Pith. sign in

REVIEW 3 major objections 5 minor 66 references

Enhancing the Robustness of Contextual ASR to Varying Biasing Information Volumes Through Purified Semantic Correlation Joint Modeling

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read The paper claims that contextual ASR can be made stable as biasing-list length varies by intersecting list-, phrase-, and token-level semantic correlations, reporting relative F1 gains of up to 21.34% on AISHELL-1 and 28.46% on KeSpeech.

desk verdict Solid extension of FineCoS/SeACo with explicit multi-granularity supervision and a purification trick, but the headline F1 numbers don't recompute and the phrase-level scorer's train/inference mismatch is unresolved. read the letter →

arxiv 2509.05908 v1 pith:WJAG6LDF submitted 2025-09-07 cs.CL cs.SDeess.AS

Yue Gu , Zhihao Du , Ying Shi , Shiliang Zhang , Qian Chen , Jiqing Han This is my paper · ORCID
classification cs.CLcs.SDeess.AS
keywords contextualASRsemanticcorrelationjointmodelinggroupcompetitivepurificationvarying-lengthbiasinglistscross-attentionspeechrecognitionphrasenon-autoregressive
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

Contextual automatic speech recognition (ASR) lets a recognition system be steered by a personalized list of phrases, such as contact names or locations, but accuracy drops as that list grows longer. The paper argues that at each decoding step only a small amount of the biasing list is actually relevant to the speech being recognized, so the right approach is to find that relevant subset instead of attending to the entire list. PSC-Joint scores relevance at three granularities—the whole list, individual phrases, and individual tokens—and keeps only the intersection of those scores. A grouped-and-competitive purification step prunes irrelevant phrases before the intersection is computed, keeping inference fast. Across biasing lists of 51 to over 2,000 phrases, the paper reports relative F1 improvements of up to 21.34% on AISHELL-1 and 28.46% on KeSpeech over existing cross-attention contextual models.

What carries the argument

The central object is the semantic correlation joint modeling (SC-Joint) predictor, a multi-granularity relevance scorer whose three heads produce list-level, phrase-level, and token-level correlation scores for each ASR decoding step. The joint score is their intersection via Eq. (21): at each vocabulary token, it multiplies the smoothed list-level score, the smoothed phrase-level score for the phrase containing that token, and the token-level score, then takes the maximum over phrases before the softmax. Around this sits a grouped-and-competitive purification (GCP) loop that splits the biasing list into random groups, keeps the top candidates per group using list- and phrase-level scores, and repeats, shrinking the list before joint modeling. Together, the intersection and the purification carry the argument: they concentrate the biasing signal on the few tokens that all granularities agree on, rather than letting the whole list dilute attention.

What would settle it

Take a trained PSC-Joint model and a test set of sentences each containing exactly one target phrase from a 1,000+-entry biasing list; at the decoding steps where that phrase occurs, compare the max-over-heads attention score (Eq. 13) with the training-time cosine similarity between the aggregated biased embedding and the phrase embedding (Eq. 11). If the inference-time score does not rank the correct phrase first in well above chance, or if F1 collapses when a second co-occurring target phrase is deleted from the list, the joint intersection is not doing the claimed work.

Watch

Extended reading notes

Core claim

The central claim is that cross-attention-based contextual ASR degrades with longer biasing lists because the model integrates too much irrelevant biasing information, not because attention stops working. The paper argues that for any given ASR intermediate representation only a limited amount of biasing information is relevant, regardless of list length. PSC-Joint operationalizes relevance as the intersection of three learned correlation scores: a list-level binary judgment of whether the current decoding step relates to the list at all, a phrase-level score selecting among competing phrases, and a token-level score over the ASR vocabulary. The intersection is computed as $Q^{\mathrm{bias}}_u = \mathrm{Softmax}(\max_m Q^{\mathrm{slist}}_u Q^{\mathrm{sphr}}_{u,m} \Phi_{m,v} Q^{\mathrm{tok}}_{u,v})$, where $\Phi_{m,v}$ is 1 when vocabulary token $v$ appears in phrase $m$, so only tokens belonging to phrases that all three levels agree on receive high probability. On a non-autoregressive ASR backbone, the paper reports relative F1 improvements of up to 21.34% on AISHELL-1 and 28.46% on KeSpeech over existing cross-attention contextual models, with the gains largest on the longest biasing lists.

Load-bearing premise

The load-bearing premise is that the phrase-level relevance score extracted from the attention matrix at inference agrees with the phrase-level score optimized during training, even though training assumes one target phrase per sentence and uses a sentence-level aggregated embedding; if the two diverge for long lists or for sentences with multiple target phrases, the intersection and the purification both lose reliability.

