Pith. sign in

REVIEW 3 major objections 7 minor 38 references

Enhancing Code-Switching ASR Leveraging Non-Peaky CTC Loss and Deep Language Posterior Injection

T0 review · 3 major / 7 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Flattening CTC's peaky frame posteriors and feeding the resulting language boundaries into the encoder cuts mixed error rate on SEAME below the D-MoE baseline.

desk verdict Consistent MER gains over D-MoE on SEAME, but the missing D-MoE ablations leave the component attribution unsupported. read the letter →

arxiv 2412.08651 v1 pith:HKLHZKBY submitted 2024-11-26 eess.AS cs.CLcs.LGcs.SD

Tzu-Ting Yang , Hsin-Wei Wang , Yi-Cheng Wang , Berlin Chen This is my paper · ORCID
classification eess.AScs.CLcs.LGcs.SD
keywords code-switchingautomaticspeechrecognitionintermediateCTClossnon-peakylanguageidentificationmixture-of-expertsposteriorinjectionSEAMEcorpus
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 show that end-to-end code-switching speech recognition, where a speaker mixes Mandarin and English within one sentence, improves when the acoustic encoder is given explicit, frame-accurate language information rather than coarse language labels. The authors take the D-MoE (disentangle-based mixture-of-experts) architecture and add a language-identification block inside the shared encoder, trained with an intermediate CTC loss whose usual sharp spikes are flattened by a non-peaky reweighting; the resulting language-boundary posteriors are then fed into the Mandarin- and English-specific encoders. On the SEAME corpus the full method lowers mixed error rate relative to D-MoE from 20.7 to 20.0 (DevMAN) and 29.0 to 28.4 (DevSGE) without a language model, and to 18.5 and 26.3 with one. If correct, the result indicates that precise alignment of language boundaries, not just language identity, is what helps a model separate cross-language homophones.

What carries the argument

The load-bearing object is the non-peaky CTC loss: a modification of the standard CTC objective in which every frame posterior $P(a_t|X)$ is divided by a sequence-wise softmax prior $P_{\mathrm{prior}}$ raised to the power $\alpha$, giving $\mathcal{L}_{\mathrm{NPC}} = -\log \sum_{\mathbf{a} \in \mathcal{A}} \exp\big(\sum_t \log P(a_t|X) - \alpha \log P_{\mathrm{prior}}\big)$ (Eqs. 5-7). Because the prior is dominated by frequent tokens such as <blank>, the reweighting flattens spikes and lets rarer tokens, here the language labels, accumulate probability mass across frames, so the posteriors read as language boundaries instead of single-frame peaks. The other pieces hang off this object: a language-identification block at the third shared-encoder layer trained on <Mandarin>/<English> targets with intermediate CTC; self-conditioned propagation of its posteriors into later blocks; and injection of those posteriors into the two language-specific encoders, with the whole training objective assembled in Eq. (8) under the disentanglement weight $\lambda$.

What would settle it

Run the model twice, once with $\alpha$ chosen on a held-out tuning split and once with $\alpha$ tuned on DevMAN/DevSGE as in Table 3, and compare MER on those sets; if the advantage over D-MoE evaporates under selection-free tuning, the reported gains are a tuning artifact. Mechanistically, plot the frame-level language posterior for utterances with known switch points and check that the non-peaky loss with $\alpha$ near 0.2-0.3 genuinely flattens the <blank>-dominated spikes and sharpens the boundary transition, as Figure 2 suggests.

