Towards few-shot isolated word reading assessment

Herman Kamper; Retief Louw; Reuben Smit

REVIEW 3 major objections 4 minor 35 references

Towards few-shot isolated word reading assessment

T0 review · 3 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Self-supervised speech features that score about 95% on adult-adult template matching reach only 65.3% balanced accuracy when child speech is matched against adult templates, showing they are not yet reliable enough for low-resource child…

desk verdict Solid, honest few-shot word assessment study; the negative-sample worry is real but doesn't sink the core claim about SSL robustness on child speech. read the letter →

arxiv 2507.12217 v1 pith:CIGXBSNV submitted 2025-07-16 cs.CL eess.AS

Reuben Smit , Retief Louw , Herman Kamper This is my paper · ORCID

classification cs.CLeess.AS

keywords readingassessmentchildspeechself-supervisedlearningfew-shotclassificationrepresentationsAfrikaanstemplatematchingdynamictimewarping

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper asks whether a handful of adult voice recordings can serve as templates for automatically judging whether a child read a word aloud correctly, without any speech recogniser for the target language. It builds a few-shot system that encodes both the child's input and the adult templates with a pretrained multilingual self-supervised model, then classifies the word as correct or incorrect by thresholding the matching distance. On adult inputs the approach works well, reaching balanced accuracy above 94%, but on child inputs it falls to 76.3% even with child templates and 65.3% with adult templates. The paper's central claim is that off-the-shelf SSL representations are not robust enough for child speech in a few-shot reading-assessment setting, and that child-specific SSL models or adaptation are needed.

What carries the argument

The central machinery is a few-shot template classifier. Each target word is represented by a small set of template recordings, and the input is compared to those templates after encoding both with a self-supervised speech model. Continuous frame-level features from mHuBERT-147 are compared with dynamic time warping using cosine distance, while discrete code sequences from k-means-quantised HuBERT-base are compared with normalised edit distance. To compress the templates, DTW barycentre averaging or edit-distance barycentre produces a single prototype per word. The system predicts 'correctly read' when the average distance from input to a word's templates falls below one threshold tuned on a development set. This mechanism lets the paper isolate how much of the child-adult gap comes from the representations themselves rather than from training data.

What would settle it

Collect a test set of genuine child mispronunciations of the target words, such as 'huis' said with a vowel error or 'wors' said as 'worsies', and run the mHuBERT template matcher with adult templates; if balanced accuracy on those true errors is close to 65.3%, the paper's limitation claim holds, and if it is substantially higher, the reported drop overstates the failure on real reading errors.

Watch

Extended reading notes

Core claim

The paper reports an ASR-free, few-shot isolated word reading assessment for low-resource Afrikaans. Using mHuBERT-147 continuous features with dynamic time warping and a single distance threshold, adult-adult template matching reaches a balanced accuracy of about 94.5–95%, whereas child inputs against adult templates reach at most 65.3% balanced accuracy with mHuBERT, and child inputs against child templates reach 76.3%. The drop is concentrated on words with phonetically similar impostors such as 'huis' versus 'muis', whose AUC falls below 70% in child-adult conditions. The paper interprets this as evidence that SSL representations encode adult speech with sufficient consistency but fail to tame the acoustic variability of child speech, and concludes that robust child-speech SSL models are the path to equitable educational technology.

Load-bearing premise

The evaluation treats negative samples as stand-ins for reading errors, but only three of the sixteen word classes have curated impostor words designed to sound like the target, so the measured child-adult gap may reflect general word discrimination rather than the specific failure to detect mispronunciations.

