Pith. sign in

REVIEW 2 major objections 5 minor 35 references

COALA: Robust Contextualized Speech-augmented Language Modeling for ASR via Contrastive Regularizer and Biasing Score Estimation

T0 review · 2 major / 5 minor · reviewed 2026-07-10 · grok-4.5

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

Pith's one-line read COALA scores audio against candidate rare words so speech language models can bias ASR without overflowing their context windows.

desk verdict Solid engineering fix for multi-entity contextual biasing in SLMs: two clean losses that stop training collapse and make large lists usable without OOM. read the letter →

arxiv 2607.08117 v1 pith:AASXFFKM submitted 2026-07-09 cs.CL

Jhih-Rong Guo , Bi-Cheng Yan , Tien-Hong Lo , Berlin Chen This is my paper · ORCID
classification cs.CL
keywords contextualbiasingspeech-augmentedlanguagemodelsASRdiscriminativescoringmulti-positivelossLibriSpeechentityfiltering
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

Speech-augmented language models are good at general transcription but still miss rare, domain-specific words. Feeding every candidate name into the prompt fails once the list grows large: the context window fills, memory blows up, and distractors hurt accuracy. COALA trains a small scorer on top of the frozen model that maps speech and entity tokens into a shared score space, then keeps only the top matches for the final recognition prompt. The paper also replaces the usual multi-class ranking loss with two new objectives that stop positive rare words from fighting each other during training, so multi-entity utterances no longer collapse. On LibriSpeech the filtered approach recovers nearly all targets even from 5 000-item lists and sharply cuts biased word error while the unfiltered baseline runs out of memory.

What carries the argument

Biasing-score estimation (length-normalized MLP scores from SLM last-token states) optimized by Multi-Positive Discriminative Loss or Decoupled Point-wise Discriminative Loss, followed by top-K selection gated by the <unbiased> token.

What would settle it

Measure Recall#20 and B-WER on LibriSpeech (or a matched multi-entity set) when the inference list size exceeds 5 000 and when every utterance is forced to contain 12–24 rare words; if either metric collapses relative to the reported top-10 results, the filtering claim fails.

Watch

Extended reading notes

Core claim

A lightweight discriminative projector on SLM hidden states can rank rare-word candidates by acoustic match, and the new multi-positive and point-wise losses let that scorer train stably when several rare words appear in one utterance, turning large biasing lists into short, accurate prompts that improve ASR.

Load-bearing premise

A scorer trained on only 120 candidates per utterance and a hard top-10 cutoff will still surface the true rare words when the real list grows to thousands of items or an utterance contains more than ten rare words.

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

2 major / 5 minor

Summary. The paper proposes COALA, a two-stage framework that equips speech-augmented language models (SLMs) with a discriminative projector mapping SLM latents to entity-matching scores (Eq. 1–2). This enables biasing-target identification (BTI) that selects a short top-K subset from large candidate lists, circumventing SLM context-window limits. To stabilize training on multi-target utterances, the authors replace the global softmax discriminative loss of prior work with Multi-Positive Discriminative Loss (MPD-Loss, Eq. 4) and Decoupled Point-wise Discriminative Loss (DPD-Loss, Eqs. 5–7). On LibriSpeech rare-word biasing lists of size N = 500/1000/5000, DPD-Loss yields Recall#20 of 99.09 % / 96.59 % (clean/other) and reduces B-WER to 3.25/9.13 (N=500), 3.86/10.59 (N=1000) and 6.96/15.17 (N=5000) after BTI, outperforming CTC-Filter, K-Prompt and the unfiltered full-list baseline (which OOMs at N=5000).