Watch

Extended reading notes

Core claim

The central claim is that CTC's well-documented 'peaky' behavior, frame posteriors collapsing into sharp spikes dominated by <blank>, degrades language identification inside a code-switching ASR encoder, and that a reweighted objective fixes it. The paper replaces the standard frame posterior $P(a_t|X)$ in the CTC likelihood with $P(a_t|X)/(P_{\mathrm{prior}})^{\alpha}$, where $P_{\mathrm{prior}}$ is the sequence-wise softmax prior of the model logits, so that high-probability spikes are suppressed and low-probability tokens keep mass across frames, as written in Eqs. (5)-(7). Applied to the third shared-encoder block, retargeted from ASR labels to language-only labels <Mandarin> and <English> and wired into the self-conditioned CTC loop, the non-peaky loss yields language posteriors that mark boundaries rather than isolated peaks; those posteriors are then injected into the language-specific encoders to sharpen their language awareness. The measured consequence is a consistent MER drop on both SEAME test sets relative to the D-MoE baseline, with the Table 3 ablation attributing most of the gain to the non-peaky LID block itself and only about 0.1 MER to the deep posterior injection.

Load-bearing premise

The gains rest on the assumption that picking the non-peaky strength $\alpha$, the disentanglement weight $\lambda$, and the LID block's location using the same DevMAN and DevSGE sets that later report the final MER does not overfit those sets.

Editorial extensions

If this is right

  • The full method beats D-MoE by 0.6-0.8 MER points across both SEAME test sets, with only a modest parameter increase (27.41M vs. 24.46M).
  • Most of the gain comes from the non-peaky LID block itself: SCCTCLID3 with $\alpha$ of 0.2 or 0.3 averages 24.3 MER versus 25.55 for plain SCCTC, while deep language posterior injection contributes only about 0.1 MER.
  • Using language-only targets at every intermediate layer (SCCTCLIDall) hurts slightly versus using them only at the third layer, indicating that low-level language cues help but high-level semantic blocks are disturbed by them.
  • The encoder improvements survive the addition of an external language model, with gains over D-MoE of 0.7 points on DevMAN and 0.8 points on DevSGE when an LM is used.

Reading between the lines

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

  • A quantitative check the paper does not run is measuring the entropy or peakiness of the LID block's posteriors across $\alpha$ values; if the flattening story is right, the optimal $\alpha$ should track the <blank>-posterior mass, which varies with the token set size.
  • The paper's own numbers suggest the deep posterior injection is nearly redundant with the self-conditioned pathway (about 0.1 MER), so the natural next experiment is injecting at multiple layers or with learned gating of the posteriors rather than a single projection.
  • The latency effect noted in the frame-level posteriors implies the language-boundary signal arrives a few frames late; a future system that corrects for this systematic lag, for instance by shifting posteriors earlier, might extract additional gains.
  • Because the non-peaky reweighting adds no trainable parameters, only the scalar $\alpha$, the same trick could carry over to any CTC-based model that consumes its own frame posteriors, such as monolingual ASR with internal subword alignment, though the paper does not test that transfer.
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

3 major / 7 minor

Summary. The paper proposes an extension of the D-MoE architecture for code-switching ASR on the SEAME corpus. The authors add a language identification (LID) block at the third encoder block, train it with a non-peaky CTC loss, use the resulting language posteriors as self-conditioning for later encoder blocks, and inject these posteriors into the language-specific encoders. The reported results are improvements in mixed error rate (MER) over D-MoE: 20.0/28.4 vs. 20.7/29.0 without a language model and 18.5/26.3 vs. 19.2/27.1 with a language model on DevMAN/DevSGE. An ablation sweeps the non-peaky strength alpha on the SCCTC backbone.

Significance. If the reported gains are robust, the paper would provide a useful and parameter-efficient way to inject language information into a mixture-of-experts encoder for code-switching ASR, building on a strong prior system from the same group. The idea of using non-peaky CTC to obtain more reliable language boundaries is interesting and the comparison across two test sets, with and without an external language model, is a strength. However, the main empirical claim is currently under-supported because the ablation is not performed on the D-MoE framework, the injection mechanism is not specified formally, and the hyperparameter selection and single-run results leave open the question of statistical robustness.

