Long talk is one of the modes of Talk with AI. The Học nè cat gives you a topic, you speak English in one go, the cat listens, and you get a report on five skills. It suits IELTS Speaking practice, a presentation or an interview.
Getting started
- Open Talk with AI, start a new conversation and pick the Long talk mode.
- Pick a Topic type: IELTS Part 1, IELTS Part 2, IELTS Part 3, Everyday, Your own topic (type your own) or Read aloud (paste your own passage, or leave it empty for a sample one).
- The cat gives the topic in its first message. Want another one? Press Another topic before you speak. IELTS Part 2 offers a one-minute preparation timer: press Start; Pause, then Resume or Reset; ready early, press Finish now - or skip it.
Học nè picks the topic from its own topic bank - no AI, no credits.
Already have a recording? Choose Upload a recording (WAV, MP3, M4A, WebM, one speaker). Học nè reads the file on your device, then analyses it as if you were speaking.
While you speak
- Press Start talking. You can speak for up to 10 minutes, and a report needs at least 30 seconds.
- You can stop to think - the cat never cuts you off. Press Pause when you need to; a paused stretch never counts as hesitation.
- Your words appear under the topic as you speak.
- If the screen turns off or you switch tabs, the talk pauses by itself. What was recorded stays in your browser: open the room again and you will see Continue your talk at ….
After Stop you have three choices: Play back, Start over (the recording is dropped) or Analyse.
The report
The report shows:
- An overall score on the scale you chose in Settings, and five skills: Pronunciation, Fluency, Intonation, Grammar, Vocabulary.
- IELTS ≈ a band range and CEFR ≈ a level. These are Học nè's estimate from the five skills it measured - never an official exam score. A talk under one minute has no estimate.
- One thing to fix first, with a button to practise it right away (for example Practise 8 words).
- Your talk, seen through one skill at a time:
- Pronunciation: each word coloured by its score; press a word to see each of its sounds. Below it, Sounds to work on - the sounds you got wrong most, with example words.
- Intonation: stressed words, rises and falls at the end of sentences; three meters: Rise and fall, Loudness, Word stress.
- Fluency: pauses, fillers ("um", "uh") and repeated words; your pace in words a minute (good: 110-160).
- Grammar: errors underlined; press one to see You said, Better and why; the sentence structures you used and ones to try next.
- Vocabulary: each word by its level A1-C2, the advanced words you used, words you repeated a lot, and Try instead - better words to use instead.
With Read aloud the text is known in advance, so the report has three skills (Pronunciation, Intonation, Fluency) and no IELTS estimate.
What is AI and what Học nè measures
- Học nè measures - with its own algorithm, always free: fluency, word levels, repeated words, the scores and the IELTS / CEFR estimate. AI listens and turns your speech into words.
- AI does three things: it finds grammar errors, suggests better words, and grades your pronunciation - it listens to up to three one-minute windows of the talk (its start, middle and end) for the Pronunciation and Intonation scores, and points out the words and sounds you said wrong with tips. Pronunciation grading rides the Analyse price - nothing extra. If you choose Algorithm in Settings → Voice → Voice grading, Học nè measures Pronunciation and Intonation instead. AI can be wrong - check when it matters.
The AI part is priced like AI grading: on AI Học nè, each 10 minutes of speech uses one of your 30 free gradings a day (shared with AI grading), then 1 credit. The price is shown next to Analyse. With your own AI it is free. If the AI part could not run, Grammar shows Not checked with Check again, and the overall score comes from the skills Học nè measured.
Speaking at length in the other modes
Every mode of Talk with AI lets one spoken turn run up to 3 minutes (an IELTS Part 2 answer is 2).
- Press Turn on the mic and speak; press again to stop. Stop to think for as long as you like: the mic never stops on a silence, only at the end of the turn's length.
- Recording has two switches, in New conversation and in the room's ⋮ menu. Auto-start recording: once the cat finishes reading, the mic opens with no press (if you have a minute to prepare, it waits for that minute). One take: when you stop, your take goes straight to the cat, with no playing back. IELTS Speaking, Debate, Role-play and Shadowing start with both on; each mode remembers your choice.
