Pith. sign in

REVIEW 5 major objections 6 minor 26 references

AdaCS: Adaptive Normalization for Enhanced Code-Switching ASR

T0 review · 5 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read Adaptive bias attention cuts code-switching speech errors by 56 percent

desk verdict Credible incremental extension of AdapITN with a reusable Vietnamese CS benchmark, but the headline WER gains are measured on synthetic text pairs and will need real-ASR validation before they mean much in production. read the letter →

arxiv 2501.07102 v1 pith:SMPEOKOJ submitted 2025-01-13 cs.CL cs.AIcs.SDeess.AS

The Chuong Chu , Vu Tuan Dat Pham , Kien Dao , Hoang Nguyen , Quoc Hung Truong This is my paper · ORCID
classification cs.CLcs.AIcs.SDeess.AS
keywords code-switchingspeechrecognitionadaptivenormalizationcontextualbiasingbiasattentionmoduleVietnameseASRinversetextlow-resourcelanguageencoder-decoder
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

AdaCS is a post-ASR normalization model for Vietnamese code-switched text, where English names and technical terms appear inside Vietnamese utterances. The paper's central claim is that inserting the same bias-attention module into both the phrase-tagging encoder and the normalizing decoder lets a model use a supplied list of expected words or phrases to locate and rewrite code-switched spans, and that this is enough to beat the previous state-of-the-art normalization system. On the paper's two new test sets, AdaCS reports relative word-error-rate reductions of 56.2% on the general set and 36.8% on the medical set over AdapITN's best results when a bias list is available. This matters because it offers a cheap, plug-in route to fixing code-switching errors in low-resource languages without retraining the acoustic model or relying on a large language model.

What carries the argument

The load-bearing object is the Bias Attention Module (BAM), which injects the bias list into both stages. For a token's hidden representation, BAM scores it against pooled vectors of every bias-list entry by inner product, selects the highest-scoring entry by argmax, runs multi-head attention with the token as query and the chosen entry's token embeddings as keys and values, and adds the attention output back to the token. Applied after the encoder, this steers the tagger's start/inside/outside classification; applied in the decoder, it steers generation of the normalized phrase. Training combines four cross-entropy losses—tagger, encoder bias-ranking, decoder bias-ranking, and next-token generation—so the model is explicitly trained to choose the right bias entry at both stages.

What would settle it

Run a production Vietnamese ASR system on real code-switched speech, feed the raw transcripts to AdaCS and AdapITN with the same bias lists, and compare word error rates; if AdaCS's advantage over AdapITN shrinks or reverses on that real output, the reported 56.2% and 36.8% reductions are artifacts of the synthetic test format.

Watch

Extended reading notes

Core claim

The paper claims that adaptive normalization, rather than model scale, is what resolves intra-sentential code-switching. AdaCS inserts the Bias Attention Module (BAM) twice: after the encoder, so the tagger marks tokens that belong to a code-switched phrase using the bias list, and inside the decoder, so each generated normalized token is conditioned on the matching bias entry. The result is dynamic tagging: given the same input sentence, AdaCS changes its start/inside/outside tags when the bias list switches from words to phrases, while AdapITN keeps the same tags; the paper presents this as the reason AdaCS normalizes more accurately. With no bias list, GPT-4o has the lowest word error rate (15.4% general, 15.0% medical), but with biases AdaCS reaches 2.8% WER on test-general using phrase bias and 4.8% on test-medical using word bias, against AdapITN's best 6.4% and 7.6%. The paper also reports that phrase bias helps on the general set but slightly hurts on the medical set, where distinct phrase entries share overlapping words.

Load-bearing premise

The evaluation assumes the paper's hand-constructed 'spoken' sentences—English phrases replaced with manually labeled Vietnamese pronunciations—faithfully represent what a real ASR system would output, because no actual acoustic-model transcript is ever passed through AdaCS.

Editorial extensions

If this is right

  • With a bias list, AdaCS beats both the previous state-of-the-art normalization system and GPT-4o on the two proposed test sets, showing that contextual biasing can substitute for model scale in this task.
  • Dynamic tagging is the mechanism: AdaCS changes its start/inside/outside tags when the bias list changes from words to phrases, so performance remains high when the list format changes.
  • Scaling experiments up to 2,000 bias entries show that CS-word error remains far below the no-bias baseline, and AdaCS stays below AdapITN at every list size on both test sets.
  • Because AdaCS and AdapITN share the same decoder design and training base, the reported gains isolate the contribution of placing the bias module in both the tagger and the decoder.

