Pith. sign in

REVIEW 4 major objections 5 minor 46 references

Discrete Tokens Exhibit Interlanguage Speech Intelligibility Benefit: an Analytical Study Towards Accent-robust ASR Only with Native Speech Data

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Training the discrete tokenizer on the speaker's native language, rather than on the target language, improves ASR word error rates on accented English, reproducing a human perceptual advantage.

desk verdict Decisive margins are small and confounded by a generic non-English-tokenizer effect, but the core observation is new and worth referee time. read the letter →

arxiv 2505.16182 v1 pith:CG4Q5EC4 submitted 2025-05-22 cs.SD eess.AS

Kentaro Onda , Keisuke Imoto , Satoru Fukayama , Daisuke Saito , Nobuaki Minematsu This is my paper · ORCID
classification cs.SDeess.AS
keywords discretetokensself-supervisedlearningforeignaccentinterlanguagespeechintelligibilitybenefitautomaticrecognitionk-meansclusteringrobustnessnativeonly
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

This paper tries to establish that a machine speech recognizer can reproduce the interlanguage speech intelligibility benefit (ISIB), the human tendency to understand accented speech better when the listener shares the speaker's native language. The lever is the discrete tokenizer used to convert speech into a token sequence: train its k-means centroids on native speech of the speaker's language, and the recognizer becomes more accurate for that accent even though the ASR itself is trained only on native English. Experiments with Japanese-, Chinese-, and Spanish-accented English show that the matched-L1 tokenizer yields the best word error rate among the tested tokenizer languages, including English. If the claim is right, accent-robust ASR can be built for language pairs where no accented recordings exist, using only native speech from both languages.

What carries the argument

The central object is the discrete tokenizer: k-means clustering applied to hidden-layer features of a HuBERT model, producing an integer sequence from each utterance. Changing the language used to train the k-means centroids changes the set of phonetic categories the tokens can express, and the paper treats this as a model of a native listener's phonological filter. The ASR model is kept fixed and trained on native English, so any performance shift caused by changing the tokenizer language is attributed to the perceptual filter imposed before recognition.

What would settle it

A concrete falsifier would be to train a tokenizer on native speech of the same L1 but with the language-specific phonemes artificially neutralized through resynthesis, and check whether the matched-L1 WER gain disappears; if the gain persists, the effect is generic centroid robustness rather than a simulated native phonological filter.

Watch

Extended reading notes

Core claim

The central claim is that discrete tokens extracted from a self-supervised speech model carry a listener-like native-language filter, so that tokenizing X-accented English with a tokenizer trained on native X speech improves downstream recognition in the same way ISIB improves human perception. On LibriSpeech960 with 2000 clusters, Japanese-trained tokenization lowered the word error rate on Japanese-accented English by 2.4 to 4.7 points absolute compared with English-trained tokenization, while the English-trained tokenizer remained best for native English. The same matched-L1 advantage appeared for Chinese- and Spanish-accented English on the L2-ARCTIC corpus, and mismatched cases showed Chinese or Spanish tokenizers outperforming the English tokenizer on all non-native English test sets. The paper reads these results as evidence that discrete tokens simulate human speech perception and that ISIB has a computational analogue that can be exploited with only native speech data.

Load-bearing premise

The load-bearing premise is that k-means clusters trained on native speech of a language faithfully simulate how a native speaker of that language perceives speech, so a tokenizer trained on language X is a valid stand-in for an X-speaking listener.

Editorial extensions

If this is right

  • If the central claim holds, accent-robust ASR for an X-accented Y pair can be built without any accented recordings by training the ASR on native Y speech and the tokenizer on native X speech.
  • Because mismatched ISIB also occurs, a high-resource language whose accent resembles the speaker's can stand in when the speaker's native language is low-resource or unknown.
  • The matched-L1 tokenizer advantage strengthens the hypothesis that SSL discrete tokens encode a listener-like perceptual filter.
  • Cluster size matters: the ISIB gain appears only with sufficiently large codebooks (500 or 2000), suggesting that small codebooks erase the perceptual detail the effect depends on.
  • The approach changes only the tokenizer, leaving the ASR architecture and training data untouched, so it can be combined with other accent-robustness techniques.

