Pith. sign in

REVIEW 5 major objections 5 minor 39 references

Dhvani: A Weakly-supervised Phonemic Error Detection and Personalized Feedback System for Hindi

T0 review · 5 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 Dhvani, a new Hindi pronunciation trainer, reports 82% F1 error detection and significant learner gains.

desk verdict Real engineering, but the 82% F1 is synthetic-only and doesn't support the benchmark claim. read the letter →

arxiv 2506.02166 v1 pith:W4C434VN submitted 2025-06-02 eess.AS cs.AIcs.LG

Arnav Rustagi , Satvik Bajpai , Nimrat Kaur , Siddharth Siddharth This is my paper
classification eess.AScs.AIcs.LG
keywords Hindipronunciationtrainingmispronunciationdetectioncomputer-assistedweaklysupervisedlearningspeechsynthesisphonemerecognitionpersonalizedfeedback
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

Dhvani is a computer-assisted pronunciation training system for Hindi, a language with about 600 million speakers but almost no dedicated pronunciation technology. The paper claims the system detects mispronounced phonemes with an F1 score of 82%, trained on real Hindi speech plus over three hours of synthetic mispronunciations generated by perturbing phoneme sequences and reading them with a text-to-speech model. When a mismatch is found, Dhvani returns personalized articulatory feedback dealing with tongue position, lip shape, airflow, and diagrams for the specific phoneme. In a study with 22 non-native learners, self-rated pronunciation improved significantly after using the feedback ($p = 0.0022$), with 10 of 13 difficult phonemes showing significant gains. If these results transfer to real learners, Dhvani would be the first scalable CAPT system for Hindi and a template for other Indian languages.

What carries the argument

The system is a weakly-supervised encoder-decoder: a recurrent-convolutional neural network (RCNN) encodes mel-spectrograms, and an attention-based recurrent neural network (ARNN) decodes phoneme sequences through a 67-unit classification head (64 Hindi phonemes plus end-of-word, end-of-sentence, and padding tokens). The paper's design decisions are to use separate decoders instead of a shared decoder across tasks, to replace location-sensitive attention with multi-head attention, and to preserve prosodic variation through energy and speed augmentation, which the authors found necessary because prosody distinguishes phonetically similar Hindi word pairs. The same decoded-phoneme sequence feeds the feedback generator, which maps each mispronounced phoneme to articulatory instructions and tongue diagrams.

What would settle it

Record a corpus of Hindi mispronunciations from genuine non-native learners, for instance speakers of Dravidian or other first languages, run Dhvani on those utterances, and compare its detection accuracy with the 82% F1 score reported on synthetic errors. If the F1 score on real learner errors is substantially lower, the system's reported performance does not transfer to its intended users.

Watch

Extended reading notes

Core claim

The paper's central claim is that Hindi's near one-to-one grapheme-to-phoneme mapping makes mispronunciation detection solvable by a single phoneme-recognition network: the system decodes an utterance into a phoneme sequence and compares it with the canonical sequence from the written Devanagari text, using classifier confidence to score error severity. To overcome the absence of real mispronounced Hindi speech, the authors synthesize training pairs by randomly adding, deleting, or modifying phonemes with probability 0.05 per phoneme and generate speech from the perturbed text with a text-to-speech system, keeping speaker embeddings matched between correct and error versions. On a held-out test set drawn from a public Hindi corpus plus synthetic errors, the model achieves an 82% F1 score under 5-fold cross-validation. The feedback study reports that learners rate the articulatory guidance highly and show statistically significant self-rated improvement.

Load-bearing premise

The load-bearing assumption is that mispronunciations synthesized by randomly perturbing phonemes and reading them aloud with a text-to-speech model are representative of the errors real non-native Hindi learners make; the system's 82% F1 score and its usefulness depend on that resemblance.

Editorial extensions

If this is right

  • Hindi learners could receive immediate, personalized pronunciation correction without a human teacher.
  • The same synthetic-error training recipe could be applied to other Indian languages with phonetic orthographies.
  • Weak supervision lowers the annotation burden: the system needs correct transcriptions plus synthetic error pairs, not expert-labeled mispronunciations.
  • If the self-reported gains reflect real pronunciation change, Dhvani-style feedback could be embedded in language-learning apps and digital tutors.
  • The reported 82% F1 score sets a benchmark against which future Hindi CAPT systems would be compared.