Editorial extensions

If this is right

  • Contextual ASR can be made stable to biasing-list length without injecting more distractors during training, which previously traded performance on short lists for performance on long lists.
  • Explicit supervision at list, phrase, and token granularities reduces both over-biasing and under-biasing compared with a single combined bias probability.
  • Grouped competitive purification keeps inference time roughly linear in list length while retaining 92.06% of target phrases on the longest evaluated list.
  • On low-inclusion, high-out-of-vocabulary lists such as person names, the approach gives its largest relative F1 gains, indicating better control of the over-biasing failure mode.

Reading between the lines

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

  • The agreement-across-granularities principle is more general than ASR: any system that retrieves from a growing context list—for example, retrieval-augmented generation—could use the intersection of coarse and fine relevance signals as a confidence filter. This is an extension the paper does not state.
  • The training/inference mismatch in the phrase-level score predicts a specific failure mode: sentences containing two or more target phrases, or lists far longer than the training lists, may misrank phrase scores. A natural test is to retrain phrase-level scoring with per-step supervision on multi-phrase sentences and measure whether the F1 gap closes.
  • Because purification groups phrases randomly, acoustically or semantically similar phrases may compete within a group and both be filtered out; grouping by semantic similarity rather than at random could raise the reported 92.06% retention rate. This is a testable variant, not a claim of the paper.
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

3 major / 5 minor

Summary. The paper proposes PSC-Joint, a contextual ASR method built on the non-autoregressive Paraformer backbone. It defines list-level, phrase-level, and token-level semantic correlation scores between ASR intermediate representations and biasing phrases, jointly models these correlations through an intersection score, and adds a grouped competitive purification (GCP) mechanism to prune the biasing list before joint modeling. Experiments on AISHELL-1 and KeSpeech with AISHELL-NER biasing lists report F1 and CER comparisons against ColDec, FineCoS, and SeACo, together with an ablation study, a retention-rate analysis for purification, and hyperparameter sensitivity analysis.

Significance. If the claims hold, the cross-granularity intersection idea is a useful contribution to contextual ASR with long biasing lists, and the GCP mechanism addresses a real computational bottleneck. The paper has concrete strengths: it evaluates on two corpora, includes ablations of the three losses, reports a retention-rate metric for purification, provides hyperparameter sensitivity heatmaps, and offers a code link. However, the headline quantitative gains do not match the numbers in Tables II and IV, and the inference-time phrase-level score is not the quantity optimized during training. These issues must be resolved before the robustness claim is established.

major comments (3)
  1. [Abstract and Section IV-C] The claimed relative F1 improvements do not match Tables II and IV. For AISHELL-1, PSC-Joint-P (GCP) vs ColDec gives per-list relative F1 changes ranging from about -1.4% on NE-51 to +4.6% on NE-101, with an average around 2.8%, not 21.34%; the text's claim of 25.57% relative improvement over ColDec on NE-101 is also inconsistent with the table (89.04 vs 84.75, about 5.1%). For KeSpeech, the largest per-list relative F1 gain over ColDec is about 15.2% on PER-414, not 28.46%, and the stated 34.49%, 33.40%, and 43.76% average improvements over the three baselines are not recovered by averaging the listed F1 scores. Please correct the abstract and text, or provide the exact computation used for each reported percentage.
  2. [Section III-C/III-D and Algorithm 1] The phrase-level correlation used at inference is not the quantity optimized during training. Training optimizes Eq. (12) on the sentence-aggregated embedding e_phr' computed in Eq. (10) under the single-phrase assumption, while inference uses Eq. (13), the per-head maximum of the cross-attention matrix. Because Eq. (21) and Algorithm 1 line 8 both depend on this inference-time Q_phr, the paper should provide evidence that Eq. (13) is calibrated with Eq. (11), for example by reporting ranking agreement or correlation on held-out multi-phrase sentences, and by measuring retention as a function of Q_phr. The paper's own retention rates (92.06% GCP vs 86.26% OCP) show that target-phrase deletion already occurs; without such evidence, the robustness claim for long lists is not fully established.
  3. [Tables II-IV] No error bars or significance tests are reported. Some gains are very small relative to the likely run-to-run variation, such as KeSpeech Test NE-51 where PSC-Joint-P (GCP) achieves 91.89 F1 versus ColDec's 91.79, while other gains are large. Without variance estimates over multiple runs or seeds, or paired significance tests such as bootstrap or matched-pair tests, the reader cannot judge whether the observed pattern is statistically reliable. Please add such analysis or explicitly state the number of repeated runs.
minor comments (5)
  1. [Equation (2)] Equation (2) is dimensionally inconsistent as written: Q_list_u is a scalar, Q_phr_u is an M-vector, and Q_tok_u is a V-vector, so (Q_list_u (Q_phr_u)^T Φ Q_tok_u) is a scalar, whereas Q_bias_u is claimed to be in R^V. Equation (21) appears to be the intended per-v computation; please fix Eq. (2) or clarify the notation.
  2. [Section III-D, Eq. (17)] The window index range [u-L'+1, u+L'-1] in Eq. (17) can fall outside the valid convolution output for boundary decoding steps; please specify the boundary handling explicitly.
  3. [Algorithm 1, line 8] The expression (Q_list_g ⊙ Q_phr_g).topk(n_top) is ambiguous because Q_list_g is a scalar or binary flag while Q_phr_g has per-step and per-phrase dimensions; please specify how the per-group scores are aggregated over decoding steps before the top-k selection.
  4. [Section IV-C] The phrase 'average relative F1 score improvements' is used multiple times without a definition; please specify whether the average is over biasing list lengths, over baseline models, or both, and state the exact formula.
  5. [Figure 7] Figure 7 is dense and the caption states that only key elements are shown; a clearer annotation of which matrix corresponds to the joint modeling score in Eq. (21) would help the reader follow the instance analysis.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the joint-correlation outputs are supervised on held-out lists and evaluated empirically, so the reported gains are not baked into the definitions.