Reading between the lines

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

  • Implicit but untested: if real ASR hypotheses contain partial words, missing diacritics, insertions, or deletions rather than clean Vietnamese pronunciations, the reported WER gains may not transfer; a test on raw acoustic-model output would settle this.
  • The same BAM recipe should transfer to other low-resource code-switching pairs, such as Mandarin-English or Hindi-English, as long as a pretrained bilingual text encoder exists, since nothing in the mechanism depends on Vietnamese.
  • The medical-domain result, where phrase bias hurts relative to word bias because distinct phrases share overlapping tokens, pinpoints argmax rank-and-select as the fragile step; deduplicating or shortening overlapping list entries could be a cheap fix worth testing.
  • The equal-weight four-term loss invites an ablation: removing each ranking loss separately would reveal whether encoder biasing, decoder biasing, or their combination carries the improvement over AdapITN.
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.

Referee Report

5 major / 6 minor

Summary. The paper proposes AdaCS, a sequence-to-sequence normalization model that augments both the tagger and the decoder with a bias attention module (BAM). BAM selects a bias entry via inner-product similarity and attends to it, allowing the model to adapt to a provided bias list. The authors construct a synthetic Vietnamese code-switching dataset by manually replacing English phrases with Vietnamese pronunciations, and compare AdaCS against Transformer, GPT-4o, and AdapITN on general and medical test sets. On the synthetic protocol with bias lists, AdaCS achieves 2.8% and 4.8% WER on the two test sets, outperforming AdapITN's best results, leading to the abstract's relative reductions of 56.2% and 36.8%.

Significance. If the reported results are externally valid, the idea of applying the same bias-attention mechanism to both phrase tagging and decoding is a clean and useful contribution for low-resource code-switching normalization. The paper also contributes a 50k-pair dataset and makes code and checkpoints available, and the controlled comparison to AdapITN is fair under the same protocol. However, external validity is not established because the evaluation uses hand-constructed text pairs rather than ASR hypotheses, and the inference bias lists are oracle-like (drawn from the test set's own English words). The relative gains are therefore best interpreted as a controlled comparison under a favorable protocol, with the generalization claims not yet supported.

major comments (5)
  1. [Section IV-A] The evaluation protocol uses synthetic text pairs rather than actual ASR hypotheses. The Input sentences are created by manually replacing English CS phrases with Vietnamese pronunciations (Table I), so they are perfectly segmented and diacritic-free; real ASR output for Vietnamese code-switched speech will contain partial transcriptions, missing or incorrect diacritics, insertions, deletions, and boundary errors. Since BAM's retrieval in Eq. (2) is based on token-level similarity to bias entries, distorted tokens are likely to yield incorrect bias_index values. Please test on real ASR hypotheses or inject realistic ASR noise, and report how the WER reductions change.
  2. [Section IV-B] The bias lists are oracle-like: Experiment 2 uses 'a random bias list ... drawn from the list of English words in the entire corresponding test set, combining English words from current sentences', and the phrase-bias experiment likewise includes the target phrases. This means the model is told the set of possible outputs for each test sentence. The claim of 'the ability to handle unseen CS phrases' (Section IV-D) is therefore only demonstrated for phrases explicitly provided in the bias list. To support the generalization claim, add experiments where the bias list excludes the target phrases, contains unrelated distractors, or includes near-miss spellings, and report the effect on WER.
  3. [Section III-D / IV-C] Training bias lists are generated as 'the English words present in the sentences within the batch, as well as random English words', i.e., word-level entries only, but the test experiments include phrase-level bias lists (Table I). The ranking losses Lenc_rank and Ldec_rank use labels that are indices of bias entries; the paper does not describe how phrase entries are encoded or labeled during training. Without phrase-level training, it is unexplained why phrase-level bias helps on test-general (2.8% WER) but hurts on test-medical (7.8% vs 4.8% for word bias). Please clarify the phrase representation in BAM and whether phrase-level training examples are used, and provide an ablation for word vs phrase biases.
  4. [Abstract / Table II] The headline reductions of 56.2% and 36.8% compare AdaCS's best bias type per test set (phrases on test-general, words on test-medical) with AdapITN's best bias type (words on both sets). Under a fixed bias type, the relative improvements are 48.4% (test-general, words: 3.3 vs 6.4), 61.6% (test-general, phrases: 2.8 vs 7.3), 36.8% (test-medical, words: 4.8 vs 7.6), and 12.4% (test-medical, phrases: 7.8 vs 8.9). The abstract should report per-condition comparisons or use a single consistent bias type, since the current phrasing is misleading.
  5. [Section IV] No error bars, multiple-seed runs, or ablations are reported. Table II and Figure 3 present single-run results on test sets of only 2,000 sentences each, so the significance of the WER differences is unknown. An ablation removing BAM from either the tagger or decoder would isolate the contribution of each module, and confidence intervals would support the robustness claims. Please provide these.
minor comments (6)
  1. [Section III-A, Eq. (1)] The notation 'score = sP^T' is ambiguous because P is described as a matrix of shape (L+1)*dmodel; clarify the intended row/column orientation and include the dummy entry dimension.
  2. [Section IV-D] The sentence 'relative WER reductions of 46.9% to 80.9%' is inaccurate; the maximum relative reduction from the Transformer baseline in Table II is 90.2% (AdaCS, phrases, test-general). Please recalculate or specify the comparison set.
  3. [Table II] The metrics N-WER and CS-WER are not formally defined; state how tagged regions are mapped to word error computations.
  4. [Section IV-C] The base pretrained model 'EnViBERT' is cited as [27], [28], but neither reference describes this model; provide a proper citation or description.
  5. [Section II] The reference for XLS-R is incorrect; [14] is XLA VS-R, not XLS-R. Please cite the actual XLS-R work.
  6. [Figure 1] Figure 1 has small labels and multiple overlapping annotation styles; please enlarge or split the figure for legibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the reported WER gains are measured under a symmetric bias-list protocol on the authors' own benchmark, and no prediction reduces to its fitted inputs by construction.