Reading between the lines

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

  • Editorial inference: the strongest test of the system is whether its F1 score survives on genuinely non-native errors; the synthetic data generation is the core uncertainty, not the architecture.
  • Editorial inference: because the human evaluation relied on self-ratings, the significant $p$-value demonstrates perceived improvement; an independent measure, such as expert listening or acoustic comparison, is needed to confirm actual pronunciation gains.
  • Editorial inference: the same phoneme-decode-and-compare approach may transfer more readily to other transparent orthographies than to English, whose many-to-many grapheme-phoneme mapping would require a dedicated error-detection head.
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

5 major / 5 minor

Summary. The paper presents Dhvani, a weakly-supervised CAPT system for Hindi built from an RCNN-ARNN phoneme recognizer adapted from WEAKLY-S, a synthetic data-generation pipeline that creates mispronounced Hindi speech with suno/bark-small, and a rule-based articulatory feedback module. The authors report an 82% F1-score for phonemic error detection on a 5-fold cross-validated dataset, and a human study with 22 non-native speakers in which self-rated pronunciation improved after using the system (Wilcoxon p = 0.0022). The central scientific claim is that Dhvani is the first effective pronunciation-training system for Hindi, substantially outperforming prior work.

Significance. If the detection claim were validated on real non-native Hindi speech and the feedback claim on objective pronunciation measures, this would be a noteworthy contribution to CAPT for Indic languages. The synthetic mispronunciation-generation pipeline and the detailed articulatory feedback design are practical assets, and the decision to release the code and model artifacts is commendable. However, the current evidence base consists of in-distribution synthetic evaluation and self-reported ratings; no real L2 Hindi corpus is used anywhere in the paper. The significance of the reported results therefore depends entirely on unvalidated assumptions about the representativeness of synthetic errors and of self-assessed pronunciation, so the contribution as presented is not yet established.

major comments (5)
  1. [Section 4.1 / Table 1] The reported 82% F1-score is computed on a test set generated by the same synthetic pipeline used for training: Section 3.2 describes 1,000 correct and 1,000 mispronounced sentence pairs created by uniformly perturbing phonemes (addition, deletion, modification) with probability 0.05 per phoneme and synthesizing them with bark-small using ten voices. This is an in-distribution evaluation on the exact distribution the model was trained on. It cannot support the claim in Table 1 of 'setting a new benchmark for Hindi pronunciation training' because real non-native errors are structured by L1 transfer (e.g., dental/retroflex substitutions, aspiration errors) rather than uniform random corruption. The manuscript provides no external L2 corpus, no expert-annotated non-native recordings, and no analysis showing that the synthetic error distribution matches real learner errors. This is a load-bearing validity issue; the detection claim rests entirely on it.
  2. [Section 4.2] The human evaluation uses only 5-point Likert self-ratings before and after using the feedback system. It does not measure actual acoustic or phonetic accuracy, and it does not involve the Dhvani detector at all, so it cannot validate the detection claim or demonstrate genuine improvement in pronunciation. The statistically significant p-value of 0.0022 shows that participants rated themselves higher after the intervention, not that their pronunciation improved. Without an objective pronunciation measure (e.g., expert annotations or acoustic features) or a control group, the conclusion in Section 4.2 of 'a meaningful improvement in pronunciation scores' is not supported by the data.
  3. [Section 4.1] The description of the 5-fold split is incomplete. The paper says speaker independence is preserved with 59 speakers in training and 19 in test, but it never specifies how the 10 synthetic TTS voices are partitioned across folds, nor whether the same synthetic utterances or voices appear in both training and test sets. If the same TTS voices or the same generated utterances are shared across folds, the F1-score can be inflated by voice memorization rather than genuine generalization. The fold construction over the synthetic portion of the dataset must be documented explicitly.
  4. [Section 2 and Section 4.1] The manuscript states that architectural changes—removing the shared decoder, replacing location-sensitive attention with multi-head attention, and adding prosodic augmentations—were 'validated through rigorous ablation studies,' but no ablation results are reported anywhere. The claim that these modifications are responsible for the reported performance is therefore unverifiable. The authors should include an ablation table with F1/PER for each architectural variant, or clearly state that such experiments were not run.
  5. [Table 1] The comparison in Table 1 is not apples-to-apples. The prior systems are evaluated on different tasks (phoneme recognition vs. mispronunciation detection), different languages (English, German, Polish, Kannada, Bangla), different datasets (TIMIT, L2-ARCTIC, Isle, etc.), and different metrics (accuracy, PER, F1). Directly concluding 'substantial improvement and setting a new benchmark' from this table is unsupported. The authors should restrict the comparison to systems measured on the same task and metric, or explicitly discuss the methodological caveats of cross-dataset comparisons.
minor comments (5)
  1. [Introduction] There is a typo in Section 1: 'enables the provision of providing comprehensive personalized feedback' should read 'enables the provision of comprehensive personalized feedback.'
  2. [Section 3.2] The error-injection process is described only by a per-phoneme probability of 0.05. The authors should report the average number of phonemes per sentence and the resulting expected number of errors per utterance, so that the difficulty of the synthetic task is interpretable.
  3. [Section 4.1] The statement that the standard deviation is 'less than 2% for all metrics' is vague. Please report the exact mean and standard deviation for each metric (F1, precision, recall, PER) and state whether F1 is computed at the phoneme level or the word level.
  4. [Section 3.4] The feedback mechanism is described as 'personalized,' but the personalization logic is not operationalized. Clarify how the feedback adapts to an individual learner's prior errors rather than simply providing the same static articulatory explanation for each mispronounced phoneme.
  5. [Section 3.4 / Reproducibility] The GitHub repository link is mentioned, but the paper should specify the license, the exact contents (model weights, data-generation scripts, evaluation scripts), and whether the synthetic dataset itself can be downloaded. This is important for reproducibility of the reported F1-score.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the reported F1 is an empirical same-distribution evaluation on held-out synthetic data; the synthetic-data representativeness concern is an external-validity risk, not a circular reduction.