Reading between the lines

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

  • An extension the paper leaves implicit is per-speaker tokenizer selection: if the speaker's native language is known at inference time, the tokenizer could be switched on the fly without retraining the ASR.
  • The paper's accent-similarity explanation for mismatched ISIB could be turned into a predictive rule: pick the tokenizer language that maximizes acoustic similarity to the speaker's accent, estimated from an accent-classification model.
  • Since the SSL pretraining language was fixed to English in the multi-accent experiments, a natural next test is whether matching both the SSL model and the tokenizer to the speaker's L1 gives a larger gain than matching the tokenizer alone.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. This manuscript reports an analytical study of discrete-token ASR for foreign-accented English. The authors train k-means tokenizers on native English, Japanese, Chinese, or Spanish speech and feed the resulting discrete units to an English ASR system trained on LibriSpeech. They report that for Japanese-accented English (ERJ), a Japanese-trained tokenizer outperforms an English-trained tokenizer, and in a multi-accent experiment (L2-ARCTIC), the tokenizer trained on the speaker's native language gives the best WER among the four tokenizers. They interpret this as evidence for an interlanguage speech intelligibility benefit in machine perception and as support for the hypothesis that SSL discrete tokens simulate human speech perception, with the stated applicative goal of accent-robust ASR using only native speech data.

Significance. If the central pattern is robust, the paper would be a useful first demonstration that tokenizer training language can shift ASR robustness for non-native speech without accented training data, and it would provide a concrete, falsifiable prediction derived from the discrete-token-as-perception hypothesis. Strengths include the systematic variation of cluster size, SSL layer, and tokenizer language; the inclusion of both a matched-L1 condition (Japanese/English) and a multi-accent extension; and the use of publicly available corpora. However, the decisive comparisons rest on small single-run WER differences, and the design does not currently separate L1-specific matching from a generic robustness advantage of non-English tokenizers. The significance of the paper depends on closing that gap.

major comments (4)
  1. [Section 3.5, Table 5] The matched-ISIB claim rests on WER gaps of 0.6–0.7 points for Chinese- and Spanish-accented English (28.6 vs 29.3 and 21.8 vs 22.4), with no confidence intervals, significance tests, or multiple seeds reported. Given that Table 2 shows en–jp differences on native English of –1.1 to +0.2, differences of this size are within plausible training noise, so the paper's conclusion that recognition accuracy was highest for every matched case is not statistically supported.
  2. [Section 3.5, Table 5] The mismatched columns show that Chinese and Spanish tokenizers improve WER over the English tokenizer not only for Chinese- and Spanish-accented English but also for Arabic, Hindi, Korean, and Vietnamese accented English—languages whose L1s are absent from the tokenizer set. This is evidence of a generic non-English-tokenizer robustness effect rather than L1-specific perception. The matched diagonal could be a byproduct of this general effect plus small fluctuations, so the ISIB attribution requires a control such as matching tokenizers on effective cluster usage or entropy, or comparing with a non-linguistic quantization baseline.
  3. [Section 2.2, Figure 2] The load-bearing premise is that a k-means tokenizer trained on native X speech is a faithful computational model of a native-X listener's perception. The manuscript adopts this premise from [10] but provides no independent evidence for it; the experiments only show that certain tokenizer languages yield lower WER. To avoid circularity, the authors should either test the perception link directly (e.g., by predicting human intelligibility judgments or by showing that the effect is specific to L1-matched phonology rather than to any out-of-domain tokenizer) or explicitly frame the result as a technical ISIB analog without claiming to validate the perceptual hypothesis.
  4. [Section 3.4, Table 4] The cluster-quality metrics do not consistently support the proposed mechanism. QE is lower for Japanese-trained k-means only for JE w10 at the 9th layer, and MTER is lower for Japanese-trained k-means only at cluster sizes 100 and 500 for JE w10, not at cluster size 2000 where the main ASR comparison is made. Thus the claim that L1-matched tokenization gives more adequate units for accented speech is only partially corroborated and does not explain the ASR result at the operating cluster size.
minor comments (5)
  1. [Section 2.1, Figure 1] There are typos: 'exracted' should be 'extracted' in Section 2.1, and 'expeted' should be 'expected' in the Figure 1 caption.
  2. [Table 3] The recognition examples 'app liciated' and 'app reciated' appear to have spacing issues; please format the examples with monospaced or aligned text for clarity.
  3. [Section 3.5] The evaluation on L2-ARCTIC is not described in the same detail as ERJ; please specify which speakers and sentences were used and the size of each accent-specific test subset.
  4. [Section 3.1] The paper does not state how many random seeds were used for k-means initialization or ASR training; at minimum, report the seed or the k-means initialization procedure.
  5. [References] The reference list is appropriate, but [10] is the authors' own prior work; please clarify in the text which aspects of the perception model are established there and which are newly tested here.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the matched-L1 WER advantage is an empirical outcome, not a fitted or definitional consequence.

