Cambridge Speaking Exam Practice: Weekly FCE & CAE Activities
Most Cambridge speaking exam practice happens backwards. A class spends the term on grammar, vocabulary, and general conversation, and then, two or three weeks before the test, the teacher pulls out a past paper and everyone rehearses the actual exam format for the first time. That's a vocabulary problem being solved with a format solution. The students usually know enough English to pass; what they haven't rehearsed is the specific, unfamiliar shape of the test itself — talking to a stranger for exactly two minutes about two photos, or reaching a joint decision with a partner they've never negotiated with under a timer. Cambridge speaking exam practice works better spread across the term than crammed into the final fortnight, and it doesn't need a stack of past papers to do it. It needs the exact task shape rehearsed often enough that it stops feeling unfamiliar.
What the Cambridge Speaking test actually asks for
Cambridge's own exam-format pages (checked live 2026-08-02) lay out the Speaking component for both levels precisely. For B2 First — still widely called the FCE speaking test by teachers, even though Cambridge renamed it years ago — the Speaking test runs 14 minutes for a pair of candidates or 20 minutes for a group of three. C1 Advanced (CAE) runs a slightly longer version of the same shape: 15 minutes per pair, or 23 minutes per group of three. Both are face to face with two examiners, one who asks the questions and one who only listens and scores.
Both exams share the same four-part structure, regardless of level:
- Part 1 — a short interview, question and answer with the examiner.
- Part 2 — an individual long turn: one candidate describes and compares two photos while the other listens.
- Part 3 — a collaborative task: the pair works through visual prompts together, aiming for a joint decision.
- Part 4 — a closing discussion that extends the collaborative task's theme.
That structure is the whole point. A candidate with broad vocabulary who has never had to fill exactly one minute of solo talk about two photos, or never had to reach an actual joint decision with a partner rather than just take turns speaking, is walking into an unfamiliar task under time pressure. Unfamiliarity, not language level, is often what costs marks.
Why exam task familiarity matters more than another vocabulary review
Dörnyei (2005) identifies L2 anxiety as one of the most consistently negative predictors of oral exam performance, and the mechanism is straightforward: novelty is what triggers it. A student who has never had to reach a joint decision with a partner in exactly two minutes, under a timer, in front of an examiner, is managing two things at once in the exam room — the language and the format. Regular, low-stakes rehearsal of the exact task shape, run often enough that it stops being new, removes the second problem so students walk in only having to manage the first. That's exam task familiarity in practice: not more grammar drilling, but repeated, unremarkable exposure to the same task shape until it isn't a source of stress anymore.
BubbleTalk's Chelsea Mode: rehearsing the collaborative task cold
BubbleTalk generates a cluster of keyword bubbles around a topic that one or two students at the screen build a spoken presentation or discussion from, tapping bubbles as they're used and branching off new ones as the talk develops. Its Chelsea Mode switches the generation logic specifically to match Cambridge B2 First Speaking Part 3: instead of a general topic, it produces a central collaborative-task question with bubbles orbiting it as discussable sub-points — the exact shape of the real exam task, not an open-ended discussion that happens to be collaborative.
What makes this more than a discussion prompt is the app's Chelsea Decision Round activity: after working through the sub-point bubbles, the pair has to reach and justify one final decision on the central question, out loud, before the round ends. That mirrors the last stage of the real exam task, where a show of good ideas isn't enough if the pair never actually agrees on anything. Sweller's (1988) cognitive load theory explains why the constantly regenerating bubble pool works at all here: with the content pre-loaded, a student's working memory goes toward stringing the language together and steering the conversation toward a decision, rather than inventing what to say next from nothing. The optional connectors — however, in addition, as a result — add a second layer on top of that. They scaffold the kind of cohesion Canale & Swain's model of communicative competence treats as its own skill, distinct from vocabulary: an examiner isn't just listening for the right words, but for whether ideas are linked logically rather than strung together with "and… and… and."
Brooklyn Trainer: the individual long turn, and a step further
Brooklyn Trainer covers the other half of the format — the individual long turn. Its 2-image mode is the classic Speaking Part 2 shape: one student compares two AI-generated photos and answers a discussion question while their partner listens and tracks a pre-loaded "vocabulary boost" of three to five level-appropriate words the speaker should aim to use naturally. Its 3-image mode shifts closer to the Part 3 collaborative shape, giving pairs a small set of images around a shared topic instead of a solo long turn.
Two pieces of research ground why this works as more than "look at pictures and talk." Maley & Duff (1982) point out that image-based prompts pull out a register no text prompt produces — speculative, interpretive language ("this might be…", "they appear to be…"). That's exactly what Cambridge photo tasks are scoring, since a candidate has to speculate and compare, not just describe. Separately, Laufer & Nation (1995) found that pre-loaded vocabulary — words a student knows in advance they're expected to use — produces measurably more complex spoken output than unprompted free speaking. Brooklyn Trainer's vocabulary boost is that pre-load: students read the target words before the clock starts, which activates them for use in the response, closer to how a well-prepared candidate walks into the real test having practiced comparison language rather than reaching for it cold.