Editorial extensions

If this is right

A deployed few-shot reading assessment using adult templates and off-the-shelf SSL features would operate at roughly 65% balanced accuracy on child speech, which is too low for reliable classroom use.
The idealised adult-adult results suggest SSL features are sufficient for template matching when acoustic variability is low, so the bottleneck is specifically child speech rather than the template-matching method.
Discrete HuBERT units gave the best balanced accuracy in child-adult development experiments, but the gain was small and came with lower recall and AUC, so discretisation alone does not close the child-adult gap.
Fine-tuned ASR systems can show high AUC while having near-random balanced accuracy under a single threshold, meaning common reporting metrics can overstate real-world usefulness for reading assessment.
Techniques like barycentre averaging did not consistently improve child-adult performance, pointing to the input representation rather than template aggregation as the limiting factor.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

A classroom system would need per-word thresholds to reach the higher AUC-level performance the paper reports for ASR models, but tuning per-word thresholds requires child development data that the low-resource scenario does not supply, so the single-threshold balanced accuracy is the realistic operating estimate.
The paper's own proposed next step, adapting adult templates to child voices, is a direct test of whether the bottleneck is acoustic mismatch: if voice conversion closes most of the roughly 30-point gap, representation scarcity is not the main issue.
A fairer benchmark for reading assessment would use actual mispronunciations rather than other correctly spoken words as negatives; the current three curated impostor classes suggest the hardest cases are near-homophones, so real-error performance could differ from the headline 65.3%.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Bottom line: I largely agree with the reader's conditional verdict, and I think the stress-test note is half right. The central finding, that off-the-shelf SSL features degrade sharply on Afrikaans child speech (adult-adult ~95% balanced accuracy vs child-adult ~65%), holds up and is presented honestly. That is the real contribution: a new evaluation benchmark and a systematic comparison of continuous and discrete SSL features, DTW/NED, and barycentre averaging in a low-resource few-shot setting. Credit where due: the paper does not oversell its results, reports both balanced accuracy and AUC, and includes a sensible discussion of why a single global threshold is the realistic operating point.

The soft spots are real but proportionate. The stress-test is right that the negatives are mostly correct productions of other words, not genuine mispronunciations or near-miss errors. For a reading assessment system, that matters: distinguishing 'kat' from 'hond' is easier than catching a child who says 'wors' when the target is 'wors' but with a devoiced final consonant, or who produces a close non-word. The paper's own Table 4 supports this concern, since the three classes with curated impostors (nee*, huis*, wors*) show much lower child-child and child-adult AUC than classes like 'sien' and 'water'. However, that same table actually strengthens the paper's broader claim about SSL struggling with child acoustic variability; it makes the benchmark less valid as a reading-error-detection test, not the paper's conclusion about SSL representations. The authors should acknowledge this explicitly and, better, extend the benchmark with genuine mispronunciation negatives if they want the title to fully match the evaluation.

Other soft spots are as the reader says: small test sets (6-15 positives per class), a single adult male speaker for templates, no confidence intervals, and no released code/data. These limit reproducibility but are common in this kind of initial empirical study. The citation pattern looks fine; the authors lean on their own prior work in places, but that is appropriate given the corpus and the discrete-unit pipeline.

Who this is for: anyone building ASR-free reading assessment in low-resource languages, and the larger speech community working on child speech robustness. It is a useful caution and a reasonable proof of concept, not a breakthrough. I would bring it to a reading group and cite it if I were working on child speech or few-shot SSL evaluation. Worth a serious referee; a good reviewer could push the authors to clarify the negative-sample construction and release the benchmark, which would make it a solid contribution.

Referee Report

3 major / 4 minor

Summary. The paper proposes an ASR-free, template-based method for few-shot isolated word reading assessment. Speech inputs and reference templates are encoded with SSL features (mHuBERT continuous, HuBERT discrete, MFCC) and compared via DTW or edit distance, with optional barycentre averaging of templates. Experiments on an Afrikaans child-speech corpus show that adult-adult classification is strong (~95% balanced accuracy) but child-adult classification degrades to ~65%. The authors also compare to English Whisper ASR baselines and a topline Whisper fine-tuned on 5 hours of child speech. The paper concludes that SSL representations are not robust enough for child speech in this few-shot setting.