major comments (3)
  1. [Section 4.3, Table 3 vs. Section 4.2, Table 2] The central claim that the proposed components produce the Table 2 gains over D-MoE is not supported by the ablation. Table 3 ablates SCCTC, LID3, and non-peaky alpha on a plain Transformer-CTC/SCCTC backbone, not on D-MoE. The SCCTC row of Table 3 (21.1/30.0) is already worse than the D-MoE row of Table 2 (20.7/29.0), so the reference base changes between the main comparison and the ablation. There is no 'D-MoE + SCCTC', 'D-MoE + LID block', or 'D-MoE + deep posterior injection' row. Consequently, the 0.5-0.8 MER improvement in Table 2 could stem from generic intermediate CTC/SCCTC or from the parameter increase from 24.46M to 27.41M, rather than from the proposed LID mechanisms. Please add controlled ablations on the D-MoE framework, or otherwise separate the contributions of each component.
  2. [Section 3.3] The deep language posterior injection mechanism is not specified enough to be reproducible. The text states that internal language posteriors are injected into the language-specific encoders, but gives no equation, tensor shape, or fusion operation, and does not say how this injection differs from the SCCTC conditioning already applied at the LID block. Please define the injection operation explicitly (e.g., concatenation, gating, cross-attention) and state at which layer or module of the language-specific encoders it is applied.
  3. [Section 4.3, Table 3] The non-peaky strength alpha is swept from 0.1 to 0.5 on the same DevMAN and DevSGE sets on which the final MER numbers in Table 2 are reported, and all results appear to be single-run point estimates with no error bars or significance tests. Selection of alpha on the evaluation sets can inflate the observed gain, and the absence of variance estimates makes it difficult to judge whether differences such as 20.1 vs. 20.3 at alpha=0.2 vs. 0.1 are meaningful. Please report multiple seeds with mean and standard deviation, or apply a separate validation partition for alpha selection before final evaluation on the test sets.
minor comments (7)
  1. [Section 3.2, Eq. (6)] The definition of P_prior is ambiguous because s_l is not defined in the text; it should state explicitly that the average is taken over frames of the posterior of the most probable token, or provide the exact formula used for the label prior, so that Eq. (5) is reproducible.
  2. [Section 7, reference [24]] The citation appears as '[24][24][25]' in Section 3.2; please use a single bracket for each reference and check for duplicated citation numbers.
  3. [Throughout] There are numerous typographical errors, including 'uue to', 'bacppropagation', 'blocps', 'LIu', 'peapiness', 'peap-first', and 'SCCTCLIDall'; these should be corrected before publication.
  4. [Section 4.2, Table 2] The text says 'an additional 0.5% reduction in MER across two test datasets', but the per-dataset reductions are 0.7/0.6 without a language model and 0.7/0.8 with a language model on DevMAN/DevSGE; please report the numbers per dataset or clarify that 0.5% is an average.
  5. [Section 3.2] The term 'language boundary alignment loss' is used for what appears to be an application of non-peaky CTC to the LID block; the text should explain how this constitutes an alignment loss and how it relates to boundary accuracy.
  6. [Figure 2] The caption does not explain the overlaid text 'THEN 有 时 候 I THINK' or the time axis; please clarify what is being visualized and how the red and green posterior curves correspond to the two languages.
  7. [Section 4.2, Table 2] The Multi-Transformer-Transducer baseline is listed without a parameter count and is not described in the experimental setup, making that comparison difficult to interpret.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: loss equations are adopted from prior work and the central evidence is an external empirical comparison; caveats are experimental-control issues, not reasoning loops.

full rationale

The proposed objective in Eq. (8) is a weighted combination of standard CTC terms (Eq. (2)), the non-peaky prior from [24][25] (Eqs. (5)-(7)), and D-MoE's disentanglement term from Eq. (1). None of these target quantities is defined in terms of the reported MER, and the non-peaky loss is explicitly cited to prior work rather than re-derived to fit SEAME results. The main evidence is an empirical comparison against external baselines (Transformer CTC, Bi-Encoder, LAE, Multi-Transformer-Transducer) plus the authors' prior D-MoE; the D-MoE self-citation is a baseline, not a load-bearing premise used to prove the new result. The two substantive caveats—selecting alpha on the same DevMAN/DevSGE partitions that produce the headline MER, and ablating on a plain SCCTC backbone rather than on D-MoE—are experimental-control and generalization concerns. They could make the reported 0.5% gain optimistic or leave its source undecomposed, but they do not make the derivation equivalent to its inputs. No circular step rises to the standard of Eq. X = Eq. Y, a fitted parameter renamed as a prediction, or a self-citation chain that forces the conclusion.