full rationale

The paper's central detection claim is an empirical F1 score (Section 4.1, Table 1) obtained by training the RCNN-ARNN model on MUCS L1 Hindi plus synthetic mispronunciations and evaluating on a held-out fold of the same synthetic dataset (Section 3.2). This is standard same-distribution supervised evaluation: the test instances are not used in training, the labels are not derived from the model's own outputs, and the F1 is not analytically forced by construction. The synthesis pipeline (suno/bark-small, p=0.05 per-phoneme addition/deletion/modification) defines the target labels, and the decoder compares predicted phonemes to canonical phonemes, but that alignment is a dataset-design choice, not a circular equation. The concern that synthetic errors may not represent real non-native Hindi errors (e.g., L1-transfer patterns such as dental/retroflex substitution or loss of aspiration) is a substantive external-validity risk and should be weighed under correctness, not circularity. The human evaluation (Section 4.2) measures self-rated improvement before and after feedback; although subjective, it tests the feedback mechanism rather than the detector, and does not reduce to the training data. There are no load-bearing self-citations; the WEAKLY-S architecture is cited to external prior work, and the modifications (separate decoders, multi-head attention, prosodic augmentation) are presented as empirical choices rather than derived from the target result. Accordingly, no specific circular step can be exhibited under the definitions in the rubric; the derivation chain is empirical, not definitional.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The system introduces no new theoretical entities. The central claim depends on free parameters controlling synthetic data generation, the assumption that synthetic errors mirror real L2 errors, and the assumption that self-reported gains reflect actual pronunciation changes.

