One meaningful difference.
Ship and sheep share the same consonants. Their vowel changes the word. A minimal pair keeps the rest of the sound pattern steady, giving you one contrast to focus on.
Other lessons focus on word stress, Mandarin tones or a consonant feature such as length. Those targets are labeled separately. In every case, the exercise asks you to compare what you hear across words.
Listen to A. B. Then X.
A and B introduce the two target categories. X is a different word: choose the reference that shares its target sound. For example, if A is ship, B is sheep and X is fill, the vowel in X matches A. You are comparing a sound category across words.
The exercise starts with audio only. Words, spelling, IPA and meanings stay hidden unless you turn on text hints, including after you answer. You can replay A, B or X. After choosing, compare the sounds again and enable the hints if useful. No microphone is required.
For tone, stress and consonant-feature lessons, match that specific pattern instead of one identical consonant or vowel. A short instruction tells you what to listen for.
What the research tells us.
Researchers use categorical ABX tasks to assess whether X shares a target category with A or B. In the study linked here, X repeated one reference word in a different speaker’s voice. Our requirement that X be a different word is an extension of that task. The study provides a research precedent for ABX assessment; it does not establish learning gains from our app. Read an ABX study protocol ↗
A comparison of identification and categorical discrimination training found benefits from both, with identification training producing larger gains on identification in that study. The discrimination task there used AX, not our cross-word ABX format. We choose ABX to make sound comparison the core experience; we do not claim it is universally superior. Read the training comparison ↗
A familiar clip is only a start.
X uses a different word so that matching an identical recording cannot solve the trial. In the cached practice libraries, A and B share one voice profile and X uses another. Voice assignments and answer positions vary independently. A word that changes within the current trial is not necessarily new to you; a genuine transfer check reserves words or voices outside the practice material.
More voices are not automatically better. A substantial replication found ambiguous evidence for an advantage of multiple talkers over a single talker. Our priority is carefully chosen, correct recordings; a voice count alone is no guarantee of learning. Read the replication study ↗
Come back while it is still a small task.
Sessions are short by design. Whether that helps learners return remains to be measured. Revisit a contrast after a break, check what you can still hear, and give difficult examples another turn. The session length is a practical choice, not a scientifically established optimum.
Research on retrieval practice supports checking remembered knowledge rather than relying only on repeated study. Applying that principle to a listening routine is a design decision that still needs its own measurement. Read the retrieval-practice study ↗
Progress with a clear meaning.
A result such as “9 of 12 correct” describes your first ABX choices in that session. Audio-only and text-assisted attempts should be reported separately. Two-choice exercises have a 50% chance baseline, and a short run cannot establish mastery. Your history is a record of practice, not a fluency certificate or an accent score.
WHAT THIS EDITION DOES
Listening exercises across nine languages, with varied voices, sound guides and feedback.
Your native language may offer useful context for an explanation. It does not decide what you can or cannot hear. Choose the contrasts that matter to you and use your own results as the starting point.
Make the habit encouraging.
Clear goals, useful feedback and visible personal progress can make practice easier to return to. A missed day does not undo what you learned. The point is a better next listen, not protecting a streak.
Gamification research reports benefits in some settings, with varied results and important limits. We treat motivational design as something to test, not a promise that points or badges teach a language. Read the gamification meta-analysis ↗