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

The method borrows CTC formalism and non-peaky CTC from prior literature. The only new free choices are alpha, lambda, and the LID block placement, all tuned on the evaluation sets. No new physical or theoretical entities are introduced.

free parameters (3)
  • alpha (non-peaky CTC exponent) = 0.2 (tied with 0.3)
    Exponent in the softmax prior of Eq. (5). Swept over 0.1 to 0.5 in Table 3 and selected on DevMAN and DevSGE, the same sets used for final MER reporting.
  • lambda (disentanglement loss weight) = 10
    Hyperparameter weighting LDisentangle in Eq. (8). Chosen to balance loss scales; inherited from the authors' D-MoE prior work, no tuning curve shown.
  • LID block layer (encoder block 3) = Layer 3 of the shared encoder
    The language identification block is placed at encoder block 3. The paper reports that using language targets at all intermediate blocks (SCCTCLIDall) is worse than at layer 3 only (SCCTCLID3), making this an empirical architecture choice.
assumptions (5)
  • standard math CTC's frame-level conditional independence and Viterbi-style alignment formulation (Eqs. 2-3) are valid for the intermediate and final losses.
    Standard CTC formalism from prior work; the paper uses it as the backbone for both LID and ASR objectives.
  • domain assumption Token-level language labels (<Mandarin>, <English>) are available for every training token in SEAME and are accurate.
    Section 3.1 trains the LID block with language-only targets; the paper does not discuss annotation errors or how boundaries are derived from the transcript.
  • domain assumption Non-peaky CTC (Eqs. 5-7) yields more accurate frame-level language boundaries than plain CTC.
    Inherited from references [24] and [25]; the paper's Fig. 2 is illustrative and no quantitative boundary accuracy metric is provided.
  • standard math The sequence-wise softmax prior P_prior computed from model logits (Eq. 6) is a valid proxy for the label prior in the non-peaky CTC loss.
    Definition given in Section 3.2; this is the standard formulation of non-peaky CTC from prior work.
  • ad hoc to paper Placing the LID block at layer 3 and conditioning later blocks on its posterior through SCCTC improves downstream ASR without harming semantic representation.
    Section 3.1 and Table 3 justify this empirically, but no mechanism or analysis is given for why layer 3 is optimal.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Enhancing Code-Switching ASR Leveraging Non-Peaky CTC Loss and Deep Language Posterior Injection." pith.science (2026). https://pith.science/paper/HKLHZKBY

@misc{pith2026241208651,
  author       = {Pith},
  title        = {Pith review of: Enhancing Code-Switching ASR Leveraging Non-Peaky CTC Loss and Deep Language Posterior Injection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HKLHZKBY}},
  note         = {Machine review of arXiv:2412.08651}
}
read the original abstract

Code-switching-where multilingual speakers alternately switch between languages during conversations-still poses significant challenges to end-to-end (E2E) automatic speech recognition (ASR) systems due to phenomena of both acoustic and semantic confusion. This issue arises because ASR systems struggle to handle the rapid alternation of languages effectively, which often leads to significant performance degradation. Our main contributions are at least threefold: First, we incorporate language identification (LID) information into several intermediate layers of the encoder, aiming to enrich output embeddings with more detailed language information. Secondly, through the novel application of language boundary alignment loss, the subsequent ASR modules are enabled to more effectively utilize the knowledge of internal language posteriors. Third, we explore the feasibility of using language posteriors to facilitate deep interaction between shared encoder and language-specific encoders. Through comprehensive experiments on the SEAME corpus, we have verified that our proposed method outperforms the prior-art method, disentangle based mixture-of-experts (D-MoE), further enhancing the acuity of the encoder to languages.