free parameters (3)
  • Error injection probability per phoneme = 0.05
    Used to systematically introduce phoneme addition, deletion, and modification errors in synthetic data (Section 3.2). This parameter determines the error distribution for both training and test sets.
  • Prosodic augmentation magnitudes = ±5 dB energy, ±10% speed
    Energy and speed modifications applied to preserve prosodic variations; the paper states performance drops without them, but no ablation is shown.
  • Number of synthetic speaker voices = 10
    Ten distinct bark-small voices used to generate the synthetic dataset (Section 3.2); the choice influences speaker diversity.
assumptions (5)
  • domain assumption Hindi has a highly consistent grapheme-to-phoneme mapping
    Used to justify phoneme-level analysis without a separate mispronunciation detection network (abstract, Section 4.1).
  • ad hoc to paper Synthetic mispronunciations are representative of real learner errors
    The entire evaluation relies on this assumption; stated implicitly in Section 3.2 where synthetic pairs are generated.
  • domain assumption Bark-small TTS produces realistic Hindi speech
    Used to generate synthetic correct and mispronounced audio (Section 3.2).
  • domain assumption MUCS telephone speech is adequate for phoneme-level error detection
    The L1 training data is from MUCS, an ASR corpus with 8 kHz telephone audio (Section 3.1).
  • ad hoc to paper Self-reported Likert ratings reflect actual pronunciation improvement
    The user study uses self-evaluation before and after intervention, assuming self-perception correlates with objective pronunciation quality (Section 4.2).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Dhvani: A Weakly-supervised Phonemic Error Detection and Personalized Feedback System for Hindi." pith.science (2026). https://pith.science/paper/W4C434VN

@misc{pith2026250602166,
  author       = {Pith},
  title        = {Pith review of: Dhvani: A Weakly-supervised Phonemic Error Detection and Personalized Feedback System for Hindi},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/W4C434VN}},
  note         = {Machine review of arXiv:2506.02166}
}
read the original abstract

Computer-Assisted Pronunciation Training (CAPT) has been extensively studied for English. However, there remains a critical gap in its application to Indian languages with a base of 1.5 billion speakers. Pronunciation tools tailored to Indian languages are strikingly lacking despite the fact that millions learn them every year. With over 600 million speakers and being the fourth most-spoken language worldwide, improving Hindi pronunciation is a vital first step toward addressing this gap. This paper proposes 1) Dhvani -- a novel CAPT system for Hindi, 2) synthetic speech generation for Hindi mispronunciations, and 3) a novel methodology for providing personalized feedback to learners. While the system often interacts with learners using Devanagari graphemes, its core analysis targets phonemic distinctions, leveraging Hindi's highly phonetic orthography to analyze mispronounced speech and provide targeted feedback.

Figures

Figures reproduced from arXiv: 2506.02166 by the authors.

Figure 1
Figure 1. Dhvani system overview: (a) Model architecture and (b) Personalized feedback mechanism Layer Hyperparameters Conv1D (x5) Out Channels: 16 Dropout: 0.25 Multi-head Attention Attention Heads: 8 Embedding Dimension: 64 Gated Recurrent Unit Hidden Size: 64 Dropout: 0.2 Training Batch Size: 32 Sequence Length: 256 tokens Optimization Optimizer: AdamW Learning Rate: 3e-6 Regularization Decoder dropout: 0.2 CNN dropout: 0.… view at source ↗
Figure 3
Figure 3. Comparison between pronunciation results before and after using the Dhvani’s feedback system for 13 challenging Hindi phonemes using a 5-point Likert scale (n=22 non-native speakers). Box plots show score distributions, with green indi￾cating statistically significant gains for 10 of 13 phonemes after using Dhvani’s feedback system. (p < 0.05, Wilcoxon Signed￾rank Test) Question Mean ± SD Prior awareness of pronunci… view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

39 extracted references · 37 canonical work pages

  1. [1]

    Introduction India’s linguistic diversity, while a cultural asset, presents sig- nificant challenges for spoken communication and linguistic co- hesion within the country. Indigenous languages remain widely in use and cannot be replaced by a single lingua franca, often necessitating individuals to learn multiple languages through- out their lives due to m...

    work page

  2. [2]

    Literature Survey CAPT has made significant strides in second language acqui- sition, yet existing models struggle with detecting pronunci- ation errors accurately. [18, 19] Currently, CAPT systems for Indian languages remain largely unexplored, mostly focusing on phoneme recognition tasks for detecting mispronunciation with limited success (Table 1). The...