full rationale

The paper's central claim is an empirical comparison of AdaCS against AdapITN under the same bias-list conditions (Section IV-B, Table II). The bias lists are drawn from the English words in the corresponding test set, so the model is given the candidate answer inventory; however, this is a contextual-biasing evaluation protocol rather than a derivation step. The model must still rank the correct bias entry with Eq. (2) and generate the normalized phrase via attention in Eq. (3), so the WER number is not equal to the bias list by construction. Training and test inputs are hand-constructed 'spoken-reference' pairs (Section IV-A), which raises external validity concerns for real ASR hypotheses, but the comparison to AdapITN is symmetric and the method is not fitted to the test outputs. I found no load-bearing self-citation: the cited prior works are baselines or backbones ([9], [25], [27], [28]) and are not invoked as an external uniqueness theorem or an ansatz that smuggles in the target result. The ranking losses L_enc_rank and L_dec_rank supervise the model to select bias entries, but the reported reductions still require correct retrieval and generation, so no prediction reduces to a fitted parameter by construction. Accordingly, the circularity score is 0; the benchmark-validity concerns belong under correctness risk, not circularity.

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

The central claim rests on a synthetic data generation protocol and an oracle-like bias list protocol. The model contributes a new attention configuration rather than new physical entities, so the invented-entity list is empty. Three hand-chosen settings and four domain assumptions are load-bearing; none are independently verified in the paper.

free parameters (3)
  • Loss weights alpha, beta, gamma, delta = 1.0 each
    Chosen by hand (Section III-D) without reported tuning; all four loss terms weighted equally.
  • Bias list size for main experiments = 1000 entries
    A predetermined size of 1000 CS words/phrases is used in Experiments 2 and 3 (Section IV-B); varied only in the scaling experiment.
  • Test-set unseen-phrase threshold = at least 90%
    Test sets are curated so at least 90% of CS phrases are absent from training (Section IV-A); this threshold is a design choice that shapes difficulty.
assumptions (4)
  • domain assumption EnViBERT pretrained representations are suitable for Vietnamese-English token sequences
    AdaCS, AdapITN, and the baseline are initialized from EnViBERT (Section IV-C); if the base model's tokenization or representations are mismatched, the reported results would change.
  • domain assumption Manually written Vietnamese transliterations represent ASR output
    Training and test inputs are generated by replacing CS phrases with manually labeled Vietnamese pronunciation (Section IV-A), not by running an ASR system; real ASR errors such as deletions, insertions, and diacritic loss are absent.
  • ad hoc to paper Bias list drawn from test set words is a realistic inference-time context list
    In Experiments 2 and 3, the bias list is drawn from the English words in the entire corresponding test set (Section IV-B), giving the model the correct target phrases; this is a protocol choice specific to this paper.
  • domain assumption Dummy bias entry B0 adequately represents 'no bias' cases
    The BAM adds a dummy entry B0 to handle tokens without bias information (Section III-A); the model's behavior for out-of-list tokens depends on this design.