Significance. The paper provides a useful, reproducible benchmark for a practical low-resource scenario and honestly reports several strong baselines. The speaker-disjoint splits and fixed development-set threshold are sound practices. However, the benchmark's negatives are mostly correct productions of other words rather than genuine mispronunciations, so the reading-error assessment claim is not yet supported. The observed child-adult drop may partly reflect difficulty in word discrimination under acoustic variability. If the task were reframed as word discrimination, the significance is reduced; if reading-error detection is intended, the evaluation needs genuine error samples.

major comments (3)

[§3.1, Table 4] The negative samples for 13 of the 16 word classes are not documented as mispronunciations; only three classes ('huis', 'nee', 'wors') have manually curated impostors. The remaining negative samples appear to be correct productions of other words drawn from the child corpus. The task therefore measures word discrimination ('is this token the target word?') rather than reading-error detection ('did the child correctly produce the target word?'). Table 4 supports this concern: the classes with curated impostors show much lower child-child and child-adult AUC (nee* 55.6/44.4, huis* 51.2/41.3) than classes like sien (100.0/91.8), suggesting the measured child-adult gap is driven by acoustic variability on easy discriminations rather than by failure to catch near-miss reading errors. Please add genuine mispronunciation negatives for all classes, or explicitly reframe the contribution as word discrimination rather than reading assessment.
[§4.1 vs §4.2, Tables 2 and 3] The abstract and introduction compare the ~65% child-adult balanced accuracy (Table 3, test set) with the ~94-95% adult-adult accuracy (Table 2, development set). These numbers come from different data splits. Because the classification threshold is tuned on the development set in §3.3 and then frozen for the test set, part of the observed drop could be due to threshold mismatch or to the different speaker sets in the two splits. Please report adult-adult and child-child results on the same held-out test split used for Table 3, or at least quantify the development-to-test shift for each condition.
[§3.1, Table 3] Test classes contain only 6-15 positive and negative samples per class, across 16 classes, giving roughly 100-250 test items in total. No confidence intervals, bootstrap estimates, or significance tests are reported. The headline 65.3% balanced accuracy therefore has substantial uncertainty; a small number of misclassifications can move the result by several percentage points. Please add confidence intervals or per-class score distributions to support the claim that the child-adult drop is reliable.

minor comments (4)

[§3.3] The text defines the true positive rate as TP/(TN + FP); the correct formula is TP/(TP + FN). The false positive rate definition (FP/(FP + TN)) is correct. Please fix this typo in the ROC description.
[§3.1] The spelling 'impostor' and 'imposter' is used inconsistently (e.g., 'impostors' and 'imposter classes'); please standardize.
[Figure 2] The discrete code sequences are printed as long digit strings over colored rows; the mapping from colors to sound types is hard to read. Consider using clearly labeled bands or a larger legend.
[§5] The sentence 'In our follow-up work, we will look into using approaches like voice conversion to align the adult templates more closely to the child inputs [11]' cites reference [11], which is the child speech corpus paper; if the intended citation is for voice conversion, please replace it.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central child-versus-adult SSL robustness finding is an independent, testable empirical result, and no reported number is a fitted input renamed as a prediction.

full rationale

The paper's load-bearing claim is empirical: template-based verification with SSL features (mHuBERT) achieves roughly 94-95% balanced accuracy for adult-adult inputs, 76.3% for child-child, and 65.3% for child-adult on a held-out test set (Tables 2-3). The only fitted parameter is the decision threshold τ, which is tuned on the development set and then frozen as a single threshold across all word classes (Section 3.3); no model parameters are optimized on the test inputs, and the reported metrics follow a standard held-out evaluation. The comparisons include external systems (English Whisper-base, Whisper fine-tuned on the templates, and a topline ASR trained on 5 hours of child speech), so the main finding is not manufactured by a fitted input. The self-citations ([11] for the Afrikaans child corpus and [14,19] for discrete speech units) are background motivation or data provenance; the conclusion that SSL representations are not robust to child-speech variability is independently demonstrated by the reported accuracy/AUC breakdown, including the poor AUC on impostor classes (nee*, huis*, wors*) in Table 4. The benchmark's negative samples being correct productions of other words rather than genuine mispronunciations is a possible construct-validity limitation, but that concerns what the benchmark measures, not circularity of the derivation. No equation reduces to its own input, and no load-bearing argument depends on an unverified self-citation.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The paper's numbers rest on two external assets it did not create: mHuBERT-147 and the Afrikaans child oral narrative dataset, one of whose authors is also an author here. Within the paper, it introduces only a new evaluation setup, not a new entity. The threshold is the only fitted number, and it is tuned on a development split, which is standard but still a free choice.