full rationale

The paper's load-bearing quantities are network outputs trained with explicit losses (Eqs. 9, 12, 14) and then evaluated on held-out AISHELL-1/KeSpeech models with AISHELL-NER biasing lists. The joint intersection (Eq. 21) is a compositional operation over three predicted correlation scores, not a restatement of the ground-truth labels; final recognition comes from interpolation (Eq. 23) and is measured by CER/F1, so the empirical claims are not self-fulfilling. The purification mechanism (Algorithm 1) filters phrases using predicted list- and phrase-level scores, and the paper explicitly measures the failure of this filtering via the retention rate (92.06% vs. 86.26% for GCP vs. OCP), showing that target-removal errors are not assumed away. The train/inference mismatch for phrase-level correlation—training uses a sentence-aggregated contrastive loss (Eqs. 10–12) while inference uses max-over-heads attention (Eq. 13)—is a calibration and robustness risk, but it is not circular: the max-attention quantity is not equal by construction to the contrastive target. Self-citations to Paraformer, FunASR, and SeACo are used as backbone/baseline tooling rather than load-bearing uniqueness theorems, and no equation reduces to its own input by definition. Apparent arithmetic inconsistencies in the abstract's relative F1 improvements are correctness concerns, not circularity concerns.

Assumptions & free parameters 7 free parameters · 4 assumptions · 0 invented entities

The method is a supervised engineering contribution. It introduces no new physical entities, but it depends on several training and inference assumptions and on multiple hyperparameters tuned on the evaluation setup. The free parameters are standard machine learning hyperparameters, not fitted constants of a theory.

free parameters (7)
  • Focal loss alpha = 0.75
    Balances positive and negative list-level labels; set in Section IV.B.
  • Focal loss gamma = 2.0
    Modulates easy versus hard list-level labels; set in Section IV.B.
  • Triangular window smoothing weight omega = 0.6
    Smooths the list-level correlation in Eq. (15); tuned in Section IV.E and used in the interpolation of Eq. (23).
  • Purification group size = 75
    Group size for divide-and-conquer purification; set close to the average training list length and tuned in Fig. 6.
  • Purification rounds n_r = 2
    Number of competitive purification rounds; chosen for computational cost in Section IV.E.
  • List-level correlation threshold thres_list = 0.5
    Binarizes Q_list during purification in Algorithm 1; set to the median of the score interval without tuning.
  • Top-k n_top = 10
    Number of phrases retained per group in each purification round; tuned in Fig. 5.
assumptions (4)
  • domain assumption Each sentence contains only one biasing phrase during training
    Stated in Section III-C: "To simplify this classification at the training stage, we assume that each sentence contains only one biasing phrase." Test lists contain thousands of phrases and the instance analysis in Section IV.G contains two target phrases.
  • domain assumption Cross-granularity consistency of relevant biasing information
    Section III-D postulates that if an ASR representation is relevant to a biasing phrase, list-, phrase-, and token-level correlations are simultaneously high. This is the basis for computing their intersection in Eq. (2) and Eq. (21).
  • domain assumption Inference-time attention max equals trained phrase-level correlation
    Training uses a contrastive loss on the sentence-aggregated embedding e_phr' (Eqs. 10-12); inference uses the maximum over attention heads (Eq. 13). The paper does not verify that these reflect the same notion of phrase relevance.
  • domain assumption AISHELL-NER named entities represent realistic biasing lists
    Section IV.A builds evaluation lists from Chinese named entities with inclusion rates from about 10% to 35% and OOV rates from about 31% to 65%; the authors assume this captures real variation in list length and content.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Enhancing the Robustness of Contextual ASR to Varying Biasing Information Volumes Through Purified Semantic Correlation Joint Modeling." pith.science (2026). https://pith.science/paper/WJAG6LDF