how reviews work

0 comments
Cite this review

Pith. "Pith review of AdaCS: Adaptive Normalization for Enhanced Code-Switching ASR." pith.science (2026). https://pith.science/paper/SMPEOKOJ

@misc{pith2026250107102,
  author       = {Pith},
  title        = {Pith review of: AdaCS: Adaptive Normalization for Enhanced Code-Switching ASR},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SMPEOKOJ}},
  note         = {Machine review of arXiv:2501.07102}
}
read the original abstract

Intra-sentential code-switching (CS) refers to the alternation between languages that happens within a single utterance and is a significant challenge for Automatic Speech Recognition (ASR) systems. For example, when a Vietnamese speaker uses foreign proper names or specialized terms within their speech. ASR systems often struggle to accurately transcribe intra-sentential CS due to their training on monolingual data and the unpredictable nature of CS. This issue is even more pronounced for low-resource languages, where limited data availability hinders the development of robust models. In this study, we propose AdaCS, a normalization model integrates an adaptive bias attention module (BAM) into encoder-decoder network. This novel approach provides a robust solution to CS ASR in unseen domains, thereby significantly enhancing our contribution to the field. By utilizing BAM to both identify and normalize CS phrases, AdaCS enhances its adaptive capabilities with a biased list of words provided during inference. Our method demonstrates impressive performance and the ability to handle unseen CS phrases across various domains. Experiments show that AdaCS outperforms previous state-of-the-art method on Vietnamese CS ASR normalization by considerable WER reduction of 56.2% and 36.8% on the two proposed test sets.

Figures

Figures reproduced from arXiv: 2501.07102 by the authors.

Figure 1
Figure 1. An overview of the AdaCS architecture, along with an illustrative example. The Bias Attention Module (BAM) is on the right side of the figure, and [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. An example of the impact of the word bias list and phrase bias list on [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Performance of AdaCS and AdapITN as the size of the bias list [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

26 extracted references · 15 canonical work pages

  1. [1]

    Contextual speech recognition in end-to-end neural network sys- tems using beam search,

    I. Williams, A. Kannan, P . Aleksic, D. Rybach, and T. N. Sainath, “Contextual speech recognition in end-to-end neural network sys- tems using beam search,” in Conference of the International Speech Communication Association, ISCA, Sep. 2018, pp. 2227–2231. doi: 10.21437/INTERSPEECH.2018-2416

    work page doi:10.21437/interspeech.2018-2416 2018

  2. [2]

    Deep Context: End-to-end Contextual Speech Recognition,

    G. Pundak, T. N. Sainath, R. Prabhavalkar, A. Kannan, and D. Zhao, “Deep Context: End-to-end Contextual Speech Recognition,” in Spoken Language Technology Workshop, IEEE, Dec. 2018, pp. 418–425. doi: 10.1109/SLT.2018.8639034

    work page arXiv 2018

  3. [3]

    Shallow-Fusion End-to-End Contextual Biasing,

    D. Zhao et al., “Shallow-Fusion End-to-End Contextual Biasing,” in Conference of the International Speech Communication Association, ISCA, Sep. 2019, pp. 1418–1422. doi: 10.21437/INTERSPEECH.2019-1209

    work page doi:10.21437/interspeech.2019-1209 2019

  4. [4]

    Deep Shallow Fusion for RNN-T Personalization,

    D. Le, G. Keren, J. Chan, J. Mahadeokar, C. Fuegen, and M. L. Seltzer, “Deep Shallow Fusion for RNN-T Personalization,” in Spoken Language Technology Workshop, IEEE, Jan. 2021, pp. 251–257. doi: 10.1109/SLT48900.2021.9383560

    arXiv 2021

  5. [6]