Figures

Figures reproduced from arXiv: 2412.08651 by the authors.

Figure 1
Figure 1. Illustration of our proposed model architecture for code-switching ASR. The symbols <M> and <E> in the figure represent the language indicator labels, <Mandarin> and <English>, respectively [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. shows the language posterior results during CTC computation. The upper part reveals that the traditional CTC method results in spike-shaped predicted tokens. In contrast, the lower part of the figure illustrates how our model, after integrating NPC loss in a novel way, better delineates language boundaries. We also observe a latency phenomenon in the language posterior probabilities at the intermediate layers, which… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

38 extracted references · 32 canonical work pages

  1. [1]

    INTRODUCTION With the widespread adoption of end -to-end (E2E) neural networks, automatic speech recognition (ASR) models have achieved significant progress and notable success across various languages. Unlike traditional ASR systems, which depend on multiple independently trained components such as acoustic and language models, E2E ASR integrates all mod...

    work page

  2. [2]

    D-MOE D-MoE consists primarily of four components: one shared encoder, two language -specific encoders for Chinese and English respectively, and one gating network. At the outset, the shared encoder module co -processes acoustic representations, significantly reducing model parameters while effectively extracting cross -lingual acoustic information. Then,...

    work page

  3. [3]

    METHODLOGY 3.1. LID information block Given the fact that the length 𝐿 of the frame-level acoustic representation X = (x𝑙 ∈ ℝ𝐷|𝑙 = 1, ⋯ , 𝐿) is typically much greater than the length 𝑆 of the token -level ground truth sequence 𝐲 = (y𝑠 ∈ V |𝑠 = 1, ⋯ , 𝑆) , CTC plays a crucial role in mapping X to 𝐲: ℒ𝐶𝑇𝐶 = −log𝑃(𝐲|X) = −log ∑ 𝑃(𝐚|X) 𝐚∈A , (2) where the set...

    work page

  4. [4]

    Experimental setup All experiments are conducted using the SEAME dataset [28], a spontaneous code -switching corpus recorded by Southeast Asian speakers

    EXPERIMENTS 4.1. Experimental setup All experiments are conducted using the SEAME dataset [28], a spontaneous code -switching corpus recorded by Southeast Asian speakers. SEAME contains both intra -sentence and inter-sentence code-switching utterances. The total duration of the recorded audio is approximately 115 hours, and detailed statistics are pr esen...

    work page

  5. [5]

    CONCLUSION In this paper, we aim to improve the acoustic encoding of E2E ASR to address code -switching challenges. By integrating LID information into intermediate encoder layers and applying language boundary alignment loss in a novel way, we have significantly enhanced the robustness of intermediate language posteriors. For future work, we plan to pair...

    work page

  6. [6]

    Any findings and implications in the paper do not necessarily reflect those of the sponsor

    ACKNOWLEDGEMENT This work was supported in part by E.SUN Bank under Grant Numbers 202308 -NTU-02 and 202408 -NTU-02. Any findings and implications in the paper do not necessarily reflect those of the sponsor. Table 3: Comparison of the MER (%) across various methods of applying SCCTC. Model 𝛂 DevMAN DevSGE Average SCCTC - 21.1 30.0 25.55 SCCTCLIDall - 21....

    work page

  7. [7]

    Transformer-Transducers for Code - Switched Speech Recognition ,

    S. Dalmia et al., “Transformer-Transducers for Code - Switched Speech Recognition ,” in Proceedings of International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2021

    work page 2021

  8. [8]

    Pushing the Limits of Semi -Supervised Learning for Automatic Speech Recognition,

    Y. Zhang et al., “Pushing the Limits of Semi -Supervised Learning for Automatic Speech Recognition,” arXiv preprint arXiv:2010.10504, 2020

    arXiv 2010

Show all 38 references
  1. [9]

    Self-Training and Pre -Training are Complementary for Speech Recognition ,

    Q. Xu et al., “Self-Training and Pre -Training are Complementary for Speech Recognition ,” in Proceedings of International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2021

    2021

  2. [10]