- The mic opens by itself only while you are in the room: if you are typing, have another panel open or have moved to another tab, it waits. After two turns in a row that ran to their full length it stops opening by itself until you press it, so an empty room is not recorded on and on.
- With One take off, after you stop you have Play back, Re-record, Record more and Submit. Pressed stop by accident? Record more carries on in the same take. So the cat answers sooner, your voice is sent to be heard the moment you stop speaking, while you play it back; Re-record throws that take away.
- Didn't catch that? Just say so, for example "Sorry, can you repeat that?": the cat reads the same line again and does not move on. Your request and the repeat are not counted in your band, scores or report. AI recognises the request and can be wrong.
- While you talk, the cat listens to answer you and grades nothing; your voice is analysed in the report. Open Session summary: at the top, the cat's verdict, Total time (the cat reading plus your mic open, never the waiting) and the Graded by row; after the scores, Keep working on: your few weakest points with one thing to do. Your speaking has an overall score and five axes - Pronunciation, Fluency, Intonation, Natural, Vocabulary - over the turns you spoke (typed messages are not counted), and shows once AI has graded how natural you sound. Grammar is in the AI's feedback.
- Who grades Your speaking is set in Settings → Voice → Voice grading. With AI (the default) - pick which AI right beside it (AI Học nè or your own) - AI listens to all of your voice in the conversation again (up to 15 minutes) to grade Pronunciation, Intonation and Fluency, and points out each word said wrong, word stress, sentences whose tune should change and what you said well - each minute of voice is one pronunciation grading. Học nè puts the scores on real examiners' scale (measured on scored tests); full marks only when AI points out nothing to fix. It grades by itself while today's free gradings last, or on your own AI (free); with no free gradings left, a Grade with AI button shows the credits it takes, for you to press. With Algorithm, Học nè measures all four skills itself. Vocabulary and the numbers (pace, pauses) are always measured by Học nè. AI can be wrong.
- The Graded by row in the report says whether AI Học nè or your own AI is grading, and you can change it right there: Your words is the AI that reads what you said (the feedback, wording, grammar, how natural you sound), Your voice the AI that listens. After a change, press Grade again to grade the whole report with it.
- In IELTS Speaking mode, once you have spoken for at least a minute, the Fluency band is measured by Học nè from your speech rate (words a minute, pauses counted in, fillers like "um" left out) and the Pronunciation band comes from AI listening to your voice; the Vocabulary and Grammar bands are the AI's estimate. Graded by AI, the report waits until AI has heard all of your voice before grading the bands (AI is listening to your voice to grade your band…), so the Pronunciation band and its comments come from the words, stress and intonation AI actually heard. All of them are estimates, not a test score.
- An IELTS Speaking test runs as the real one does: the examiner asks your full name and then what to call you (unless you already said), never asks again what you have already answered, corrects nothing while it runs (the fixes are in the report), repeats a Part 1 question only word for word and never explains a word (Part 3 may), and in Part 2's minute to prepare you get Notes to jot down ideas - only you see them, and they are not graded. Typed answers alone give no Fluency, Pronunciation or Speaking band, because the real test is spoken. A test of fewer than 3 parts gives a band estimated from the parts you took.
- In the Session summary, the Each answer tab takes you through your answers one at a time: the question, what to fix marked on the words you said, and your voice. Press a marked word to see how to fix it, press any other word to hear from there, and the words light up with your voice. Upgrade this answer asks AI to rewrite your answer from your own ideas, suggest one more idea and pick new phrases; AI can be wrong, so check it before you learn from it. Say this answer again lets you say it and scores it, a long answer in short parts. This practice never changes the session's scores.
- A turn longer than 15 seconds has no Score pronunciation button - its words are already scored in the Session summary.
In the room
On the same line as Tap to speak are the things that decide the cat's next reply. Change them whenever you like.
- Level (left): pick a step, in CEFR, HSK, JLPT or TOPIK depending on the language, or by school year when you talk in your own language. The step decides how hard the cat's words are, how long it talks, how fast it reads, whether the translation stands open and how closely it corrects you. Auto is the default: the cat adjusts to how you talk. Turn it off to pick a step; each language remembers its own. The Session summary estimates the level you talked at (AI can be wrong) and offers one step up or down; each language remembers its own step.