Significance. If the reported gains hold under broader conditions, COALA supplies a practical, lightweight (777 M) recipe for injecting large entity catalogs into SLMs without OOM or severe distractor interference. The explicit diagnosis of inter-positive gradient conflict and the two new losses that remove the need for an auxiliary log term are useful contributions; Figure 4 and Table 1 give direct empirical support. Public code and a clear two-stage training protocol further raise the work’s utility for the community. The main caveats are a comparatively weak unbiased baseline and the fixed top-K protocol, yet the relative improvements over strong filter baselines remain informative for contextual ASR.

major comments (2)
  1. [Table 2] Table 2 shows that COALA’s unbiased B-WER (23.39 / 39.49) is substantially worse than the unbiased numbers of the cited baselines (Bias Qwen 8.40/18.40, RNN-T+IB 12.96/28.09, etc.). While relative gains after BTI are large, absolute B-WER comparisons across systems are therefore difficult to interpret; the paper should either match base-model strength or explicitly qualify the absolute numbers.
  2. [§3.4, Figure 3] §3.4 trains the scorer with only M=120 candidates per utterance and then applies a fixed top-K=10 (+ <unbiased> threshold) at inference for lists up to N=5000. Figure 3 and the accompanying text acknowledge that utterances with >11 rare words cannot be fully recovered. An ablation on training-list size M and an adaptive-K or recall-oriented selection policy would strengthen the claim of robustness for the multi-entity regime shown in Figure 1.
minor comments (5)
  1. [Title / Abstract] Title and abstract advertise a “Contrastive Regularizer,” yet the body never uses that term; MPD-Loss and DPD-Loss are presented simply as refined discriminative objectives. Align terminology.
  2. [Figure 2] Figure 2 caption and panel (b) contain typos (“scroing”, “Entity sample and scroing”).
  3. [Title / Abstract] Acronym expansion of COALA differs slightly between title (“COntextualizedASRLeveraging biAsing scoring”) and abstract; standardize.
  4. [Tables 1–2] No error bars or statistical significance tests accompany the WER/recall figures; even a simple bootstrap over utterances would help.
  5. [§2.1] The special token is written both as <unbiased> and “unbiased”; keep notation consistent.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: empirical ML paper with standard ranking/binary losses evaluated on external WER/recall metrics; no derivation reduces to inputs by construction.

full rationale

COALA is a standard empirical ASR paper. The core contributions are (1) an MLP projector that maps SLM last-hidden states to scalar matching scores (Eqs. 1–2) and (2) two new training objectives (MPD-Loss Eq. 4, DPD-Loss Eqs. 5–7) that remove inter-positive competition present in the prior discriminative loss (Eq. 3). Both losses are ordinary ranking / point-wise binary-classification objectives; their gradients do not algebraically force the reported Recall#20 or B-WER numbers. Evaluation uses the external LibriSpeech rare-word protocol of Le et al. (2021) with independently constructed biasing lists of size N ∈ {500,1000,5000}; metrics (U-WER, B-WER, Recall@X, Recall#X) are computed from held-out transcripts and are not functions of the training loss. The only internal loop is ordinary train-then-test on the same corpus, which is not circularity under the stated criteria. No self-citation is load-bearing, no uniqueness theorem is imported, no fitted constant is renamed a prediction, and no ansatz is smuggled via citation. The acknowledged top-K=10 limitation for utterances with >10 entities is an explicit experimental caveat, not a circular reduction. Hence score 0 with empty steps.

Assumptions & free parameters 5 free parameters · 4 assumptions · 3 invented entities

The central claim rests on standard ASR modeling assumptions plus a handful of engineering choices (top-K, candidate count at train time, rare-word definition inherited from prior work). No new physical entities are postulated; the ‘invented’ items are the two loss functions and the discriminative projector, which are algorithmic constructs with direct experimental handles.