full rationale

The paper's derivation chain is an experimental comparison, not a mathematical derivation. For each tokenizer-language condition, the k-means model is trained on native speech and the ASR (trained on LibriSpeech) is evaluated on held-out accented English; no parameter is fitted to the reported WERs, and the matched-L1 pattern could have failed (and does fail at cluster size 100 in Table 1). The paper's use of its own prior work [10] supplies the interpretive hypothesis that discrete tokens simulate human perception, but the paper explicitly treats this as a hypothesis to validate ('This could also validate the hypothesis...'), and the ISIB observation is independent evidence for that hypothesis rather than a consequence of citing it. There is no uniqueness theorem, no ansatz smuggled through citation as a forced choice, and no equation in the paper that makes the predicted WER equal to an input. Concerns about small WER margins, lack of significance testing, and the alternative explanation of generic non-English-tokenizer robustness are statistical/correctness risks, not circularity.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The empirical claim rests on the perceptual-simulation premise inherited from prior GSLM work, on the choice of cluster size, layer, and 30-hour subsets, and on the assumption that English-HuBERT features are a common space for all tokenizers. No new entities are introduced.

free parameters (3)
  • k-means cluster size = 2000 (500 in earlier experiments)
    ISIB for Japanese-accented English is not observed at 100 clusters and appears only at 500 or 2000, so the central result depends on this hand-set hyperparameter.
  • HuBERT feature layer = 12th layer for Tables 1/4; 9th layer for Table 5
    Layer choice changes the size of the ISIB effect; Table 5 uses layer 9 because it gave the best native-English WER on validation, which is a model selection choice.
  • k-means training subset size = approximately 30 hours per language
    The exact random subset is not pinned by seed or file list, and the strength of the effect may depend on this data selection.
assumptions (4)
  • domain assumption Discrete tokens from SSL models simulate human speech perception, so a k-means tokenizer trained on language X represents how a native X speaker perceives speech.
    Inherited from the authors' prior work [10], used in Sections 2.2 and 3.1; all ISIB interpretations depend on this mapping.
  • domain assumption English-pretrained HuBERT features provide a common acoustic space suitable for training k-means tokenizers on Japanese, Chinese, and Spanish speech.
    Used in Sections 3.3 and 3.5, where the SSL model is fixed to English while the k-means language varies.
  • domain assumption The ERJ segmental pronunciation scores reliably identify the strongest Japanese accents, so the JE w10 subset is a valid high-accent group.
    Used in Sections 3.2 and 3.4; no independent validation of these scores is provided.
  • domain assumption A joint CTC/attention ASR trained on English-native tokens can decode streams from other-language tokenizers without additional adaptation.
    The experimental design in Sections 3.2 to 3.5 trains ASR only on LibriSpeech, so the token stream must transfer; no adaptation is performed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Discrete Tokens Exhibit Interlanguage Speech Intelligibility Benefit: an Analytical Study Towards Accent-robust ASR Only with Native Speech Data." pith.science (2026). https://pith.science/paper/CG4Q5EC4

@misc{pith2026250516182,
  author       = {Pith},
  title        = {Pith review of: Discrete Tokens Exhibit Interlanguage Speech Intelligibility Benefit: an Analytical Study Towards Accent-robust ASR Only with Native Speech Data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CG4Q5EC4}},
  note         = {Machine review of arXiv:2505.16182}
}
read the original abstract

