If ten repetitions of *ship* and *sheep* still sound interchangeable to you, more mindless repetitions may not solve the problem. You need a difference your ears can use as a target.
That does not mean listening alone will fix your speaking. A more useful cycle is: notice a contrast, imitate it, record yourself, then test it in new material.
What the study actually supports
In a study published in *System*, Bradford Lee, Luke Plonsky and Kazuya Saito worked with 115 Japanese tertiary-level English learners. Over two weeks, they compared perception-focused instruction, centred on identification and discrimination, with production-focused instruction, centred on articulation and feedback. The targets included both individual sounds and suprasegmental features such as rhythm and stress, with pre-, post- and delayed post-tests.
All four treatment groups showed substantial gains in pronunciation accuracy. A closer comparison suggested that perception-based instruction may have been more effective across the target features in this context.
That is not a universal rule that listening always beats speaking, nor proof that headphones alone create fluent speech. The learners, teaching sequence, feedback and targets were specific. Treat the result as a reason to calibrate your listening, not a reason to postpone production indefinitely.
A six-step listen–speak loop
- Choose one contrast: *ship/sheep*, or stressed versus unstressed words—not an entire phonetic chart.
- Judge before seeing the text: identify which item you heard, or where the prominence falls.
- Check immediately: write down the direction of your error and form a small hypothesis, such as “I am lengthening both vowels”.
- Imitate with that hypothesis: hear a reliable model once, then speak once with a purpose.
- Record and replay: check only the chosen target, not everything at once.
- Change the material: try a new word, sentence or speaker to see whether the contrast survives outside the memorised clip.
Listening is not a prerequisite course you must finish before speaking. It is the calibration before and after each attempt.
A ten-minute example
- Two minutes: judge eight *ship/sheep* items without seeing the spelling, then check the answers.
- Two minutes: revisit the two hardest contrasts and describe the difference you can hear.
- Two minutes: put them into *The ship is near the shore* and *I saw a sheep on the hill*; listen and imitate.
- Two minutes: replay your recording, checking the target vowel and the rhythm of the whole sentence.
- Two minutes: try two new sentences to test whether you are relying on one memorised recording.
For rhythm, replace word identification with pointing to the stressed word or comparing syllable counts. Return the physical or listening task to real speech; do not stop at tapping a desk.
What perception practice cannot solve alone
You may hear a contrast clearly but still need tongue- and mouth-position guidance to produce it. You may identify isolated words accurately yet lose them in connected speech; that calls for sentence-level practice, not more isolated quizzes. A missed speech-recognition result may also reflect noise or model limits, not necessarily your ears or accent.
Connecting the loop with MirrorSay
Do the discrimination task with reliable audio, then use MirrorSay to speak a short sentence containing the target. Hear the reference, shadow it and replay your own recording. Its word-level timing comparison helps with sentence alignment and pauses; it does not certify individual phonemes.
For a fuller listening design, see High-variability phonetic training; for the limits of shadowing, read What shadowing research says.
Sources and limitations
- Lee, B., Plonsky, L., & Saito, K. (2020). The effects of perception- vs. production-based pronunciation instruction. *System*, 88, 102185. DOI · Open manuscript at UCL
- Kartushina, N., & Frauenfelder, U. H. (2014). On the effects of L2 perception and of individual differences in L1 production on L2 pronunciation. *Frontiers in Psychology*, 5, 1246. Open article. Perception and production are not always strongly linked; a listening score does not directly prove pronunciation accuracy.
We located and read the original research and open abstracts for this guide. The routine is a self-study application, not a reproduction of the experiments.