free parameters (3)

classification threshold tau = not reported numerically
Selected per representation type by maximizing balanced accuracy on the development set (Section 3.3). All test-set numbers depend on this choice, and it is a genuine fitted quantity in the pipeline.
SSL feature extraction layer = unspecified
The abstract and Section 2.1 say intermediate layers are used, but the exact layer index is not given; different layers give different DTW distances.
DBA/EDB algorithm settings = iteration count not specified
Section 2.3 says DBA is repeated for a number of iterations and EDB runs until no improvement, without stating the number of iterations or convergence tolerance; these settings affect the barycentre templates.

assumptions (3)

domain assumption DTW with cosine distance and normalised edit distance are valid dissimilarity measures for comparing isolated word pronunciations.
Section 2.2 uses these distances to decide MATCH/NOMATCH. The paper does not validate that smaller distances correlate with correct readings rather than merely with speaker or channel similarity.
domain assumption The Afrikaans child oral narrative corpus and the manually extracted word samples are representative of child reading performance and contain genuine positive/negative examples.
Section 3.1 takes this dataset as ground truth; the negative samples are not characterized as mispronunciations, which is the weakest point.
domain assumption SSL features from models trained on mostly adult speech transfer their phonetic structure to Afrikaans, a language that is only partly seen during pretraining.
Section 2.1 relies on this transfer for both adult and child inputs; the paper's own results suggest transfer is strong for adults and weak for children.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Towards few-shot isolated word reading assessment." pith.science (2026). https://pith.science/paper/CIGXBSNV

@misc{pith2026250712217,
  author       = {Pith},
  title        = {Pith review of: Towards few-shot isolated word reading assessment},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CIGXBSNV}},
  note         = {Machine review of arXiv:2507.12217}
}

read the original abstract

We explore an ASR-free method for isolated word reading assessment in low-resource settings. Our few-shot approach compares input child speech to a small set of adult-provided reference templates. Inputs and templates are encoded using intermediate layers from large self-supervised learned (SSL) models. Using an Afrikaans child speech benchmark, we investigate design options such as discretising SSL features and barycentre averaging of the templates. Idealised experiments show reasonable performance for adults, but a substantial drop for child speech input, even with child templates. Despite the success of employing SSL representations in low-resource speech tasks, our work highlights the limitations of SSL representations for processing child data when used in a few-shot classification system.

Figures

Figures reproduced from arXiv: 2507.12217 by the authors.

**Figure 1.** Few-shot isolated word reading assessment. SSLencoded input from a child is compared to templates from an adult to determine whether the input is correctly read. (SSL) representations from an existing pretrained model. It has been shown that SSL models can provide robust features, even for unseen languages and domains [12, 13]. We use these representations to calculate a distance between each input sample and the … view at source ↗

**Figure 2.** Discrete codes for “boom” and “hond” for child and adult samples coloured by sound type. dicts that all words are read correctly – a meaningless prediction – giving perfect recall and random precision. This shows the problem with relying solely on F1: artificially inflated recall results in an inflated F1 score. The Whisper ASR model finetuned on roughly 2 minutes of templates achieves the highest AUC (87.9%). But,… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

35 extracted references · 33 canonical work pages

[1]

kat” “hond

Introduction One of the earliest goals when learning to read is to correctly pro- duce a familiar word when it is presented in isolation. However, even for this simple oral reading task, automatic assessment sys- tems are only available in a handful of the world’s well-resourced languages [1–6]. A major challenge is that child speech is inher- ently diffi...