    FunASR: a Fundamental End-To-End Speech Recognition Toolkit,

    Z. Gao et al., “FunASR: a Fundamental End-To-End Speech Recognition Toolkit,” in Proceedings of International Speech Communication Association (INTERSPEECH), 2023

    2023

  3. [11]

    MMSpeech: Multi-Modal Multi -Task Encoder-Decoder Pre -Training for Speech Recognition ,

    X. Zhou et al., “MMSpeech: Multi-Modal Multi -Task Encoder-Decoder Pre -Training for Speech Recognition ,” arXiv preprint arXiv:2212.00500, 2022

    2022 arXiv

  4. [12]

    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 ,” in Proceedings of International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2012

    2012

  5. [13]

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

    K. Li et al., “Towards Code-switching ASR for End -to-end CTC Models,” in Proceedings of International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2019

    2019

  6. [14]

    Minimum Word Error Training for Non - Autoregressive Transformer -Based Code -Switching ASR,

    Y. Peng et al., “Minimum Word Error Training for Non - Autoregressive Transformer -Based Code -Switching ASR,” in Proceedings of International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2022

    2022

  7. [15]

    Towards Zero-Shot Code -Switched Speech Recognition,

    B. Yan et al., “Towards Zero-Shot Code -Switched Speech Recognition,” in Proceedings of International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2023

    2023

  8. [16]

    A Study of Pronunciation Problems of English Learners in China,

    F. Zhang et al., “A Study of Pronunciation Problems of English Learners in China,” in Asian Social Science, 2009

    2009

  9. [17]

    Integrating Knowledge in End -To-End Automatic Speech Recognition for Mandarin-English Codeswitching,

    C. -Y. Li et al., “Integrating Knowledge in End -To-End Automatic Speech Recognition for Mandarin-English Codeswitching,” in Proceedings of International Conference on Asian Language Processing (IALP), 2019

    2019

  10. [18]

    LAE: Language-Aware Encoder for Monolingual and Multilingual ASR,

    J. Tian et al., “LAE: Language-Aware Encoder for Monolingual and Multilingual ASR,” arXiv preprint arXiv:2206.02093, 2022

    2022 arXiv

  11. [19]

    Language-Specific Characteristic Assistance for Code-Switching Speech Recognition,

    T. Song et al., “Language-Specific Characteristic Assistance for Code-Switching Speech Recognition,” in Proceedings of International Speech Communication Association (INTERSPEECH), 2022

    2022

  12. [20]

    Bi-encoder Transformer Network for Mandarin- English Code-Switching Speech Recognition Using Mixture of Experts ,

    Y. Lu et al., “Bi-encoder Transformer Network for Mandarin- English Code-Switching Speech Recognition Using Mixture of Experts ,” in Proceedings of International Speech Communication Association (INTERSPEECH), 2020

    2020

  13. [21]

    Multi -Encoder-uecoder Transformer for Code-Switching Speech Recognition ,

    X. Zhou et al ., “Multi -Encoder-uecoder Transformer for Code-Switching Speech Recognition ,” arXiv preprint arXiv:2006.10414, 2020

    2006 arXiv

  14. [22]

    Language-Specific Characteristic Assistance for Codeswitching Speech Recognition ,

    T. Song et al., “Language-Specific Characteristic Assistance for Codeswitching Speech Recognition ,” in Proceedings of International Speech Communication Association (INTERSPEECH), 2022

    2022

  15. [23]

    An Effective Mixture -of-Experts Approach for Code -Switching Speech Recognition Leveraging Encoder Disentanglement ,

    T.-T. Yang et al., “An Effective Mixture -of-Experts Approach for Code -Switching Speech Recognition Leveraging Encoder Disentanglement ,” in Proceedings of International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2024

    2024

  16. [24]

    Intermediate Loss Regularization for CTC- Based Speech Recognition,

    J. Lee et al., “Intermediate Loss Regularization for CTC- Based Speech Recognition,” in Proceedings of International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2021

    2021

  17. [25]