@misc{pith2026250905908,
  author       = {Pith},
  title        = {Pith review of: Enhancing the Robustness of Contextual ASR to Varying Biasing Information Volumes Through Purified Semantic Correlation Joint Modeling},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WJAG6LDF}},
  note         = {Machine review of arXiv:2509.05908}
}
read the original abstract

Recently, cross-attention-based contextual automatic speech recognition (ASR) models have made notable advancements in recognizing personalized biasing phrases. However, the effectiveness of cross-attention is affected by variations in biasing information volume, especially when the length of the biasing list increases significantly. We find that, regardless of the length of the biasing list, only a limited amount of biasing information is most relevant to a specific ASR intermediate representation. Therefore, by identifying and integrating the most relevant biasing information rather than the entire biasing list, we can alleviate the effects of variations in biasing information volume for contextual ASR. To this end, we propose a purified semantic correlation joint modeling (PSC-Joint) approach. In PSC-Joint, we define and calculate three semantic correlations between the ASR intermediate representations and biasing information from coarse to fine: list-level, phrase-level, and token-level. Then, the three correlations are jointly modeled to produce their intersection, so that the most relevant biasing information across various granularities is highlighted and integrated for contextual recognition. In addition, to reduce the computational cost introduced by the joint modeling of three semantic correlations, we also propose a purification mechanism based on a grouped-and-competitive strategy to filter out irrelevant biasing phrases. Compared with baselines, our PSC-Joint approach achieves average relative F1 score improvements of up to 21.34% on AISHELL-1 and 28.46% on KeSpeech, across biasing lists of varying lengths.

Figures

Figures reproduced from arXiv: 2509.05908 by the authors.

Figure 1
Figure 1. The schematic diagram of semantic correlation joint modeling method. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Schematic diagram of (a) current attention-based contextual NAR [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Schematic diagram of the purified semantic correlation joint modeling approach. (a) Non-autoregressive ASR backbone model: Paraformer. (b) Multi [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Comparison of contextual ASR models tested on varying-length [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Sensitivity analysis of hyperparameter tuning for PSC-Joint-P (OCP) [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: Sensitivity analysis of hyperparameter tuning for PSC-Joint-P (GCP) [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: An instance. Note that, for demonstration, only key elements of [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

66 extracted references · 64 canonical work pages

  1. [1]

    Connec- tionist temporal classification: labelling unsegmented sequence data with recurrent neural networks,

    A. Graves, S. Fern ´andez, F. J. Gomez, and J. Schmidhuber, “Connec- tionist temporal classification: labelling unsegmented sequence data with recurrent neural networks,” inICML, vol. 148, 2006, pp. 369–376

    work page 2006

  2. [2]

    Speech recognition with deep recurrent neural networks,

    A. Graves, A. Mohamed, and G. E. Hinton, “Speech recognition with deep recurrent neural networks,” inICASSP, 2013, pp. 6645–6649

    work page 2013

  3. [3]

    Towards end-to-end speech recognition with recurrent neural networks,

    A. Graves and N. Jaitly, “Towards end-to-end speech recognition with recurrent neural networks,” inICML, vol. 32, 2014, pp. 1764–1772

    work page 2014

  4. [4]

    Attention-based models for speech recognition,

    J. Chorowski, D. Bahdanau, D. Serdyuk, K. Cho, and Y . Bengio, “Attention-based models for speech recognition,” inNIPS, 2015, pp. 577–585

    work page 2015

  5. [5]

    Listen, attend and spell: A neural network for large vocabulary conversational speech recognition,

    W. Chan, N. Jaitly, Q. V . Le, and O. Vinyals, “Listen, attend and spell: A neural network for large vocabulary conversational speech recognition,” inICASSP, 2016, pp. 4960–4964

    work page 2016

  6. [6]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, and et.al., “Attention is all you need,” inNeurIPS, 2017, pp. 5998–6008

    work page 2017

  7. [7]

    Con- former: Convolution-augmented transformer for speech recognition,

    A. Gulati, J. Qin, C. Chiu, N. Parmar, Y . Zhang, J. Yu, and et.al., “Con- former: Convolution-augmented transformer for speech recognition,” in INTERSPEECH, 2020, pp. 5036–5040

    work page 2020

  8. [8]