In this study, we gained insight that contributes to achieving accent-robust ASR using only native speech data. In human perception of non-native speech, the phenomenon known as "interlanguage speech intelligibility benefit" (ISIB) is observed, where non-native listeners who share the native language with the speaker understand the speech better compared even to native listeners. Based on the idea that discrete tokens extracted from self-supervised learning (SSL) models represent the human perception of speech, we conducted an analytical study on the robustness of discrete token-based ASR to non-native speech, varying the language used for training the tokenization, which is viewed as a technical implementation of ISIB. The results showed that ISIB actually occurred in the discrete token-based ASR. Since our approach relies only on native speech data to simulate the behavior of human perception, it is expected to be applicable to a wide range of accents for which speech data is scarce.

Figures

Figures reproduced from arXiv: 2505.16182 by the authors.

Figure 1
Figure 1. The reproduction of native speech of language Y by (a) native speakers of language Y, (b) native speakers of language X: the output speech of (b) is expeted to be “X-accented Y”. … … (a) (b) Perception Recognition X-accented Y This is light. This is right. Recognition by Native speaker of Y Recognition by Native speaker of X Discrete Tokens of Y Discrete Tokens of X [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The recognition of “X-accented Y” speech by (a) na￾tive speakers of language Y, (b) native speakers of language X: (b) is expected to be more accurate through the ISIB effect. 2. Related studies 2.1. Generative Spoken Language Model Generative Spoken Language Model (GSLM) [14] is a lan￾guage model trained only with speech data without any text. It was proposed under the concept of “textless NLP,” based on the idea t… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

46 extracted references · 41 canonical work pages

  1. [10]

    Foreign accent, comprehensi- bility, and intelligibility in the speech of second language learn- ers,

    M. J. Munro and T. M. Derwing, “Foreign accent, comprehensi- bility, and intelligibility in the speech of second language learn- ers,” Language Learning, vol. 45, no. 1, pp. 73–97, 1995

    work page 1995

  2. [1]

    Discrete Tokens Exhibit Interlanguage Speech Intelligibility Benefit: an Analytical Study Towards Accent-robust ASR Only with Native Speech Data

    Introduction Automatic speech recognition (ASR) is widely used in ap- plications such as voice-controlled devices and transcription. ASR systems are required to be robust against audio varia- tions caused by factors such as speakers and recording envi- ronments. However, it is well known that ASR performance tends to degrade when recognition is performed ...

    work page Pith review arXiv 2025

  3. [2]

    textless NLP,

    Related studies 2.1. Generative Spoken Language Model Generative Spoken Language Model (GSLM) [14] is a lan- guage model trained only with speech data without any text. It was proposed under the concept of “textless NLP,” based on the idea that text is not always necessary for human lan- guage acquisition. In GSLM, “pseudo-text” is used as a sub- stitute ...

    work page

  4. [3]

    X-accented Y

    Empirical analyses Figure 2 illustrates an overview of discrete token-based ASR for foreign-accented speech, aiming to simulate ISIB. Based on the concept of [10], shown in the section 2.2, discrete tokens are regarded as results of human perception. When recogniz- ing “X-accented Y” speech, by using discrete tokens trained on language X, the native langu...

    work page

  5. [4]

    mismatched

    Conclusions In this study, we conducted comparative experiments on dis- crete token-based ASR for foreign-accented speech varying the language used for k-means clustering. The results showed that training k-means on the speaker’s native language led to the best recognition performance among the tested languages, including the spoken language. This shows t...

    work page

  6. [5]

    A smaller value indi- cates that the discretization is more suitable for the given data

    Quantization Error (QE): The squared Euclidean distance between the original SSL features and the centroids of the clus- ters to which the features are assigned. A smaller value indi- cates that the discretization is more suitable for the given data

    work page

  7. [6]

    mismatched

    Mean Token Error Rate (MTER): The average of token error rate (TER) computed across all possible pairs of multi- ple utterances of the same content. TER is defined as the edit distance between two discrete token sequences generated from two utterances of the same content. A lower MTER suggests that the tokenization is more robust to non-linguistic variati...

    work page 2000

  8. [7]

