The listening method behind the app

Public Cambridge sample materials guide the task architecture, pacing and distractor logic. The app does not preserve or lightly rewrite any published dialogue. Every script, question, character, option and explanation is newly written.

Our north star

A Cambridge-faithful simulator when support is off, and an intelligent listening coach when support is on.

The B1 target, task purpose and answer logic stay the same in both modes. Training mode improves access; it does not lower the level.

Meaning before matching

Students must understand the whole message, not choose an option because one word appears in the audio.

Real distractors

Wrong options are naturally mentioned, suggested or implied, then clearly ruled out by context.

One defensible answer

The recording supplies enough evidence for one answer only, and the explanation identifies that evidence.

Natural communication

Every line belongs in a believable school-age situation. Scripts stop when the communicative purpose is complete.

Ten reference patterns we use

These are high-level task patterns, not dialogue templates. We reuse the listening purpose and distractor mechanism, while changing the communicative situation, sequence of ideas, wording and question construction.

Part 1

Practical elimination

All three objects are mentioned. Two are already available or unsuitable, so one becomes the practical choice.

Build rule: every visual is relevant, but only one fits the final situation.

Part 1

Past versus still true

The speaker mentions an earlier problem and a different detail that remains true now.

Build rule: use time markers such as at first, before, now and still.

Part 1

Route or sequence

Several places or actions appear, but the question asks which one is first, next or last.

Build rule: the answer depends on order, not the most memorable noun.

Part 1

Change of plan

Two activities are considered and rejected before the speakers settle on the final one.

Build rule: stop the dialogue immediately after the final decision is secure.

Part 2

Main reason

A speaker describes several positive details, but one reason matters most to the question.

Build rule: distinguish background detail from the speaker's central reason.

Part 2

Qualified opinion

A speaker begins with a reservation, adds a condition and then gives an overall view.

Build rule: the correct option paraphrases the final position, not the first reaction.

Part 2

Expectation versus experience

Someone expects one difficulty, experiences another and explains what they learned.

Build rule: use contrast to test the real conclusion.

Part 2

Advice after alternatives

Several possible solutions are discussed before one recommendation becomes clear.

Build rule: make every option credible, but one best summarises the advice.

Part 3

One organised information journey

A continuous talk moves through six answer-bearing facts in a natural order.

Build rule: gaps follow audio order and use exact spoken information without filler.

Part 4

One complete interview arc

The conversation develops from origins and challenges to attitudes, advice, results and future thinking.

Build rule: six questions sample the whole recording, not only the first six replies.

Pacing calibrated against an official public sample

Cambridge does not publish fixed word-count limits for each recording. We used the spoken content in B1 Preliminary for Schools Listening Sample Test 1 as a practical pacing benchmark, then set broad guardrails so our original scripts remain natural rather than copied or padded.

Part	Official Sample Test 1 benchmark	This app's original bank	Release guardrail
Part 1	About 79-92 spoken words in Sample Test 1	75-84 words	70-105 words
Part 2	About 85-105 spoken words in Sample Test 1	96-108 words	80-125 words
Part 3	About 304 spoken words in one talk	290-292 words	250-340 words
Part 4	About 444 spoken words in one interview	415-430 words	390-480 words

Why ranges? Authentic task purpose and coherent language matter more than forcing every script to the same number. A script fails if it is short because evidence is missing or long because irrelevant filler was added.

Official references: Sample Test 1 tapescript · current exam format · handbook for teachers

Project goals retained in this build

Goal from the project	How the app protects it
Official 7 / 6 / 6 / 6 structure	Exam rounds use seven Part 1 items, six Part 2 items, six Part 3 gaps and six Part 4 questions: 25 answer points.
Cambridge-style pacing at normal TTS speed	Every bank script now sits inside part-specific guardrails calibrated against the official public sample tapescript.
Original content, not copied material	Only high-level task architecture and listening methodology are modelled. No source dialogue is retained with noun or adjective substitutions; all wording, situations, sequencing, questions and distractors are original.
Natural-speed listening	1.00x is the default and fixed speed in Exam mode. Slower playback is clearly identified as Training-mode support.
Two listens under exam conditions	Exam mode permits exactly two complete plays and blocks a third. Training replay remains flexible.
Male-first voice pairing	The first unique character in a two-person recording uses the chosen male voice; the second uses the chosen female voice.
Two voices and two characters in Part 4	Every interview uses a male Presenter and one female guest. No narrator, third character or voice switching is allowed.
UK / US voice choices	Settings provide UK, US or any-English accent preference plus separate male and female browser voice selectors.
Transcript and audio consistency	TTS and the displayed script are generated from the same transcript data. Speaker names are formatted but never spoken.
Script-style transcripts	Every turn has a bold speaker label, a separate row and clear spacing. Monologues remain one coherent speech block.
Transcript control	Text is absent until deliberate reveal or submission, the reveal state is explicit, and it resets for every new item.
Authentic, non-meta dialogue	Scripts use believable school-age situations and are rejected if characters discuss being in a test or if unrelated tail text remains.
Part 1 image and topic alignment	Each option receives an original semantic cue card tied to its meaning. Written option labels are available in Training mode but hidden in Exam mode to prevent direct word matching.
Part 1 distractor methodology	All three choices are relevant, while correction, sequence, practical elimination or a final decision identifies one answer.
Part 2 main-message methodology	Questions target overall meaning, reason, attitude, advice or decision rather than isolated word matching.
Part 3 exact-answer discipline	Six gaps follow audio order. Canonical answers are no more than two words, a number, date or time, with sensible variants accepted.
Part 3 sound and prediction support	Training mode predicts the information type and offers optional first-letter and word-count hints without changing the B1 answer.
Part 4 full interview arc	Six questions sample early, middle and late exchanges and cover at least four listening skills such as reason, attitude, process and reflection.
Broad school-age topic bank	Categories include school, hobbies, sport, travel, home, food, technology, health, jobs, visitors and community activities.
Training and Exam separation	Support use is tracked. Training gives coaching; Exam mode fixes speed, hides support and delays correctness feedback.
Visible settings and accessibility	A persistent Settings control provides appearance, font size, accent and voice choices, with keyboard operation and live status.
Dark and light appearance	Learners may follow the device setting or choose light or dark mode, with readable contrast in both.
Mobile, tablet and desktop use	Navigation, controls, cue cards, transcripts, method panels and settings are responsive without horizontal overflow at 390 px.
Progress and supported-attempt clarity	Scores, completion, streaks and support use are tracked separately so transcript-assisted work is not confused with independent performance.
Reproducible recursive QA	The exact deployable files include structural, methodology and browser-runtime validators and are retested after fresh ZIP extraction.

Release rule

Structural QA: PASSMethodology QA: PASSRuntime QA: PASSOverall release status: PASS

A prewritten report is never accepted on its own. The checks must run against the same files that are packaged for deployment.