    Neural Inverse Text Normalization,

    M. Sunkara, C. Shivade, S. Bodapati and K. Kirchhoff, "Neural Inverse Text Normalization," ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Toronto, ON, Canada, 2021, pp. 7573-7577

    work page 2021

  6. [7]

    Correction of Automatic Speech Recognition with Transformer Sequence-To-Sequence Model,

    O. Hrinchuk, M. Popova and B. Ginsburg, "Correction of Automatic Speech Recognition with Transformer Sequence-To-Sequence Model," ICASSP 2020 - 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Barcelona, Spain, 2020, pp. 7074-7078

    work page 2020

  7. [8]

    Instant One-Shot Word-Learning for Context-Specific Neural Sequence-to-Sequence Speech Recognition,

    C. Huber, J. Hussain, S. St ¨uker and A. Waibel, "Instant One-Shot Word-Learning for Context-Specific Neural Sequence-to-Sequence Speech Recognition," 2021 IEEE Automatic Speech Recognition and Understand- ing Workshop (ASRU), Cartagena, Colombia, 2021, pp. 1-7

    work page 2021

  8. [10]

    Four-in-One: a Joint Approach to Inverse Text Normalization, Punctu- ation, Capitalization, and Disfluency for Automatic Speech Recognition,

    “Four-in-One: a Joint Approach to Inverse Text Normalization, Punctu- ation, Capitalization, and Disfluency for Automatic Speech Recognition,” in 2022 IEEE Spoken Language Technology Workshop (SLT), 2022 IEEE Spoken Language Technology Workshop (SLT), Jan. 2023. doi: 10.1109/slt54892.2023.10023257

    arXiv 2022

Show all 26 references
  1. [11]

    Towards Contextual Spelling Correction for Customization of End-to- End Speech Recognition Systems,

    X. Wang, Y . Liu, J. Li, V . Miljanic, S. Zhao, and H. A. Khalil, “Towards Contextual Spelling Correction for Customization of End-to- End Speech Recognition Systems,” IEEE/ACM transactions on audio, speech, and language processing, vol. 30, pp. 3089–3097, Mar. 2022, doi: 10.1...

    2022

  2. [12]

    Improving Contextual Spelling Correction by External Acoustics Attention and Semantic Aware Data Augmentation,

    X. Wang, Y . Liu, J. Li and S. Zhao, "Improving Contextual Spelling Correction by External Acoustics Attention and Semantic Aware Data Augmentation," ICASSP 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Rhodes Island, Greece, 20...

    2023

  3. [13]

    Robust Speech Recognition via Large-Scale Weak Supervision,

    A. Radford, J. W . Kim, T. Xu, G. Brockman, C. McLeavey, and I. Sutskever, “Robust Speech Recognition via Large-Scale Weak Supervision,” arXiv.org, vol. abs/2212.04356, Dec. 2022, doi: 10.48550/arXiv.2212.04356

    2022 doi arXiv

  4. [14]

    XLA VS-R: Cross-Lingual Audio-Visual Speech Repre- sentation Learning for Noise-Robust Speech Perception,

    H. Han et al., “XLA VS-R: Cross-Lingual Audio-Visual Speech Repre- sentation Learning for Noise-Robust Speech Perception,” arXiv.org, vol. abs/2403.14402, Mar. 2024, doi: 10.48550/arxiv.2403.14402

    2024 doi arXiv

  5. [15]

    Google USM: Scaling Automatic Speech Recognition Beyond 100 Languages,

    Y . Zhang et al., “Google USM: Scaling Automatic Speech Recognition Beyond 100 Languages,” arXiv.org, vol. abs/2303.01037, Mar. 2023, doi: 10.48550/arXiv.2303.01037

    2023 doi arXiv

  6. [16]

    Reducing Language Confusion for Code-Switching Speech Recognition with Token-Level Language Diarization,

    H. Liu, H. Xu, L. P . Garcia, A. W . H. Khong, Y . He and S. Khudanpur, "Reducing Language Confusion for Code-Switching Speech Recognition with Token-Level Language Diarization," ICASSP 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASS...