    Hierarchical Multitask Learning with CTC,

    R. Sanabria et al., “Hierarchical Multitask Learning with CTC,” in Proceedings of IEEE Spoken Language Technology Workshop (SLT), 2018

    2018

  18. [26]

    Relaxing the Conditional Independence Assumption Of CTC-Based ASR By Conditioning on Intermediate Predictions ,

    J. Nozaki et al., “Relaxing the Conditional Independence Assumption Of CTC-Based ASR By Conditioning on Intermediate Predictions ,” in Proceedings of International Speech Communication Association (INTERSPEECH), 2021

    2021

  19. [27]

    Mask CTC: Non-Autoregressive End-To- End ASR with CTC and Mask Predict ,

    Y. Higuchi et al., “Mask CTC: Non-Autoregressive End-To- End ASR with CTC and Mask Predict ,” in Proceedings of International Speech Communication Association (INTERSPEECH), 2020

    2020

  20. [28]

    AlignRefine: Non-Autoregressive Speech Recognition via Iterative Realignment,

    E. A. Chi et al., “AlignRefine: Non-Autoregressive Speech Recognition via Iterative Realignment,” in Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL-HLT), 2021

    2021

  21. [29]

    Why Does CTC Result in Peaky Behavior?

    A. Zeyer et al., “Why Does CTC Result in Peaky Behavior? ” arXiv preprint arXiv:2105.14849, 2021

    2021 arXiv

  22. [30]

    Less Peaky and More Accurate CTC Forced Alignment by Label Priors,

    R. Huang et al., “Less Peaky and More Accurate CTC Forced Alignment by Label Priors,” in Proceedings of International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2024

    2024

  23. [31]

    Peap-First CTC: Reducing the Peak Latency of CTC Models by Applying Peak -First Regularization,

    Z. Tian et al., “Peap-First CTC: Reducing the Peak Latency of CTC Models by Applying Peak -First Regularization,” in Proceedings of International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2023

    2023

  24. [32]

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

    H. Liu, et al ., “Reducing Language Confusion for Code - Switching Speech Recognition with Token -Level Language Diarization,” in Proceedings of International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2023

    2023

  25. [33]

    Enhancing Code-Switching Speech Recognition with Interactive Language Biases ,

    H. Liu, et al ., “Enhancing Code-Switching Speech Recognition with Interactive Language Biases ,” in Proceedings of International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2024

    2024

  26. [34]

    SEAME: A Mandarin -English Code- Switching Speech Corpus in South -East Asia,

    u.-C. Lyu et al., “SEAME: A Mandarin -English Code- Switching Speech Corpus in South -East Asia,” in Proceedings of International Speech Communication Association (INTERSPEECH), 2010

    2010

  27. [35]

    The ASRU 2019 Mandarin -English Code- Switching Speech Recognition Challenge: Open uatasets, Tracps, Methods and Results,

    X. Shi et al ., “The ASRU 2019 Mandarin -English Code- Switching Speech Recognition Challenge: Open uatasets, Tracps, Methods and Results, ” arXiv preprint arXiv:2007.05916, 2020

    2019 arXiv

  28. [36]

    AISHELL -1: An Open-Source Mandarin Speech Corpus and A Speech Recognition Baseline ,

    H. Bu et al ., “AISHELL -1: An Open-Source Mandarin Speech Corpus and A Speech Recognition Baseline ,” in Proceedings of the international coordinating committee on speech databases and speech I/O systems and assessment (O- COCOSDA), 2017

    2017

  29. [37]

    Librispeech: An ASR Corpus Based on Public uomain Audio Boops,

    V . Panayotov et al., “Librispeech: An ASR Corpus Based on Public uomain Audio Boops, ” in Proceedings of International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2015

    2015

  30. [38]

    Low-Resource Speech Recognition and Dialect Identification of Irish in A Multi-Task Framework,

    L. Lonergan et al., “Low-Resource Speech Recognition and Dialect Identification of Irish in A Multi-Task Framework,” arXiv preprint arXiv:2405.01293, 2024

    2024 arXiv

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.