| App / Mode | Cambridge task rehearsed | CEFR levels | Formation | What it trains |
|---|---|---|---|---|
| BubbleTalk — Chelsea Mode | Speaking Part 3, Collaborative Task | B1–C2 (Chelsea Mode B2/C1) | One or two at the screen | Joint decision-making, cohesive connectors |
| Brooklyn Trainer — 2 images | Speaking Part 2, Individual Long Turn | B1–C1 | Pairs | Solo comparison, speculative language |
| Brooklyn Trainer — 3 images | Speaking Part 3-adjacent, Collaborative Task | B1–C1 | Pairs | Comparison across a small image set |
Building a weekly rotation instead of a cram week
A rotation that spreads these across a term looks less dramatic than a mock-exam week, which is exactly the point — the goal is for the format to stop being an event. Early in a course preparing for B2 First, run Brooklyn Trainer's 2-image mode weekly, one pair at a time while the rest of the class listens and later gives feedback on the comparison question; this builds the solo long-turn habit before adding a second task shape. Once that's routine, alternate in BubbleTalk's Chelsea Mode every other week, moving one or two students to the screen for the collaborative task while everyone else works through their own sub-point bubbles on paper as a listening-and-noting exercise. In the final few weeks before the real exam, run both back to back in the same session — Brooklyn Trainer's long turn followed immediately by a Chelsea Mode round — to rehearse the actual sequence of the test, not just its individual pieces in isolation.
The same rotation carries over directly to CAE speaking practice for a C1 Advanced class — nothing about the task shapes changes, only the CEFR level setting in both apps. Set Brooklyn Trainer to C1 and its vocabulary boost shifts to hedging and evaluative language ("it is conceivable that…", "arguably…") instead of B2's simpler comparison collocations; set BubbleTalk to C1 in Chelsea Mode and the sub-point bubbles expect more independently developed, nuanced reasoning rather than straightforward opinion-plus-example. Same rotation, same weekly habit, just a different level dial.
What this doesn't replace
Neither app puts a student in front of an actual examiner, and neither rehearses Speaking Parts 1 or 4 — the opening interview and the closing discussion, both built around spontaneous question-and-answer rather than a fixed task shape. That's a real gap: a class that only ever runs Brooklyn Trainer and BubbleTalk has rehearsed the two most format-heavy parts of the test thoroughly and the two most conversational parts not at all. The honest framing is that these two apps take the two-thirds of the exam that has a genuinely fixed, learnable shape and make it a weekly habit instead of a once-a-term mock. They're not a substitute for a full timed mock exam closer to the test date, which is still the only way to rehearse the whole sequence, examiner and all, in one sitting.
I ran BubbleTalk's Chelsea Mode with a B2 exam class the week before their real FCE test, and the Chelsea Decision Round was the part that actually surprised me. Two of the strongest pairs on paper — good vocabulary, fluent individually — stalled at the decision stage, because they'd been taking turns talking rather than actually listening to and building on what the other person said. The bubbles weren't the problem; neither of them had ever had to end a two-minute conversation with an answer they both had to agree on and defend. That's not a vocabulary gap a review session fixes — it's exactly the kind of exam task familiarity a weekly rotation builds before it costs marks on the real day.
Both apps live inside the same ENK account as the rest of the library, gated by monthly AI credits rather than a separate exam-prep subscription — 100 credits free every month, or from €5/month for 300 credits on the pricing page. Browse the full app library to see how BubbleTalk and Brooklyn Trainer sit alongside the rest of the Cambridge-relevant apps covered in the best ESL classroom games roundup, or start from ESL speaking activities for the broader speaking-practice picture these two apps sit inside. If a BubbleTalk topic surfaces a genuinely contestable claim, You're Wrong is a natural next step to develop it into a structured discussion; for the visual-interpretation side, Iconic Stories builds the same speculative, image-based language Brooklyn Trainer trains, but through narrative sequencing rather than comparison.
FAQ
What's the difference between FCE and CAE speaking practice? The task shape is identical — four parts, an individual long turn, a collaborative task, a joint decision — the difference is the level of language expected inside that shape. CAE speaking practice asks for more independently developed, evaluative language (speculating, hedging, weighing options) rather than B2's simpler opinion-plus-example pattern. Both BubbleTalk and Brooklyn Trainer handle this with a CEFR level setting rather than a separate app.
Does BubbleTalk's Chelsea Mode really rehearse Cambridge B2 First Speaking Part 3? It rehearses the task's structure — a central question with orbiting sub-points leading to a joint decision — not the exact exam materials, since Cambridge doesn't publish those for reuse. What it builds is comfort with the shape of the task: working through prompted sub-points toward an agreed answer under a timer, which is the part students who've never seen the format tend to freeze on.
Can Brooklyn Trainer be used for C1 Advanced as well as B2 First? Yes — it supports B1, B2, and C1 directly, with the vocabulary boost and question complexity adjusting to match. The 2-image mode maps to the Speaking Part 2 long turn at any of those three levels; the 3-image mode adds the collaborative dimension for classes further along in exam prep.
How often should a class rehearse Cambridge speaking format before the real exam? Weekly, starting well before the exam date, is what actually builds exam task familiarity — a single mock exam two weeks out tests whether students can perform the format once, not whether the format itself has stopped being a source of anxiety. Dörnyei's (2005) research on oral exam anxiety points the same way: it's repeated, low-stakes exposure that reduces the novelty effect, not one high-stakes rehearsal close to the date.
Written by Prof. Vito Schiuma, designer of English, no kidding.