work page
[2]

boom” and “hond

Methodology Our overall few-shot assessment approach is illustrated in Fig- ure 1. We test different features for encoding the inputs and templates, look at different distance measures based on the types of features used, and then consider ways to aggregate different templates of the same word into a single representative example. arXiv:2507.12217v1 [cs.C...

arXiv 2025
[3]

muis” as impostor for “huis

Experimental setup 3.1. Data We construct a few-shot evaluation benchmark using an existing dataset of Afrikaans child speech [11]. This dataset contains oral stories from four- and five-year-old children from low-income communities in the Western Cape region of South Africa. Al- though Afrikaans is a Germanic language with more than 7 million native spea...

work page
[4]

boom” “hond

Experiments and results To determine whether SSL representations are a viable option for automatic isolated word reading assessment in a few-shot setting, we examine model performance in three experiments. The first two are idealised scenarios: we test adult inputs using adult templates, and child inputs with child templates (which would normally not be a...

work page
[5]

Conclusion This paper presented an initial investigation into a few-shot isolated word reading assessment approach for low-resource child speech using SSL representations. We introduced a simple, template-based classification method and tested various represen- tations, including continuous and discrete self-supervised speech features, barycentre-averaged...

work page
[6]

This work was supported by a grant from the Het Jan Marais Fonds (HJMF)

Acknowledgements: We want to thank Danel Adendorff and Kyle Janse van Rensburg for their assistance with data collection. This work was supported by a grant from the Het Jan Marais Fonds (HJMF)

work page
[7]

Automatic assessment of expressive oral reading,

D. Bolanos, R. A. Cole, W. H. Ward, G. A. Tindal, P. J. Schwanen- flugel, and M. R. Kuhn, “Automatic assessment of expressive oral reading,” Speech Communication, 2013

work page 2013
[8]

Validation of automated scoring of oral reading,

J. Balogh, J. Bernstein, J. Cheng, A. Van Moere, B. Townshend, and M. Suzuki, “Validation of automated scoring of oral reading,” Educational and Psychological Measurement, 2012

work page 2012

Show all 35 references

[9]

A framework for large-scale automatic fluency assessment,

W. A. Silva, L. C. Carchedi, J. G. Junior, J. V . de Souza, E. Bar- rere, and J. F. de Souza, “A framework for large-scale automatic fluency assessment,” International Journal of Distance Education Technologies, 2021

2021
[10]

An ASR-based reading tutor for practicing reading skills in the first grade: Improv- ing performance through threshold adjustment,

Y . Bai, F. Hubers, C. Cucchiarini, and H. Strik, “An ASR-based reading tutor for practicing reading skills in the first grade: Improv- ing performance through threshold adjustment,” in IberSPEECH, 2021

2021
[11]

Reading accuracy assessment in first graders: Comparing assess- ments by teachers and machines,

W. Harmsen, F. Hubers, R. van Hout, C. Cucchiarini, and H. Strik, “Reading accuracy assessment in first graders: Comparing assess- ments by teachers and machines,” SSRN, 2025

2025
[12]

Can artificial intelligence identify reading flu- ency and level? Comparison of human and machine performance,

M. Yıldız, H. K. Keskin, S. Oyucu, D. K. Hartman, M. Temur, and M. Aydo˘gmus ¸, “Can artificial intelligence identify reading flu- ency and level? Comparison of human and machine performance,” Reading & Writing Quarterly, 2025

2025
[13]

Analysis of chil- dren’s speech: Duration, pitch and formants,

S. Lee, A. Potamianos, and S. S. Narayanan, “Analysis of chil- dren’s speech: Duration, pitch and formants,” in Eurospeech, 1997

1997
[14]

On the difficulties of automatic speech recognition for kindergarten-aged children,

G. Yeung and A. Alwan, “On the difficulties of automatic speech recognition for kindergarten-aged children,” in Interspeech, 2018

2018
[15]

Automatic speech recognition (ASR) systems for children: A systematic literature review,