    2023

  7. [17]

    Towards Context- Aware End-to-End Code-Switching Speech Recognition,

    Z. Qiu, Y . Li, X. Li, F. Metze, and W . M. Campbell, “Towards Context- Aware End-to-End Code-Switching Speech Recognition,” in Conference of the International Speech Communication Association, ISCA, Oct. 2020, pp. 4776–4780. doi: 10.21437/INTERSPEECH.2020-1980

    2020 doi

  8. [18]

    Bayesian Transformer Using Disentangled Mask Attention,

    J.-T. Chien and Y .-H. Huang, “Bayesian Transformer Using Disentangled Mask Attention,” in Interspeech 2022, Sep. 2022, pp. 1761–1765. doi: 10.21437/interspeech.2022-10457

    2022 doi

  9. [19]

    En- hancing Code-Switching Speech Recognition With Interactive Language Biases,

    H. Liu, L. P . Garcia, X. Zhang, A. W . H. Khong, and S. Khudanpur, “En- hancing Code-Switching Speech Recognition With Interactive Language Biases,” Apr. 2024, doi: 10.1109/icassp48485.2024.10448335

    2024

  10. [20]

    A first speech recognition system for Mandarin-English code-switch conversational speech,

    N. T. Vu et al., "A first speech recognition system for Mandarin-English code-switch conversational speech," 2012 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Kyoto, Japan, 2012, pp. 4889-4892, doi: 10.1109/ICASSP .2012.6289015

    2012

  11. [21]

    On the End-to-End Solution to Mandarin-English Code- Switching Speech Recognition,

    Z. Zeng, Y . Khassanov, V . T. Pham, H. Xu, E. S. Chng, and H. Li, “On the End-to-End Solution to Mandarin-English Code- Switching Speech Recognition,” in Conference of the International Speech Communication Association, ISCA, Sep. 2019, pp. 2165–2169. doi: 10.21437/INTERSPEEC...

    2019 doi

  12. [22]

    Contextual RNN-T for Open Domain ASR,

    M. Jain, G. Keren, J. Mahadeokar, G. Zweig, F. Metze, and Y . Saraf, “Contextual RNN-T for Open Domain ASR,” in Conference of the Inter- national Speech Communication Association, ISCA, Oct. 2020, pp. 11–15. doi: 10.21437/INTERSPEECH.2020-2986

    2020 doi

  13. [23]

    Phoebe: Pronunciation-aware Contextualization for End-to-end Speech Recognition,

    A. Bruguier, R. Prabhavalkar, G. Pundak, and T. N. Sainath, “Phoebe: Pronunciation-aware Contextualization for End-to-end Speech Recognition,” in International Conference on Acoustics, Speech, and Signal Processing, IEEE, May 2019, pp. 6171–6175. doi: 10.1109/ICASSP .2019.8682441

    2019

  14. [24]

    Towards Code-switching ASR for End-to-end CTC Models,

    K. Li, J. Li, G. Y e, R. Zhao and Y . Gong, "Towards Code-switching ASR for End-to-end CTC Models," ICASSP 2019 - 2019 IEEE Interna- tional Conference on Acoustics, Speech and Signal Processing (ICASSP), Brighton, UK, 2019, pp. 6076-6080

    2019

  15. [25]

    Attention is All you Need,

    A. Vaswani et al., “Attention is All you Need,” in Neural Information Processing Systems, Curran Associates Inc., Jun. 2017, pp. 5998–6008. [Online]. Available: https://proceedings.neurips.cc/paper/2017/file/ 3f5ee243547dee91fbd053c1c4a845aa-Paper.pdf

    2017

  16. [26]

    GPT-4 Technical Report,

    OpenAI, “GPT-4 Technical Report,” arXiv.org, vol. abs/2303.08774, Mar. 2023, doi: 10.48550/arXiv.2303.08774

    2023 doi arXiv

  17. [27]

    BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding,

    J. Devlin, M.-W . Chang, K. Lee, and K. Toutanova, “BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding,” in North American Chapter of the Association for Computational Linguistics, Association for Computational Linguistics, Oct. 2018, pp. 4171–418...

    2018 doi

  18. [28]

    Improving Vietnamese Named Entity Recognition from Speech Using Word Capitalization and Punctuation Recovery Models,

    T. B. Nguyen, Q. M. Nguyen, T. T. H. Nguyen, Q. T. Do, and C. M. Luong, “Improving Vietnamese Named Entity Recognition from Speech Using Word Capitalization and Punctuation Recovery Models,” in Conference of the International Speech Communication Association, ISCA, Oct. 2020, ...

    2020 doi

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.