Do Pronunciation Scores Improve When Learners Try Again?

By Talkio Research

Published September 16, 2026. First-party observational study; not an independent educational trial.

Abstract

Objective. This study examines whether pronunciation scores improve when learners repeat the same exercise. Our primary question was whether the subsequent saved retry received a higher score than the preceding retry. We also examined how quickly learners tried again.

Method. We analysed a privacy-protected sample of 2,889 eligible pronunciation-exercise pairs selected from Talkio product data. Each pair contains the first two saved retries for the same sentence exercise, ordered before any outcome was compared. The sample is not a count of all Talkio pronunciation activity, learners or sessions. To reduce the risk of inferring platform scale, we do not publish the source window, its length or account counts.

Findings. The average pronunciation rating increased from 3.65 to 3.80 on Talkio's five-point practice scale. The subsequent retry received a higher rating in 23.0% of pairs, the same rating in 66.6%, and a lower rating in 10.4%. In addition, 98.3% of subsequent retries occurred within five minutes of the preceding retry.

Interpretation. The results are consistent with repetition helping learners perform better on their next recorded attempt: the average score increased, and higher subsequent ratings were more than twice as common as lower ones. Because the study is observational, it cannot establish that repetition caused the difference or that the score change represents lasting improvement in pronunciation ability.

Introduction

Repeating a difficult word or phrase gives a learner an immediate opportunity to apply feedback, adjust a sound or rhythm and try again. This listen–record–review–repeat cycle is a practical form of focused pronunciation practice. The question examined here is whether the next recorded attempt tends to receive a better score.

This study looks at that narrow moment by comparing two consecutive saved retries for the same exercise. It does not attempt to measure overall language proficiency, fluency in spontaneous conversation or learning over time. The focus is improvement in the automated pronunciation score on the subsequent attempt, with retry timing included as supporting context.

Research questions and hypotheses

We examined three research questions:

1. Does the subsequent saved retry tend to receive a higher pronunciation score than the preceding retry?
2. How is the subsequent score distributed across higher, unchanged and lower outcomes?
3. How often does the subsequent retry occur within five minutes?

Our primary hypothesis was that the subsequent retry would have a higher average pronunciation score, with score increases occurring more often than decreases. Our secondary hypothesis was that most subsequent retries would occur shortly after the preceding attempt. These were exploratory hypotheses for this company analysis; they were not preregistered. A higher automated practice rating was not treated as evidence of lasting language learning.

Method

Study design and sample

We conducted a descriptive, paired-observation study using aggregate records from Talkio's production system. The analysis used a selected, privacy-protected sample of 2,889 eligible exercise pairs. It should be read as a sample of product interactions, not as a census of Talkio learners, customers, sessions or pronunciation records.

The source dates, duration of the sampling window and number of contributing accounts are intentionally not disclosed. No conversation transcript, recording, sentence text, name, email address or account identifier is published.

Pair construction

For each eligible sentence exercise, saved retries were ordered by recording timestamp and record ID. We compared the first two saved retries in that fixed order; we did not choose the best, highest or latest score.

The two records had to have a later timestamp for the second retry, matching exercise ownership, sentence text, language and stored proficiency label, and valid stored ratings between one and five for pronunciation, accuracy, fluency and completeness.

The exercise's main record is updated after a retry, so it cannot serve as an untouched first-attempt baseline. The separate retry history preserves the paired observations used here. As a result, the first observation in each pair is itself a saved retry. This is not an initial-attempt-to-final-attempt comparison.

Outcomes and analysis

The primary outcome was the change in pronunciation rating between the two saved retries, including the mean change and whether the subsequent score was higher, unchanged or lower. The secondary outcome was whether the subsequent retry occurred within five minutes of the preceding retry.

Percentages use eligible exercise pairs as the denominator, with one pair per exercise, and are rounded to one decimal place. A rating increase means crossing a category on the practice scale. It is not a percentage improvement in pronunciation ability. The average rating is a descriptive summary of an ordinal scale, not an equal-interval measure of language skill.

Findings

Outcome	Result in the selected sample
Exercise pairs analysed	2,889
Average rating, preceding → subsequent retry	3.65 → 3.80
Higher rating on the subsequent retry	23.0%
Unchanged rating	66.6%
Lower rating on the subsequent retry	10.4%
Subsequent retry within five minutes	98.3%

The results support the primary hypothesis within this sample. The average score increased by approximately 0.15 points, and higher subsequent ratings were observed more than twice as often as lower ratings. The timing result supports the secondary hypothesis: almost all paired subsequent retries occurred within five minutes.

The unchanged group is important. Although 89.6% of pairs maintained or increased their rating, maintaining a rating is not improvement. Reporting the three outcome groups separately avoids overstating the finding.

Discussion

Feedback is easier to act on when it points to something a learner can try immediately. A word-level result can suggest where to listen more carefully, and repetition creates another opportunity to adjust the pronunciation or rhythm. The upward average score and the imbalance between higher and lower outcomes are consistent with that repeat-and-adjust process helping on the next attempt.

The timing observation shows that repeated attempts in this sample were usually close together. It does not show how frequently all Talkio learners practise, or whether feedback caused them to return. The rating observation shows what happened on two saved attempts, without establishing why the score changed.

A practical interpretation is to use feedback as a prompt for one focused next attempt, then test the same sound or expression in a new sentence. Rehearsing one sentence and communicating clearly in an unfamiliar situation remain different tasks.

Robustness checks

More frequent practisers can contribute more exercises. We therefore repeated the rating comparison by calculating each learner's average change first and giving each learner equal weight. The average change remained positive: approximately +0.19 rating points, compared with +0.15 when each exercise pair had equal weight.

Restricting the analysis to pairs in which the second retry occurred within 24 hours left the higher-rating proportion at 23.0%. These checks describe stability within the selected sample; they do not remove selection bias or make the analysis a controlled experiment.

Limitations

There was no control group, independent listener assessment or delayed follow-up. The analysis therefore does not establish lasting pronunciation improvement, transfer to spontaneous conversation, exam performance or superiority over another learning method.

Conclusion

In this privacy-protected sample, pronunciation scores improved on average on the subsequent retry, and higher scores were more common than lower scores. Most scores were unchanged, but the overall pattern supports repetition as a useful part of focused pronunciation practice: learners can apply feedback and make another attempt while the phrase is still fresh. The study does not prove that repetition caused the improvement or that a better next-attempt score becomes lasting pronunciation ability.