeakUp listens to one real conversation and builds a two-week roadmap from what it actually hears.
Your first diagnostic conversation is free. Membership unlocks the roadmaps.
Not sure yet? Build your plan in a minute, no signupTwo people, the same gap
Between what you mean and what reaches the other person. The words are only half of it.
You're learning the language
You can recite rules you don't apply when you stop being careful. The diagnosis works from what you actually produced in conversation, not from what you can explain — then drills those patterns, and only those.
- Named patterns pulled from your own transcript
- A note on why your first language pushes you that way
- Drills that ladder from recognising a form to producing it
You already speak it fluently
Then grammar isn't your bottleneck — delivery is. Filler words, a pace that outruns the room, a flat line where the emphasis belongs, a story with no turn in it. All of that is measurable, and all of it is trainable.
- Pauses that land, instead of um and uh
- Pitch that marks the word you actually mean
- A story with a hook, a turn and an ending
English or French, whichever you speak
The language you are working on and the language you read the site in are two separate choices. Pick either combination.
English
Whether it is your first language or your fourth. Learners fix the forms they actually produce; fluent speakers work on the delivery that decides whether a point lands.
Working on English FRFrench
The same method in French: a conversation that finds your agreement, tense and partitive habits, and a rhythm drill built on where French really puts its weight — the end of the group. The speech side is younger than the English one; the grammar side is the same engine.
Working on FrenchSwitch the language you're learning any time in Settings. Each one keeps its own roadmap, its own profile and its own progress.
One cycle, start to finish
Every round is the same four steps. You always know where you are and what unlocks next.
- 1
Talk
A few minutes of normal conversation. No test, no score — the app is listening for patterns while you chat.
- 2
Get diagnosed
What you do gets named — “uses present tense for past events”, “fills every gap with um” — with real examples from your own transcript.
- 3
Follow one roadmap
A tree sized to about two weeks at your cadence. The habits that show up most get the most stops; the easiest fixes come first.
- 4
Prove it changed
Clearing the tree unlocks a check-in conversation, which ends in a before/after report: improved, still needs reps, or newly spotted.
A course that could only have been made for you
A roadmap built from YOUR habits
Not a generic course — every exercise targets something heard in your own conversations.
Real AI conversations
Talk naturally, get diagnosed silently. No flashcard walls, no corrections mid-sentence.
Proof you're improving
Every round ends with a before/after report on your real grammar and your real delivery.
Nine formats, recognition laddering into production
Each one is matched to where the habit actually lives — on the page, in your mouth, or in how the whole thing is put together. The later ones take the scaffolding away.
Select
Pick the right form. Recognition first — the distractors are built from the mistakes you actually make.
Type
Fill the blank yourself. No options to lean on, so you have to produce the form rather than recognise it.
Arrange
Build the sentence from shuffled tiles. This is where word order stops being theoretical.
Find
Spot the error in a full sentence yourself, then fix it. The self-monitoring skill that transfers to real speech.
Produce
A situation, and nothing else. You write the line — no gap, no tiles, no options. This is the first format that asks for the form with no scaffolding at all.
Under time
The same production, with a clock on it. Habits come back the moment you have to answer before you can think, which is also when you actually speak.
Repair
Your own sentence, from your own transcript, handed back for you to fix. Nothing generic about it — this exercise did not exist before you said the thing that made it.
Say it
Read the line out loud. Some drills check the recogniser heard the word; others measure how you said it — one real pause, a pitch that moves, no filled hesitations.
Write it
A short piece to a named reader, in the register the task asks for. Marked on whether it did the job, not only whether the grammar held.
Yesterday she ____ to the office early.
A step only clears at 100%
Miss anything and you run it again until it's perfect. Every run is recorded, but only your first attempt at each item counts as evidence about you — so the roadmap can't be gamed by grinding.
It's stricter than most apps on purpose. Clearing the whole tree is what unlocks your next conversation.
Your profile is the product
One living document describes what the app believes about you — and you can read all of it.
Named patterns, each with a status, how hard it is to fix, how often it shows up, and real examples pulled from your own transcripts. The profile screen renders exactly the same data the AI reads when it builds your next roadmap. Nothing is hidden behind a score.
I want to discuss problem
I want to discuss the problem
The part that listens to how you sound, not just what you say
Pace, pitch range, pauses, filler words and the shape of your story — measured against your own baseline, never against “native speakers”. Speaking drills are already part of every session; this is the instrument around them.
Record
Speak for a minute and see how it actually sounded — the pitch line, every pause, and where your emphasis landed.
Modulation
Train your range against targets built from your own voice.
Rhythm
Put the weight where the language actually puts it.
Speech models are downloaded to your device and run there. Your audio and video never leave it — only the measurements are saved to your account.
Those models are a few hundred megabytes the first time you open the Studio. Worth knowing before you start it on mobile data.
In beta: it is being rolled out account by account, so it may not be switched on the day you join. Speaking and rhythm drills inside a session do not wait for it.
Find out what you actually sound like
It takes one conversation. Most people are surprised twice: by a rule they can recite and still don't use, and by a habit they had no idea was audible.
Your first diagnostic conversation is free. Membership unlocks the roadmaps.