V . Bhardwaj, M. T. Ben Othman, V . Kukreja, Y . Belkhier, M. Bajaj, B. S. Goud, A. U. Rehman, M. Shafiq, and H. Hamam, “Automatic speech recognition (ASR) systems for children: A systematic literature review,”Applied Sciences, 2022

2022
[16]

A survey about ASR for children

F. Claus, H. G. Rosales, R. Petrick, H.-U. Hain, and R. Hoffmann, “A survey about ASR for children.” inSLaTE, 2013

2013
[17]

Speech recognition for automatically assessing Afrikaans and isiXhosa preschool oral narratives,

C. Jacobs, A. Smith, D. Klop, O. Klejch, F. de Wet, and H. Kamper, “Speech recognition for automatically assessing Afrikaans and isiXhosa preschool oral narratives,” in ICASSP, 2025

2025
[18]

Acoustic word embed- dings for zero-resource languages using self-supervised contrastive learning and multilingual adaptation,

C. Jacobs, Y . Matusevych, and H. Kamper, “Acoustic word embed- dings for zero-resource languages using self-supervised contrastive learning and multilingual adaptation,” in SLT, 2021

2021
[19]

Unsupervised cross-lingual representation learning for speech recognition,

A. Conneau, A. Baevski, R. Collobert, A. Mohamed, and M. Auli, “Unsupervised cross-lingual representation learning for speech recognition,” in Interspeech, 2021

2021
[20]

A comparison of discrete and soft speech units for improved voice conversion,

B. Van Niekerk, M.-A. Carbonneau, J. Za¨ıdi, M. Baas, H. Seut´e, and H. Kamper, “A comparison of discrete and soft speech units for improved voice conversion,” inICASSP, 2022

2022
[21]

Mel frequency cepstral coef- ficient and its applications: A review,

Z. K. Abdul and A. K. Al-Talabani, “Mel frequency cepstral coef- ficient and its applications: A review,”IEEE Access, 2022

2022
[22]

HuBERT: Self-supervised speech representa- tion learning by masked prediction of hidden units,

W.-N. Hsu, B. Bolte, Y .-H. H. Tsai, K. Lakhotia, R. Salakhutdinov, and A. Mohamed, “HuBERT: Self-supervised speech representa- tion learning by masked prediction of hidden units,” IEEE/ACM Transactions on Audio, Speech, and Language Processing, 2021

2021
[23]

WavLM: Large-scale self-supervised pre-training for full stack speech processing,

S. Chen, C. Wang, Z. Chen, Y . Wu, S. Liu, Z. Chen, J. Li, N. Kanda, T. Yoshioka, X. Xiao, J. Wu, L. Zhou, S. Ren, Y . Qian, Y . Qian, M. Zeng, X. Yu, and F. Wei, “WavLM: Large-scale self-supervised pre-training for full stack speech processing,” Journal of Selected Topics in ...

2022
[24]

mHuBERT-147: A Compact Multilingual HuBERT Model,

M. Z. Boito, V . Iyer, N. Lagos, L. Besacier, and I. Calapodescu, “mHuBERT-147: A Compact Multilingual HuBERT Model,” in Interspeech, 2024

2024
[25]

Spoken-term discovery using discrete speech units,

B. van Niekerk, J. Za ¨ıdi, M.-A. Carbonneau, and H. Kamper, “Spoken-term discovery using discrete speech units,” inInterspeech, 2024

2024
[26]

Are discrete units neces- sary for spoken language modeling?

T. A. Nguyen, B. Sagot, and E. Dupoux, “Are discrete units neces- sary for spoken language modeling?” Journal of Selected Topics in Signal Processing, 2022

2022
[27]

A global averaging method for dynamic time warping, with applications to clustering,

F. Petitjean, A. Ketterlin, and P. Ganc ¸arski, “A global averaging method for dynamic time warping, with applications to clustering,” Pattern Recognition, 2011

2011
[28]

Median strings,

T. Kohonen, “Median strings,” Pattern Recognition Letters, 1985

1985
[29]

Robust speech recognition via large-scale weak su- pervision,

A. Radford, J. W. Kim, T. Xu, G. Brockman, C. McLeavey, and I. Sutskever, “Robust speech recognition via large-scale weak su- pervision,” in ICML, 2023

2023
[30]

Scikit-learn: Machine learning in python,

F. Pedregosa, G. Varoquaux, A. Gramfort, V . Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V . Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, and ´E. Duchesnay, “Scikit-learn: Machine learning in python,” Journal of machine lea...