    work page

  3. [3]

    a novel methodology for providing personalized feedback to learners. While the system often interacts with learners using Devanagari graphemes, its core analysis targets phonemic dis- tinctions, leveraging Hindi’s highly phonetic orthography to an- alyze mispronounced speech and provide targeted feedback. Index Terms: CAPT system, automatic pronunciation ...

    work page

  4. [4]

    Phoneme-to-Phoneme (P2P): Perturbs phonetic transcrip- tions to simulate errors but lacks variability in prosody and is unable to generate new speech signals. [19]

    work page

  5. [5]

    [20, 21]

    Text-to-Speech (T2S): Uses neural TTS to convert phoneme sequences into speech, offering improved variability, but pro- ducing the same output for identical inputs. [20, 21]

    work page

  6. [6]

    [5] The WEAKLY-S model, proposed by Korzekwa et al., used the robust S2S technique and bypassed the need for detailed phonetic transcription

    Speech-to-Speech (S2S): Introduces mispronunciations into aligned mel-spectrograms, maintaining prosody and voice timbre while generating diverse speech. [5] The WEAKLY-S model, proposed by Korzekwa et al., used the robust S2S technique and bypassed the need for detailed phonetic transcription. It could work on word-level mispronun- ciation detection with...

    work page

  7. [7]

    Dhvani: A Weakly-supervised Phonemic Error Detection and Personalized Feedback System for Hindi

    Mispronunciation Detection Network (MDN): This network detects pronunciation errors at the word-level by comparing the input speech signal with canonical phonemes. arXiv:2506.02166v1 [eess.AS] 2 Jun 2025 Research Work Methodology Language Dataset Metrics Feedback Bhatt et al. [8] HMM Hindi TIMIT Accuracy: 72.2% ✗ Dash et al. [9] SD-DNN Hindi Original Spee...

    work page Pith review arXiv 2025

  8. [8]

    Both networks share an RCNN encoder that processes mel- spectrogram representations and an attention-based RNN de- coder

    Phoneme Recognition Network (PRN): This network recog- nizes phonemes in the speech signal, enhancing the model’s robustness by processing both L1 and L2 data. Both networks share an RCNN encoder that processes mel- spectrogram representations and an attention-based RNN de- coder. This shared architecture serves as a regularizer, prevent- ing overfitting ...

    work page

Show all 39 references
  1. [9]

    Methodology 3.1. Dataset The MUCS dataset (for L1 Hindi Speech), a subset of the Multi- lingual and Code-switching ASR Challenge dataset curated by the MUCS organization and available on OpenSLR, was uti- lized for this research. [17] The dataset consists of telephone- quality...

  2. [10]

    Evaluation 4.1. Model Performance To ensure robust evaluation, we employed a rigorous 5-fold cross-validation methodology on our dataset, which comprises 95.05 hours of training data and 5.55 hours of test data. The dataset was carefully partitioned to maintain speaker indepen...

  3. [11]

    By carefully adapting to Hindi’s distinctive pho- netic features, our system significantly improves upon exist- ing CAPT technologies

    Conclusion and Future Work Our research presents a promising CAPT system for Hindi, ful- filling a critical gap in language learning technology for Indian languages. By carefully adapting to Hindi’s distinctive pho- netic features, our system significantly improves upon exist-...

  4. [12]

    Acknowledgments The authors would like to thank Harish and Bina Shah School of AI & Computer Science at Plaksha University for providing seed financial support for this research

  5. [13]

    Census of India,

    Office of the Registrar General & Census Commissioner, India, “Census of India,” 2011, data available at: https://censusindia.gov. in/census.website/data/census-tables

    2011

  6. [14]

    National Education Policy 2020,

    Ministry of Human Resource Development, Government of In- dia, “National Education Policy 2020,” 2020, policy avail- able at: https://www.education.gov.in/sites/upload files/mhrd/ files/NEP Final English 0.pdf

    2020

  7. [15]

    Importance of correct pronunci- ation in spoken english: Dimension of second language learners’ perspective,

    Z. A. Bakar and M. Abdullah, “Importance of correct pronunci- ation in spoken english: Dimension of second language learners’ perspective,” Pertanika Journal of Social Sciences and Humani- ties, vol. 23, no. 8, pp. 143–158, 2015

    2015

  8. [16]