    Accented speech recognition with accent-specific codebooks,

    D. Prabhu, P. Jyothi, S. Ganapathy, and V . Unni, “Accented speech recognition with accent-specific codebooks,” in EMNLP 2023 , 2023, pp. 7175–7188

    work page 2023

Show all 46 references
  1. [8]

    Acknowledgements This work was supported by AIST KAKUSEI project (FY2024)

  2. [9]

    Racial disparities in automated speech recognition,

    A. Koenecke, A. Nam, E. Lake, J. Nudell, M. Quartey, Z. Menge- sha, C. Toups, J. R. Rickford, D. Jurafsky, and S. Goel, “Racial disparities in automated speech recognition,” PNAS, vol. 117, no. 14, pp. 7684–7689, 2020

    2020

  3. [11]

    Intelligible, comprehensible, non-native models in ESL/EFL pronunciation teaching,

    J. M. Murphy, “Intelligible, comprehensible, non-native models in ESL/EFL pronunciation teaching,” System, vol. 42, pp. 258–269, 2014

    2014

  4. [12]

    Revisiting the intelligibility and nativeness principles,

    J. Levis, “Revisiting the intelligibility and nativeness principles,” Journal of Second Language Pronunciation, vol. 6, no. 3, pp. 310– 328, 2020

    2020

  5. [13]

    Data augmentation improves recognition of foreign accented speech,

    T. Fukuda, R. Fernandez, A. Rosenberg, S. Thomas, B. Ramab- hadran, A. Sorin, and G. Kurata, “Data augmentation improves recognition of foreign accented speech,” in Interspeech 2018 , 2018, pp. 2409–2413

    2018

  6. [14]

    Synthetic cross-accent data augmentation for automatic speech recognition,

    P. Klumpp, P. Chitkara, L. Sari, P. Serai, J. Wu, I.-E. Veliche, R. Huang, and Q. He, “Synthetic cross-accent data augmentation for automatic speech recognition,” in arXiv, 2023, p. 2303.00802

    2023 arXiv

  7. [15]

    HuBERT: Self-supervised speech repre- sentation learning by masked prediction of hidden units,

    W.-N. Hsu, B. Bolte, Y .-H. H. Tsai, K. Lakhotia, R. Salakhutdi- nov, and A. Mohamed, “HuBERT: Self-supervised speech repre- sentation learning by masked prediction of hidden units,”T-ASLP, vol. 29, pp. 3451–3460, 2021

    2021

  8. [16]

    Im- proving self-supervised pre-training using accent-specific code- books,

    D. Prabhu, A. Gupta, O. Nitsure, P. Jyothi, and S. Ganapathy, “Im- proving self-supervised pre-training using accent-specific code- books,” in Interspeech 2024, 2024, pp. 2310–2314

    2024

  9. [17]

    Accented speech recognition: A survey,

    A. Hinsvark, N. Delworth, M. D. Rio, Q. McNamara et al. , “Accented speech recognition: A survey,” in arXiv, 2021, p. 2104.10747

    2021 arXiv

  10. [18]

    A pilot study of GSLM-based simulation of foreign accentuation only using native speech corpora,

    K. Onda, J. Park, N. Minematsu, and D. Saito, “A pilot study of GSLM-based simulation of foreign accentuation only using native speech corpora,” in Interspeech 2024, 2024, pp. 3600–3604

    2024

  11. [19]

    The interlanguage speech intelligi- bility benefit,

    T. Bent and A. R. Bradlow, “The interlanguage speech intelligi- bility benefit,” JASA, vol. 114, no. 3, pp. 1600–1610, 2003

    2003

  12. [20]

    Accent, listening assessment and the potential for a shared-L1 advantage: A DIF perspective,

    L. Harding, “Accent, listening assessment and the potential for a shared-L1 advantage: A DIF perspective,” Language Testing, vol. 29, no. 2, pp. 163–180, 2012

    2012

  13. [21]

    Listening with a foreign-accent: The interlanguage speech intelligibility benefit in Mandarin speakers of english,

    X. Xie and C. A. Fowler, “Listening with a foreign-accent: The interlanguage speech intelligibility benefit in Mandarin speakers of english,” Journal of Phonetics , vol. 41, no. 5, pp. 369–378, 2013

    2013

  14. [22]

    On gener- ative spoken language modeling from raw audio,

    K. Lakhotia, E. Kharitonov, W.-N. Hsu, Y . Adi et al., “On gener- ative spoken language modeling from raw audio,” TACL, vol. 9, pp. 1336–1354, 2021

    2021

  15. [23]

    How should we extract discrete audio tokens from self-supervised models?