free parameters (5)
  • top-K selection size = 10
    Fixed at K=10 for the ASR prompt; directly determines how many entities reach the decoder and is not derived from data.
  • training candidate list size M = 120
    Set to 120 entities per utterance in Stage 2; chosen for GPU memory, not cross-validated against final N=5000 performance.
  • CTC / CE loss weights (Stage 1) = 0.3 / 1.0
    Hand-set to 0.3 and 1.0; affect the quality of the frozen audio tokens that the scorer later reads.
  • auxiliary log-loss weight for Bias-Loss baseline = 0.1
    Weight 0.1 required for the prior discriminative loss to converge; our proposed losses avoid it, but the comparison still depends on this hyper-parameter.
  • MLP hidden dimension D′ of discriminative projector
    Architecture hyper-parameter of f(·); value not stated, yet the entire scoring space depends on it.
assumptions (4)
  • domain assumption Rare vs common word partition of LibriSpeech vocabulary (5K common, 209.2K rare) is a valid proxy for domain-specific entities.
    Inherited from Le et al. 2021 and used to construct every biasing list; if the partition is unrepresentative of real personalization catalogs the reported gains may not transfer.
  • ad hoc to paper Length-normalized average of token-level projector outputs is a sufficient sequence-level match score (Eq. 1).
    No theoretical justification that mean pooling is optimal; alternatives (max, attention pooling) are not ablated.
  • domain assumption Freezing the Stage-1 ASR parameters and training only a new LoRA + projector preserves acoustic grounding needed for scoring.
    Stated in §3.4; if the frozen representations are poorly aligned the scorer cannot recover.
  • standard math Softmax / sigmoid ranking objectives with the stated positive/negative sampling correctly identify entities that improve downstream ASR.
    Standard contrastive / binary classification math; used without modification beyond the multi-positive and point-wise rewrites.
invented entities (3)
  • Multi-Positive Discriminative Loss (MPD-Loss) independent evidence
    purpose: Remove inter-positive competition by localizing softmax to each positive vs all negatives.
    New objective introduced in Eq. (4); independent evidence is the improved multi-entity score distributions in Figure 4(b) and the Recall numbers in Table 1.
  • Decoupled Point-wise Discriminative Loss (DPD-Loss) independent evidence
    purpose: Treat each entity as an independent binary classification so absolute scores form a stable zero-centered decision boundary.
    New objective in Eqs. (5)–(7); independent evidence is the clean positive/negative separation in Figure 4(c) and best Recall#20/50.
  • Discriminative projector f(·)
    purpose: Map SLM last-hidden states into a scalar match intensity independent of the LM vocabulary distribution.
    Two-layer MLP defined in Eq. (2); its utility is measured only inside this paper’s ranking metrics, but the architecture is standard and falsifiable by ablation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of COALA: Robust Contextualized Speech-augmented Language Modeling for ASR via Contrastive Regularizer and Biasing Score Estimation." pith.science (2026). https://pith.science/paper/AASXFFKM

@misc{pith2026260708117,
  author       = {Pith},
  title        = {Pith review of: COALA: Robust Contextualized Speech-augmented Language Modeling for ASR via Contrastive Regularizer and Biasing Score Estimation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AASXFFKM}},
  note         = {Machine review of arXiv:2607.08117}
}
read the original abstract

Contextual biasing seeks to integrate external knowledge into automatic speech recognition (ASR) systems to accurately recognize domain-specific entities. In this paper, we propose COALA (Contextualized ASR Leveraging Biasing Scoring), a robust framework designed to enhance speech-augmented language models (SLMs) in complex multi-entity scenarios. Considering the inherent context-window limitations of SLMs, identifying relevant target entities from a large-scale biasing list is crucial for effective recognition. To this end, COALA maps SLM latent representations into a specialized discriminative space to quantify the matching intensity between audio segments and candidate entities. Furthermore, we address the training collapse in prior study when handling multi-target utterances-where multiple rare words co-occur. Experimental results on the LibriSpeech benchmark demonstrate that COALA consistently achieves superior contextual biasing performance across various biasing list scales.

Figures

Figures reproduced from arXiv: 2607.08117 by the authors.