    Weakly-supervised word-level pronunciation error detection in non-native english speech,

    D. Korzekwa, J. Lorenzo-Trueba, T. Drugman, S. Calamaro, and B. Kostek, “Weakly-supervised word-level pronunciation error detection in non-native english speech,” arXiv preprint arXiv:2106.03494, 2021

    2021 arXiv

  9. [17]

    Computer-assisted pronunciation training—speech synthesis is almost all you need,

    D. Korzekwa, J. Lorenzo-Trueba, T. Drugman, and B. Kostek, “Computer-assisted pronunciation training—speech synthesis is almost all you need,” Speech Communication, vol. 142, pp. 22– 33, 2022

    2022

  10. [18]

    Structural analysis of hindi phonetics and a method for extraction of phonetically rich sentences from a very large hindi text corpus,

    S. Malviya, R. Mishra, and U. S. Tiwary, “Structural analysis of hindi phonetics and a method for extraction of phonetically rich sentences from a very large hindi text corpus,” in2016 Conference of The Oriental Chapter of International Committee for Coordi- nation and Standar...

    2016

  11. [19]

    Detection of phonemic aspiration for spoken hindi pronunciation evaluation,

    V . V . Patil and P. Rao, “Detection of phonemic aspiration for spoken hindi pronunciation evaluation,” Journal of Phonetics , vol. 54, pp. 202–221, 2016. [Online]. Available: https://www. sciencedirect.com/science/article/pii/S0095447015000947

    2016

  12. [20]

    Confusion analysis in phoneme based speech recognition in hindi,

    S. Bhatt, A. Dev, and A. Jain, “Confusion analysis in phoneme based speech recognition in hindi,” Journal of Ambient Intelli- gence and Humanized Computing, vol. 11, no. 10, pp. 4213–4238, 2020

    2020

  13. [21]

    Automatic speech recognition with articulatory information and a unified dic- tionary for hindi, marathi, bengali and oriya

    D. Dash, M. J. Kim, K. Teplansky, and J. Wang, “Automatic speech recognition with articulatory information and a unified dic- tionary for hindi, marathi, bengali and oriya.” in INTERSPEECH, 2018, pp. 1046–1050

    2018

  14. [22]

    Deep neural networks for kannada phoneme recognition,

    R. Pradeep and K. S. Rao, “Deep neural networks for kannada phoneme recognition,” in2016 Ninth International Conference on Contemporary Computing (IC3). IEEE, 2016, pp. 1–6

    2016

  15. [23]

    Development of kannada speech corpus for prosodically guided phonetic search engine,

    M. Shridhara, B. K. Banahatti, L. Narthan, V . Karjigi, and R. Kumaraswamy, “Development of kannada speech corpus for prosodically guided phonetic search engine,” in 2013 interna- tional conference oriental COCOSDA held jointly with 2013 con- ference on Asian spoken language r...

    2013

  16. [24]

    Bangla phoneme recognition using hybrid features,

    M. R. A. Kotwal, M. S. Hossain, F. Hassan, G. Muhammad, M. N. Huda, and C. M. Rahman, “Bangla phoneme recognition using hybrid features,” in International Conference on Electrical & Computer Engineering (ICECE 2010). IEEE, 2010, pp. 718– 721

    2010

  17. [25]

    Read—a bangla phoneme recognition system,

    H. Mukherjee, C. Halder, S. Phadikar, and K. Roy, “Read—a bangla phoneme recognition system,” in Proceedings of the 5th International Conference on Frontiers in Intelligent Computing: Theory and Applications: FICTA 2016, Volume 1 . Springer, 2017, pp. 599–607

    2016

  18. [26]

    End-to-end automatic pronunciation error detection based on improved hybrid ctc/attention architecture,

    L. Zhang, Z. Zhao, C. Ma, L. Shan, H. Sun, L. Jiang, S. Deng, and C. Gao, “End-to-end automatic pronunciation error detection based on improved hybrid ctc/attention architecture,” Sensors, vol. 20, no. 7, 2020. [Online]. Available: https: //www.mdpi.com/1424-8220/20/7/1809

    2020

  19. [27]

    Improved mispronunciation detection system using a hybrid CTC-ATT based approach for L2 english speakers,