    P. Mousavi, J. Duret, S. Zaiem, L. Della Libera, A. Ploujnikov, C. Subakan, and M. Ravanelli, “How should we extract discrete audio tokens from self-supervised models?” in Interspeech 2024, 2024, pp. 2554–2558

    2024

  16. [24]

    wav2vec 2.0: A framework for self-supervised learning of speech representa- tions,

    A. Baevski, Y . Zhou, A. Mohamed, and M. Auli, “wav2vec 2.0: A framework for self-supervised learning of speech representa- tions,” in NeurIPS 2020, vol. 33, 2020, pp. 12 449–12 460

    2020

  17. [25]

    The Interspeech 2024 challenge on speech processing using discrete units,

    X. Chang, J. Shi, J. Tian, Y . Wu, Y . Tang, Y . Wu, S. Watanabe, Y . Adi, X. Chen, and Q. Jin, “The Interspeech 2024 challenge on speech processing using discrete units,” in Interspeech 2024, 2024, pp. 2559–2563

    2024

  18. [26]

    Second language speech learning: Theory, findings, and problems,

    J. E. Flege, “Second language speech learning: Theory, findings, and problems,” Speech perception and linguistic experience: Is- sues in cross-language research, vol. 92, pp. 233–277, 1995

    1995

  19. [27]

    appre- ciated

    was used for Japanese. For evaluation, the ERJ corpus [28] was used. This corpus consists of English speech read by both Japanese learners and native speakers of American English. In our experiments, 460 phoneme-balanced sentences were used. For Japanese speakers’ speech, pron...

  20. [28]

    Ex- ploration of efficient end-to-end ASR using discretized input from self-supervised learning,

    X. Chang, B. Yan, Y . Fujita, T. Maekaku, and S. Watanabe, “Ex- ploration of efficient end-to-end ASR using discretized input from self-supervised learning,” in Interspeech 2023 , 2023, pp. 1399– 1403

    2023

  21. [29]

    Exploring speech recognition, translation, and understanding with discrete speech units: A comparative study,

    X. Chang, B. Yan, K. Choi, J.-W. Jung et al., “Exploring speech recognition, translation, and understanding with discrete speech units: A comparative study,” in ICASSP 2024, 2024, pp. 11 481– 11 485

    2024

  22. [30]

    Towards universal speech discrete tokens: A case study for ASR and TTS,

    Y . Yang, F. Shen, C. Du, Z. Ma, K. Yu, D. Povey, and X. Chen, “Towards universal speech discrete tokens: A case study for ASR and TTS,” in ICASSP 2024, 2024, pp. 10 401–10 405

    2024

  23. [31]

    MMM: Multi-layer multi-residual multi-stream discrete speech represen- tation from self-supervised learning model,

    J. Shi, X. Ma, H. Inaguma, A. Sun, and S. Watanabe, “MMM: Multi-layer multi-residual multi-stream discrete speech represen- tation from self-supervised learning model,” in Interspeech 2024, 2024, pp. 2569–2573

    2024

  24. [32]

    ESPnet: End- to-end speech processing toolkit,

    S. Watanabe, T. Hori, S. Karita, T. Hayashi et al., “ESPnet: End- to-end speech processing toolkit,” in Interspeech 2018, 2018, pp. 2207–2211

    2018

  25. [33]

    Joint CTC-attention based end-to-end speech recognition using multi-task learning,

    S. Kim, T. Hori, and S. Watanabe, “Joint CTC-attention based end-to-end speech recognition using multi-task learning,” in ICASSP 2017, 2017, pp. 4835–4839

    2017

  26. [34]

    Lib- rispeech: An ASR corpus based on public domain audio books,

    V . Panayotov, G. Chen, D. Povey, and S. Khudanpur, “Lib- rispeech: An ASR corpus based on public domain audio books,” in ICASSP 2015, 2015, pp. 5206–5210