    End-to-end speech recognition: A survey,

    R. Prabhavalkar, T. Hori, T. N. Sainath, R. Schl ¨uter, and S. Watanabe, “End-to-end speech recognition: A survey,”IEEE ACM Trans. Audio Speech Lang. Process., vol. 32, pp. 325–351, 2024

    work page 2024

Show all 66 references
  1. [9]

    Bringing contextual information to google speech recognition,

    P. S. Aleksic, M. Ghodsi, A. H. Michaely, C. Allauzen, K. B. Hall, B. Roark, and et.al., “Bringing contextual information to google speech recognition,” inINTERSPEECH, 2015, pp. 468–472

    2015

  2. [10]

    Composition-based on-the-fly rescoring for salient n-gram biasing,

    K. Hall, E. Cho, C. Allauzen, F. Beaufays, N. Coccaro, K. Nakajima, and et.al., “Composition-based on-the-fly rescoring for salient n-gram biasing,” inINTERSPEECH, 2015, pp. 1418–1422

    2015

  3. [11]

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

    I. Williams, A. Kannan, P. S. Aleksic, D. Rybach, and T. N. Sainath, “Contextual speech recognition in end-to-end neural network systems using beam search,” inINTERSPEECH, 2018, pp. 2227–2231

    2018

  4. [12]

    End-to-end contextual speech recognition using class language models and a token passing decoder,

    Z. Chen, M. Jain, Y . Wang, M. L. Seltzer, and C. Fuegen, “End-to-end contextual speech recognition using class language models and a token passing decoder,” inICASSP. IEEE, 2019, pp. 6186–6190

    2019

  5. [13]

    Shallow-fusion end-to-end contextual biasing,

    D. Zhao, T. N. Sainath, D. Rybach, P. Rondon, D. Bhatia, B. Li, and et.al., “Shallow-fusion end-to-end contextual biasing,” inINTER- SPEECH, 2019, pp. 1418–1422

    2019

  6. [14]

    Streaming end-to-end speech recognition for mobile devices,

    Y . He, T. N. Sainath, R. Prabhavalkar, I. McGraw, R. Alvarez, D. Zhao, and et.al., “Streaming end-to-end speech recognition for mobile devices,” inICASSP, 2019, pp. 6381–6385

    2019

  7. [15]

    Class lm and word mapping for contextual biasing in end-to-end asr,

    R. Huang, O. Abdel-hamid, X. Li, and G. Evermann, “Class lm and word mapping for contextual biasing in end-to-end asr,” inINTERSPEECH, 2020, pp. 4348–4351

    2020

  8. [16]

    Minimum word error rate training with language model fusion for end-to-end speech recognition,

    Z. Meng, Y . Wu, N. Kanda, L. Lu, X. Chen, G. Ye, and et.al., “Minimum word error rate training with language model fusion for end-to-end speech recognition,” inINTERSPEECH, 2021, pp. 2596–2600

    2021

  9. [17]

    Implementing contextual biasing in GPU decoder for online ASR,

    I. Nigmatulina, S. R. Madikeri, E. Villatoro-Tello, P. Motl ´ıcek, J. Zuluaga-Gomez, K. Pandia, and et.al., “Implementing contextual biasing in GPU decoder for online ASR,” inINTERSPEECH, 2023, pp. 4494–4498

    2023

  10. [18]

    Personal- ization of ctc-based end-to-end speech recognition using pronunciation- driven subword tokenization,

    Z. Lei, E. Pusateri, S. Han, L. Liu, M. Xu, T. Ng, and et.al., “Personal- ization of ctc-based end-to-end speech recognition using pronunciation- driven subword tokenization,” inICASSP, 2024, pp. 10 096–10 100

    2024

  11. [19]

    Ctc-assisted llm-based contextual asr,

    G. Yang, Z. Ma, Z. Gao, S. Zhang, and X. Chen, “Ctc-assisted llm-based contextual asr,” inSLT, 2024, pp. 126–131

    2024

  12. [20]

    Mala-asr: Multimedia-assisted llm-based asr,

    G. Yang, Z. Ma, F. Yu, Z. Gao, S. Zhang, and X. Chen, “Mala-asr: Multimedia-assisted llm-based asr,”arXiv preprint arXiv:2406.05839, 2024

    2024 arXiv

  13. [21]

    Contextualized streaming end-to-end speech recognition with trie-based deep biasing and shallow fusion,