    N. Baranwal and S. Chilaka, “Improved mispronunciation detection system using a hybrid CTC-ATT based approach for L2 english speakers,” CoRR, vol. abs/2201.10198, 2022. [Online]. Available: https://arxiv.org/abs/2201.10198

    2022 arXiv

  20. [28]

    L2-arctic: A non- native english speech corpus,

    G. Zhao, E. Chukharev-Hudilainen, S. Sonsaat, A. Silpachai, I. Lucic, R. Gutierrez-Osuna, and J. Levis, “L2-arctic: A non- native english speech corpus,” 2018

    2018

  21. [29]

    Multi- lingual and code-switching asr challenges for low resource indian languages,

    A. Diwan, R. Vaideeswaran, S. Shah, A. Singh, S. Raghavan, S. Khare, V . Unni, S. Vyas, A. Rajpuria, C. Yarra et al., “Multi- lingual and code-switching asr challenges for low resource indian languages,” arXiv preprint arXiv:2104.00235, 2021

    2021 arXiv

  22. [30]

    Cnn-rnn-ctc based end-to- end mispronunciation detection and diagnosis,

    W.-K. Leung, X. Liu, and H. Meng, “Cnn-rnn-ctc based end-to- end mispronunciation detection and diagnosis,” in ICASSP 2019- 2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2019, pp. 8132–8136

    2019

  23. [31]

    L2-gen: A neural phoneme paraphrasing approach to l2 speech synthesis for mispronunciation diagnosis,

    D. Zhang, A. Ganesan, S. Campbell, and D. Korzekwa, “L2-gen: A neural phoneme paraphrasing approach to l2 speech synthesis for mispronunciation diagnosis,” 2022

    2022

  24. [32]

    Controllable accented text- to-speech synthesis with fine and coarse-grained intensity render- ing,

    R. Liu, B. Sisman, G. Gao, and H. Li, “Controllable accented text- to-speech synthesis with fine and coarse-grained intensity render- ing,” IEEE/ACM Transactions on Audio, Speech, and Language Processing, 2024

    2024

  25. [33]

    Non-autoregressive tts with explicit duration modelling for low-resource highly expressive speech,

    R. Shah, K. Pokora, A. Ezzerg, V . Klimkov, G. Huybrechts, B. Pu- trycz, D. Korzekwa, and T. Merritt, “Non-autoregressive tts with explicit duration modelling for low-resource highly expressive speech,” arXiv preprint arXiv:2106.12896, 2021

    2021 arXiv

  26. [34]

    Mispronunciation detection in non- native (l2) english with uncertainty modeling,

    D. Korzekwa, J. Lorenzo-Trueba, S. Zaporowski, S. Calamaro, T. Drugman, and B. Kostek, “Mispronunciation detection in non- native (l2) english with uncertainty modeling,” in ICASSP 2021- 2021 IEEE international conference on acoustics, speech and sig- nal processing (ICASSP). ...

    2021

  27. [35]

    Computer-assisted pronunciation training—speech synthesis is almost all you need,

    D. Korzekwa, J. Lorenzo-Trueba, T. Drugman, and B. Kostek, “Computer-assisted pronunciation training—speech synthesis is almost all you need,” Speech Communication , vol. 142, p. 22–33, Jul. 2022. [Online]. Available: http://dx.doi.org/10.1016/ j.specom.2022.06.003

    2022

  28. [36]

    Bark: A transformer-based text-to-audio model,

    Suno, “Bark: A transformer-based text-to-audio model,” 2024, model available at: https://github.com/suno-ai/bark

    2024

  29. [37]

    Delhi: NCERT, 2015

    NCERT, Antra Part-I: Hindi textbook for class XI . Delhi: NCERT, 2015

    2015

  30. [38]

    New Delhi: NCERT, 2006, hindi text- book for class XI

    NCERT, Antara Bhag-1. New Delhi: NCERT, 2006, hindi text- book for class XI

    2006

  31. [39]

    Delhi: NCERT, 2018

    NCERT, Aaroh Part-II: Hindi textbook for class XII . Delhi: NCERT, 2018

    2018

Pith tools

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