    2015

  27. [35]

    JVS corpus: free Japanese multi-speaker voice corpus,

    S. Takamichi, K. Mitsui, Y . Saito, T. Koriyama, N. Tanji, and H. Saruwatari, “JVS corpus: free Japanese multi-speaker voice corpus,” 2019

    2019

  28. [36]

    Development of English speech database read by Japanese to support CALL research,

    N. Minematsu, Y . Tomiyama, K. Yoshimoto, K. Shimizu, S. Nak- agawa, M. Dantsuji, and S. Makino, “Development of English speech database read by Japanese to support CALL research,” in ICA 2004, 2004, pp. 557–560

    2004

  29. [37]

    Investigation of robust- ness of hubert features from different layers to domain, accent and language variations,

    P. Kumar, V . N. Sukhadia, and S. Umesh, “Investigation of robust- ness of hubert features from different layers to domain, accent and language variations,” in ICASSP 2022, 2022, pp. 6887–6891

    2022

  30. [38]

    Training Japanese listeners to identify En- glish/r/and/l: Iv. some effects of perceptual learning on speech production,

    A. R. Bradlow, D. B. Pisoni, R. Akahane-Yamada, and Y . Tohkura, “Training Japanese listeners to identify En- glish/r/and/l: Iv. some effects of perceptual learning on speech production,” JASA, vol. 101, no. 4, pp. 2299–2310, 1997

    1997

  31. [39]

    The sociolinguistics of variety identifica- tion and categorisation: free classification of varieties of spo- ken English amongst non-linguist listeners,

    R. M. McKenzie, “The sociolinguistics of variety identifica- tion and categorisation: free classification of varieties of spo- ken English amongst non-linguist listeners,” Language Aware- ness, vol. 24, no. 2, pp. 150–168, 2015

    2015

  32. [40]

    Perception of foreign accent in spontaneous L2 English speech,

    J. Yuan, Y . Jiang, and Z. Song, “Perception of foreign accent in spontaneous L2 English speech,” in Speech Prosody 2010, 2010

    2010

  33. [41]

    The interlanguage speech in- telligibility benefit as bias toward native-language phonology,

    H. Wang and V . J. Van Heuven, “The interlanguage speech in- telligibility benefit as bias toward native-language phonology,” i- Perception, vol. 6, no. 6, p. 2041669515613661, 2015

    2015

  34. [42]

    Evidence against the mismatched interlanguage speech intelligibility benefit hypothesis,

    R. M. Stibbard and J.-I. Lee, “Evidence against the mismatched interlanguage speech intelligibility benefit hypothesis,”JASA, vol. 120, no. 1, pp. 433–442, 2006

    2006

  35. [43]

    AISHELL-1: An open-source Mandarin speech corpus and a speech recognition baseline,

    H. Bu, J. Du, X. Na, B. Wu, and H. Zheng, “AISHELL-1: An open-source Mandarin speech corpus and a speech recognition baseline,” in O-COCOSDA 2017, 2017, pp. 1–5

    2017

  36. [44]

    MLS: A large-scale multilingual dataset for speech research,

    V . Pratap, Q. Xu, A. Sriram, G. Synnaeve, and R. Collobert, “MLS: A large-scale multilingual dataset for speech research,” in Interspeech 2020, 2020, pp. 2757–2761

    2020

  37. [45]

    L2-ARCTIC: A non-native English speech corpus,

    G. Zhao, S. Sonsaat, A. Silpachai, I. Lucic, E. Chukharev- Hudilainen, J. Levis, and R. Gutierrez-Osuna, “L2-ARCTIC: A non-native English speech corpus,” inInterspeech 2018, 2018, pp. 2783–2787

    2018

  38. [46]

    What can an accent identifier learn? probing phonetic and prosodic information in a wav2vec2-based accent identification model,

    M. Yang, R. C. M. C. Shekar, O. Kang, and J. H. L. Hansen, “What can an accent identifier learn? probing phonetic and prosodic information in a wav2vec2-based accent identification model,” in Interspeech 2023, 2023, pp. 1923–1927

    2023

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.