    D. Le, M. Jain, G. Keren, S. Kim, Y . Shi, J. Mahadeokar, and et.al., “Contextualized streaming end-to-end speech recognition with trie-based deep biasing and shallow fusion,” inINTERSPEECH, 2021, pp. 1772– 1776

    2021

  14. [22]

    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,” inSLT, 2021, pp. 251–257

    2021

  15. [23]

    Tree-constrained pointer generator for end-to-end contextual speech recognition,

    G. Sun, C. Zhang, and P. C. Woodland, “Tree-constrained pointer generator for end-to-end contextual speech recognition,” inASRU, 2021, pp. 780–787

    2021

  16. [24]

    Selective biasing with trie-based contextual adapters for personalised speech recognition using neural transducers,

    P. Harding, S. Tong, and S. Wiesler, “Selective biasing with trie-based contextual adapters for personalised speech recognition using neural transducers,” inINTERSPEECH, 2023, pp. 256–260

    2023

  17. [25]

    Minimising biasing word errors for contextual ASR with the tree-constrained pointer generator,

    G. Sun, C. Zhang, and P. C. Woodland, “Minimising biasing word errors for contextual ASR with the tree-constrained pointer generator,”IEEE ACM Trans. Audio Speech Lang. Process., vol. 31, pp. 345–354, 2023

    2023

  18. [26]

    Phoneme-aware encoding for prefix-tree-based contextual asr,

    H. Futami, E. Tsunoo, Y . Kashiwagi, H. Ogawa, S. Arora, and S. Watan- abe, “Phoneme-aware encoding for prefix-tree-based contextual asr,” in ICASSP, 2024, pp. 10 641–10 645

    2024

  19. [27]

    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,” inSLT, 2018, pp. 418–425

    2018

  20. [28]

    Contextual speech recognition with difficult negative training examples,

    U. Alon, G. Pundak, and T. N. Sainath, “Contextual speech recognition with difficult negative training examples,” inICASSP, 2019, pp. 6440– 6444

    2019

  21. [29]

    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,” inINTERSPEECH, 2020, pp. 11–15

    2020

  22. [30]

    Cif-based collaborative decoding for end-to-end contextual speech recognition,

    M. Han, L. Dong, S. Zhou, and B. Xu, “Cif-based collaborative decoding for end-to-end contextual speech recognition,” inICASSP, 2021, pp. 6528–6532

    2021

  23. [31]

    RNN-T based open-vocabulary keyword spotting in mandarin with multi-level detection,

    Z. Liu, T. Li, and P. Zhang, “RNN-T based open-vocabulary keyword spotting in mandarin with multi-level detection,” inICASSP, 2021, pp. 5649–5653

    2021

  24. [32]

    Context-aware transformer transducer for speech recogni- tion,

    F. Chang, J. Liu, M. Radfar, A. Mouchtaris, M. Omologo, A. Rastrow, and et.al., “Context-aware transformer transducer for speech recogni- tion,” inASRU, 2021, pp. 503–510

    2021

  25. [33]

    Improving end-to-end contextual speech recognition with fine-grained contextual knowledge selection,

    M. Han, L. Dong, Z. Liang, M. Cai, S. Zhou, Z. Ma, and et.al., “Improving end-to-end contextual speech recognition with fine-grained contextual knowledge selection,” inICASSP, 2022, pp. 8532–8536. JOURNAL OF LATEX CLASS FILES, VOL. 18, NO. 9, SEPTEMBER 2020 12

    2022

  26. [34]

    Fast contextual adaptation with neural associa- tive memory for on-device personalized speech recognition,

    T. Munkhdalai, K. C. Sim, A. Chandorkar, F. Gao, M. Chua, T. Strohman, and et.al., “Fast contextual adaptation with neural associa- tive memory for on-device personalized speech recognition,” inICASSP, 2022, pp. 6632–6636

    2022

  27. [35]

    Contextualized end-to-end speech recognition with contextual phrase prediction network,

    K. Huang, A. Zhang, Z. Yang, P. Guo, B. Mu, T. Xu, and et.al., “Contextualized end-to-end speech recognition with contextual phrase prediction network,” inINTERSPEECH, 2023, pp. 4933–4937

    2023

  28. [36]

    Approx- imate nearest neighbour phrase mining for contextual speech recogni- tion,

    M. J. R. Bleeker, P. Swietojanski, S. Braun, and X. Zhuang, “Approx- imate nearest neighbour phrase mining for contextual speech recogni- tion,” inINTERSPEECH, 2023, pp. 939–943

    2023

  29. [37]

    Two stage contextual word filtering for context bias in unified streaming and non- streaming transducer,