2011
[31]

Target speaker ASR with Whisper,

A. Polok, D. Klement, M. Wiesner, S. Khudanpur, J. ˇCernock`y, and L. Burget, “Target speaker ASR with Whisper,” in ICASSP, 2025

2025
[32]

Adapt- ing Whisper for regional dialects: Enhancing public services for vulnerable populations in the United Kingdom,

M. Torgbi, A. Clayman, J. J. Speight, and H. T. Madabushi, “Adapt- ing Whisper for regional dialects: Enhancing public services for vulnerable populations in the United Kingdom,” in COLING, 2025

2025
[33]

D. W. Hosmer Jr, S. Lemeshow, and R. X. Sturdivant, Applied Logistic Regression. John Wiley & Sons, 2013

2013
[34]

An analysis of the causes of increased error rates in children’s speech recognition,

Q. Li and M. Russell, “An analysis of the causes of increased error rates in children’s speech recognition,” inICSLP, 2002

2002
[35]

A study of speech recognition for children and the elderly,

J. G. Wilpon and C. N. Jacobsen, “A study of speech recognition for children and the elderly,” inICASSP, 1996

1996

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [1]

kat” “hond

Introduction One of the earliest goals when learning to read is to correctly pro- duce a familiar word when it is presented in isolation. However, even for this simple oral reading task, automatic assessment sys- tems are only available in a handful of the world’s well-resourced languages [1–6]. A major challenge is that child speech is inher- ently diffi...

work page

[2] [2]

boom” and “hond

Methodology Our overall few-shot assessment approach is illustrated in Fig- ure 1. We test different features for encoding the inputs and templates, look at different distance measures based on the types of features used, and then consider ways to aggregate different templates of the same word into a single representative example. arXiv:2507.12217v1 [cs.C...

arXiv 2025

[3] [3]

muis” as impostor for “huis

Experimental setup 3.1. Data We construct a few-shot evaluation benchmark using an existing dataset of Afrikaans child speech [11]. This dataset contains oral stories from four- and five-year-old children from low-income communities in the Western Cape region of South Africa. Al- though Afrikaans is a Germanic language with more than 7 million native spea...

work page

[4] [4]

boom” “hond

Experiments and results To determine whether SSL representations are a viable option for automatic isolated word reading assessment in a few-shot setting, we examine model performance in three experiments. The first two are idealised scenarios: we test adult inputs using adult templates, and child inputs with child templates (which would normally not be a...

work page

[5] [5]

Conclusion This paper presented an initial investigation into a few-shot isolated word reading assessment approach for low-resource child speech using SSL representations. We introduced a simple, template-based classification method and tested various represen- tations, including continuous and discrete self-supervised speech features, barycentre-averaged...

work page

[6] [6]

This work was supported by a grant from the Het Jan Marais Fonds (HJMF)

Acknowledgements: We want to thank Danel Adendorff and Kyle Janse van Rensburg for their assistance with data collection. This work was supported by a grant from the Het Jan Marais Fonds (HJMF)

work page

[7] [7]

Automatic assessment of expressive oral reading,

D. Bolanos, R. A. Cole, W. H. Ward, G. A. Tindal, P. J. Schwanen- flugel, and M. R. Kuhn, “Automatic assessment of expressive oral reading,” Speech Communication, 2013

work page 2013

[8] [8]

Validation of automated scoring of oral reading,

J. Balogh, J. Bernstein, J. Cheng, A. Van Moere, B. Townshend, and M. Suzuki, “Validation of automated scoring of oral reading,” Educational and Psychological Measurement, 2012

work page 2012