- AI (right): change the AI that answers, and choose Thinking: longer thinking gives a more careful reply that comes more slowly. With AI Học nè, the price shows right under your choice.
- How the cat listens (in the AI menu, or Settings → Voice): Hears voice (the default) has the cat read the words written down and hear the recording too, so it can hear you sing, play an instrument or put on music; Reads words has AI write down what you said and gives it to the cat, a little faster.
- An IELTS Speaking test has no level: the examiner talks as in the real test. The test opens Face to face with subtitles off, as the real one is; one press brings them back. Role-play and Debate open Face to face too.
- What you need to see while you speak stays pinned in a corner of the room, even with subtitles off: Part 2's task card, preparation clock and Notes, the scene in a role-play, the motion in a debate, the words a deck still needs. They appear once the cat has finished reading its line (at once if you open the mic before that). The clock goes when you open the mic; the card and the notes stay until Part 3, as in the real test.
- In the other modes (except Shadowing), the cat can also hand you a clock, Notes or a task of its own when it sets a long speaking task or when you ask, say "give me a minute to think". The task is written by AI and may be wrong. The clock goes when you open the mic or answer; the task and the notes stay until your second answer, then rest under the cat's line.
- Notes: on a wide screen the sheet has a column of its own on the left, as tall as the room; on a phone it sits under the task card, and the mic's bar steps aside while the keyboard is up. Folding the task card gives it more room too. Enter starts a new line,
-or1.continues a list by itself, and the sheet grows with what you write. Press Esc to leave it, then Space to open the mic. Your notes are kept with the conversation; only you see them and they don't count towards your speaking score. After the talk, read them again in the Session summary, and press AI review to get a score and tips on better notes (AI can be wrong). The AI reads them only when you press it.
Naturalness
Every mode of Talk with AI, Long talk too, has a Naturalness card: how natural you sound to a fluent speaker, scored out of 10 on four sides:
- Words and Phrasing - the AI reads what you said: whether you pick the words and phrases fluent speakers use or translate word for word, and whether your sentences sound spoken or like written text. The AI can be wrong.
- Flow and Intonation - taken from Your speaking, only when you speak with the mic.
Typed answers alone give the first two; Shadowing (repeating the cat's lines) gives the last two. The card comes with the AI's summary at no extra credit.
In the Session summary, Natural is an axis of the chart and makes up 20 parts of Your speaking (not in Shadowing: the words are the cat's). Its four sides are on the Natural tab under Your speaking in detail.
In IELTS Speaking, naturalness is not a fifth criterion: it is already part of all four official criteria, so Học nè uses it to set the overall level the four bands sit around. Học nè measured this on real tests scored by examiners: with naturalness in, the bands come closer to the examiners'. If you answer mostly in a language other than English, the band is 3 at most (almost entirely: band 1), because the real test rates English only. On the other hand, bands 1-2 are only for isolated words: once you have said at least three English sentences of three words or more, Vocabulary and Grammar are never under 3; at 40 words a minute or more, Fluency is never under 3 - Học nè counts it, by IELTS's own description of band 2.
History and privacy
- Each talk is a conversation in Your chats with AI, with its overall score beside it. Open it again to see its report.
- Học nè keeps the report, your words as text and your voice recordings so you can listen again. Only you can hear them. Deleting a conversation deletes its recordings; deleting your account deletes them all. To keep none, turn off Keep my voice recordings in Settings → Voice - you can also delete every recording there.
- A message you spoke shows as a voice message: press ▶ to hear yourself, with the words AI heard under it. In the report, every sentence to fix and every word said wrong has Your voice beside it. On the history list, ▶ replays a whole conversation: the cat's voice, then yours; a line you typed is read in a model voice.
- Only you can see your talks. To delete one: ⋮ → Delete.
Data sources
Word levels come from the CEFR-J Wordlist Version 1.5 (A1-B2), Tono Laboratory, Tokyo University of Foreign Studies, and the Octanove Vocabulary Profile C1/C2 Version 1.0 (C1-C2), Octanove Labs, licensed CC BY-SA 4.0.