    Z. Yang, S. Sun, X. Wang, Y . Zhang, L. Ma, and L. Xie, “Two stage contextual word filtering for context bias in unified streaming and non- streaming transducer,” inINTERSPEECH, 2023, pp. 3257–3261

    2023

  30. [38]

    Effective training of attention-based contextual biasing adapters with synthetic audio for personalised ASR,

    B. Naowarat, P. Harding, P. D’Alterio, S. Tong, and B. A. S. Hasan, “Effective training of attention-based contextual biasing adapters with synthetic audio for personalised ASR,” inINTERSPEECH, 2023, pp. 1264–1268

    2023

  31. [39]

    Cb- conformer: Contextual biasing conformer for biased word recognition,

    Y . Xu, B. Liu, Q. Huang, X. Song, Z. Wu, S. Kang, and et.al., “Cb- conformer: Contextual biasing conformer for biased word recognition,” inICASSP, 2023, pp. 1–5

    2023

  32. [40]

    Slot-triggered contextual bias- ing for personalized speech recognition using neural transducers,

    S. Tong, P. Harding, and S. Wiesler, “Slot-triggered contextual bias- ing for personalized speech recognition using neural transducers,” in ICASSP, 2023, pp. 1–5

    2023

  33. [41]

    Locality enhanced dynamic biasing and sam- pling strategies for contextualASR,

    M. A. Jalal, P. P. Parada, G. Pavlidis, V . Moschopoulos, K. Saravanan, C. Kontoulis, and et.al., “Locality enhanced dynamic biasing and sam- pling strategies for contextualASR,” inASRU, 2023, pp. 1–8

    2023

  34. [42]

    Adaptive contextual biasing for transducer based streaming speech recognition,

    T. Xu, Z. Yang, K. Huang, P. Guo, A. Zhang, B. Li, and et.al., “Adaptive contextual biasing for transducer based streaming speech recognition,” inINTERSPEECH, 2023, pp. 1668–1672

    2023

  35. [43]

    Gated contextual adapters for selective contextual biasing in neural transducers,

    A. Alexandridis, K. M. Sathyendra, G. P. Strimel, F. Chang, A. Rastrow, N. Susanj, and et.al., “Gated contextual adapters for selective contextual biasing in neural transducers,” inICASSP, 2023, pp. 1–5

    2023

  36. [44]

    Robust acoustic and semantic contextual biasing in neural transducers for speech recognition,

    X. Fu, K. M. Sathyendra, A. Gandhe, J. Liu, G. P. Strimel, R. McGowan, and et.al., “Robust acoustic and semantic contextual biasing in neural transducers for speech recognition,” inICASSP, 2023, pp. 1–5

    2023

  37. [45]

    Dual-mode nam: Effective top-k context injection for end-to-end asr,

    Z. Wu, T. Munkhdalai, P. Rondon, G. Pundak, K. C. Sim, and C. Li, “Dual-mode nam: Effective top-k context injection for end-to-end asr,” inINTERSPEECH, 2023, pp. 221–225

    2023

  38. [46]

    Contex- tualized automatic speech recognition with attention-based bias phrase boosted beam search,

    Y . Sudo, M. Shakeel, Y . Fukumoto, Y . Peng, and S. Watanabe, “Contex- tualized automatic speech recognition with attention-based bias phrase boosted beam search,” inICASSP, 2024, pp. 10 896–10 900

    2024

  39. [47]

    Improving asr contextual biasing with guided attention,

    J. Tang, K. Kim, S. Shon, F. Wu, and P. Sridhar, “Improving asr contextual biasing with guided attention,” inICASSP, 2024, pp. 12 096– 12 100

    2024

  40. [48]

    Promptasr for contextualized asr with controllable style,

    X. Yang, W. Kang, Z. Yao, Y . Yang, L. Guo, F. Kuang, and et.al., “Promptasr for contextualized asr with controllable style,” inICASSP, 2024, pp. 10 536–10 540

    2024

  41. [49]

    Seaco- paraformer: A non-autoregressive asr system with flexible and effective hotword customization ability,

    X. Shi, Y . Yang, Z. Li, Y . Chen, Z. Gao, and S. Zhang, “Seaco- paraformer: A non-autoregressive asr system with flexible and effective hotword customization ability,” inICASSP, 2024, pp. 10 346–10 350

    2024

  42. [50]

    Mask CTC: non-autoregressive end-to-end ASR with CTC and mask predict,

    Y . Higuchi, S. Watanabe, N. Chen, T. Ogawa, and T. Kobayashi, “Mask CTC: non-autoregressive end-to-end ASR with CTC and mask predict,” inINTERSPEECH, 2020, pp. 3655–3659

    2020

  43. [51]