Figure 1
Figure 1. Distribution of utterances by number of target entities on the LibriSpeech corpus. space and dynamically boost the emission probability of target entities during the decoding stage, often triggered by specific prefixes (e.g., ’call’ or ’play’). On a separate front, training￾time biasing methods, such as attention-based biasing adapters [14, 15] and trie-based pointer generators [16, 17], which are trained to align a… view at source ↗
Figure 2
Figure 2. (a) The proposed COALA framework for biasing scoring, featuring a frozen backbone with trainable LoRA modules and a discriminative projector. (b) The sampling and scoring process where positive (E +) and negative (E −) entities are assigned matching intensities scores S. (c) Comparison of objective functions: discriminative loss exhibits inter-positive competition, while multi-positive discriminative loss (MPD-Loss)… view at source ↗
Figure 3
Figure 3. B-WER of COALA-based ASR using biasing prompts from various filter methods on testsets across different numbers of target enti￾ties. The COALA is optimized via DPD-Loss. (a) Bias-Loss (b) MPD-Loss (c) DPD-Loss [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

35 extracted references · 35 canonical work pages

  1. [1]

    Introduction Driven by the monolithic nature and streamlined training pro- cess, end-to-end (E2E) automatic speech recognition (ASR) systems [1, 2, 3, 4] have gained widespread attention across both academia and industry. More recently, as witnessed by the remarkable success of large language models (LLMs) in the nat- ural language processing community, e...

    work page Pith review arXiv 2026

  2. [2]

    Methods 2.1. Biasing Scoring As illustrated in Figure 2 (a), we define the acoustic features extracted from the input audio via the audio encoder block as X, and the corresponding ground-truth transcription asY. For each utterance, a biasing listE={e 0, e1, e2, . . . , eM }is con- structed, containingM+ 1entities. Within this list,e 0 is a special token, ...

    work page

  3. [3]

    Datasets We evaluate our proposed method on the LibriSpeech corpus [18], following the contextual biasing experimental setup es- tablished in prior study [19]

    Experiments 3.1. Datasets We evaluate our proposed method on the LibriSpeech corpus [18], following the contextual biasing experimental setup es- tablished in prior study [19]. In this setup, 5K high-frequency terms are identified as common words, while the remaining 209.2K lower-frequency terms are categorized as rare words. For each utterance, we constr...

    work page

  4. [4]

    The Performance of Biasing Scoring Table 1 compares the performance of biasing scoring across dif- ferent objectives

    Results 4.1. The Performance of Biasing Scoring Table 1 compares the performance of biasing scoring across dif- ferent objectives. Unlike the baseline Bias-Loss, which must be implemented as a joint objective (discriminative and log loss)

    work page

  5. [5]

    Notably, DPD-Loss achieves a Recall#20 of 99.09% on test-clean, signifying that target entities are al- most invariably ranked within the top-20 candidates

    to prevent training collapse on multi-target data, our pro- posed MPD-Loss and DPD-Loss converge effectively as stan- dalone objectives. Notably, DPD-Loss achieves a Recall#20 of 99.09% on test-clean, signifying that target entities are al- most invariably ranked within the top-20 candidates. In con- trast, prompt-based baselines like CTC-Filter and K-Pro...

    work page

  6. [6]

    The COALA is optimized via DPD-Loss

    + BPB [26] 14.10 (5.05 / 3.90) 27.90 (8.81 / 6.60) - - 7.00 (3.21 / 2.70) 13.50 (6.28 / 5.50) 7.70 (3.47 / 3.00) 15.80 (7.34 / 6.40) - - COALA 23.39 (4.54 / 2.34) 39.49 (8.75 / 5.48) 1.77 (1.90 / 1.92) 6.12 (4.86 / 4.73) 20.38 (4.36 / 2.49) 35.01 (8.87 / 6.09) 21.98 (4.56 / 2.53) 37.54 (9.18 / 6.16) OOM OOM + BTI (DPD-Loss)23.39 (4.54 / 2.34) 39.49 (8.75 ...

    work page

  7. [7]

    Conclusion In this paper, we presented COALA, a robust contextual bi- asing framework tailored for SLMs. To address the limita- tions of traditional discriminative losses in multi-entity scenar- ios—namely, the gradient conflicts that hinder the optimiza- tion of multiple positive targets—we introduced MPD-Loss and DPD-Loss. Our analysis confirms that by ...

    work page

  8. [8]

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

    Acknowledgments This work was supported in part by Realtek Semiconductor Cor- poration under Grant Numbers 113KK01103 and 114KK01005. Any findings and implications in the paper do not necessarily reflect those of the sponsors

    work page

Show all 35 references
  1. [9]