    CIF: continuous integrate-and-fire for end-to-end speech recognition,

    L. Dong and B. Xu, “CIF: continuous integrate-and-fire for end-to-end speech recognition,” inICASSP, 2020, pp. 6079–6083

    2020

  44. [52]

    Non- autoregressive transformer for speech recognition,

    N. Chen, S. Watanabe, J. Villalba, P. Zelasko, and N. Dehak, “Non- autoregressive transformer for speech recognition,”IEEE Signal Process. Lett., vol. 28, pp. 121–125, 2021

    2021

  45. [53]

    Non- autoregressive transformer ASR with ctc-enhanced decoder input,

    X. Song, Z. Wu, Y . Huang, C. Weng, D. Su, and H. M. Meng, “Non- autoregressive transformer ASR with ctc-enhanced decoder input,” in ICASSP, 2021, pp. 5894–5898

    2021

  46. [54]

    Relaxing the conditional independence assumption of ctc-based ASR by conditioning on intermediate predic- tions,

    J. Nozaki and T. Komatsu, “Relaxing the conditional independence assumption of ctc-based ASR by conditioning on intermediate predic- tions,” inINTERSPEECH, 2021, pp. 3735–3739

    2021

  47. [55]

    Paraformer: Fast and accurate parallel transformer for non-autoregressive end-to-end speech recognition,

    Z. Gao, S. Zhang, I. McLoughlin, and Z. Yan, “Paraformer: Fast and accurate parallel transformer for non-autoregressive end-to-end speech recognition,” inINTERSPEECH, 2022, pp. 2063–2067

    2022

  48. [56]

    Glancing transformer for non-autoregressive neural machine transla- tion,

    L. Qian, H. Zhou, Y . Bao, M. Wang, L. Qiu, W. Zhang, and et.al., “Glancing transformer for non-autoregressive neural machine transla- tion,” inACL/IJCNLP, 2021, pp. 1993–2003

    2021

  49. [57]

    Minimum word error rate training for attention-based sequence- to-sequence models,

    R. Prabhavalkar, T. N. Sainath, Y . Wu, P. Nguyen, Z. Chen, C. Chiu, and et.al., “Minimum word error rate training for attention-based sequence- to-sequence models,” inICASSP, 2018, pp. 4839–4843

    2018

  50. [58]

    Long short-term memory,

    S. Hochreiter and J. Schmidhuber, “Long short-term memory,”Neural Comput., vol. 9, no. 8, pp. 1735–1780, 1997

    1997

  51. [59]

    Focal loss for dense object detection,

    T. Lin, P. Goyal, R. B. Girshick, K. He, and P. Doll ´ar, “Focal loss for dense object detection,” inICCV, 2017, pp. 2999–3007

    2017

  52. [60]

    Supervised contrastive learning,

    P. Khosla, P. Teterwak, C. Wang, A. Sarna, Y . Tian, P. Isola, and et.al., “Supervised contrastive learning,” inNeurIPS, 2020

    2020

  53. [61]

    AISHELL- NER: named entity recognition from chinese speech,

    B. Chen, G. Xu, X. Wang, P. Xie, M. Zhang, and F. Huang, “AISHELL- NER: named entity recognition from chinese speech,” inICASSP, 2022, pp. 8352–8356

    2022

  54. [62]

    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, pp. 1–5

    2017

  55. [63]

    Kespeech: An open source speech dataset of mandarin and its eight subdialects,

    Z. Tang, D. Wang, Y . Xu, J. Sun, and et.al., “Kespeech: An open source speech dataset of mandarin and its eight subdialects,” inNeurIPS Datasets and Benchmarks, 2021

    2021

  56. [64]

    Funasr: A fundamental end-to-end speech recognition toolkit,

    Z. Gao, Z. Li, J. Wang, H. Luo, X. Shi, M. Chen, and et.al., “Funasr: A fundamental end-to-end speech recognition toolkit,” inINTERSPEECH, 2023, pp. 1593–1597

    2023

  57. [65]

    SAN-M: memory equipped self-attention for end-to-end speech recognition,

    Z. Gao, S. Zhang, M. Lei, and I. McLoughlin, “SAN-M: memory equipped self-attention for end-to-end speech recognition,” inINTER- SPEECH, 2020, pp. 6–10

    2020

  58. [66]

    Specaugment: A simple data augmentation method for automatic speech recognition,

    D. S. Park, W. Chan, Y . Zhang, C. Chiu, B. Zoph, E. D. Cubuk, and et.al., “Specaugment: A simple data augmentation method for automatic speech recognition,” inINTERSPEECH, 2019, pp. 2613–2617

    2019

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.