    We maintained full control over the re- search content and remain responsible for the accuracy and in- tegrity of the experimental results and the presented figures

    Generative AI Use Disclosure The authors utilized Gemini-3.1-Pro to refine the wording clar- ity of this manuscript. We maintained full control over the re- search content and remain responsible for the accuracy and in- tegrity of the experimental results and the presented figures

  2. [10]

    Con- nectionist temporal classification: labelling unsegmented se- quence data with recurrent neural networks,

    A. Graves, S. Fern ´andez, F. Gomez, and J. Schmidhuber, “Con- nectionist temporal classification: labelling unsegmented se- quence data with recurrent neural networks,” inProc. ICML, 2006

    2006

  3. [11]

    Sequence transduction with recurrent neural net- works,

    A. Graves, “Sequence transduction with recurrent neural net- works,”arXiv preprint arXiv:1211.3711, 2012

    2012 arXiv

  4. [12]

    Neural machine translation by jointly learning to align and translate,

    D. Bahdanau, K. Cho, and Y . Bengio, “Neural machine translation by jointly learning to align and translate,” inProc. ICLR, 2015

    2015

  5. [13]

    Hybrid transducer and attention based encoder-decoder modeling for speech-to-text tasks,

    Y . Tang, A. Sun, H. Inaguma, X. Chen, N. Dong, X. Ma, P. Tomasello, and J. Pino, “Hybrid transducer and attention based encoder-decoder modeling for speech-to-text tasks,” inProceed- ings of the Annual Meeting of the Association for Computational Linguistics (Volume 1: Long P...

    2023

  6. [14]

    Qwen-audio: Advancing universal audio understand- ing via unified large-scale audio-language models,

    Y . Chu, J. Xu, X. Zhou, Q. Yang, S. Zhang, Z. Yan, C. Zhou, and J. Zhou, “Qwen-audio: Advancing universal audio understand- ing via unified large-scale audio-language models,”arXiv preprint arXiv:2311.07919, 2023

    2023 arXiv

  7. [15]

    Salmonn: Towards generic hearing abilities for large language models,

    C. Tang, W. Yu, G. Sun, X. Chen, T. Tan, W. Li, L. Lu, Z. Ma, and C. Zhang, “Salmonn: Towards generic hearing abilities for large language models,”arXiv preprint arXiv:2310.13289, 2023

    2023 arXiv

  8. [16]

    Qwen2-audio technical report,

    Y . Chu, J. Xu, Q. Yang, H. Wei, X. Wei, Z. Guo, Y . Leng, Y . Lv, J. He, J. Linet al., “Qwen2-audio technical report,”arXiv preprint arXiv:2407.10759, 2024

    2024 arXiv

  9. [17]

    Prepending or cross-attention for speech-to-text? an empirical comparison,

    T. K. Lam, M. Gaido, S. Papi, L. Bentivogli, and B. Haddow, “Prepending or cross-attention for speech-to-text? an empirical comparison,” inProceedings of the 2025 Conference of the Na- tions of the Americas Chapter of the Association for Computa- tional Linguistics: Human Lang...

    2025

  10. [18]

    Wavllm: Towards robust and adap- tive speech large language model,

    S. Hu, L. Zhou, S. Liu, S. Chen, L. Meng, H. Hao, J. Pan, X. Liu, J. Li, S. Sivasankaranet al., “Wavllm: Towards robust and adap- tive speech large language model,” inFindings of the Association for Computational Linguistics: EMNLP 2024, 2024, pp. 4552– 4572

    2024

  11. [19]

    Weighted finite-state trans- ducers in speech recognition,

    M. Mohri, F. Pereira, and M. Riley, “Weighted finite-state trans- ducers in speech recognition,”Computer Speech & Language, 2002

    2002

  12. [20]

    Shallow-fusion end-to-end contextual biasing

    D. Zhao, T. N. Sainath, D. Rybach, P. Rondon, D. Bhatia, B. Li, and R. Pang, “Shallow-fusion end-to-end contextual biasing.” in Proc. Interspeech, 2019

    2019

  13. [21]

    Contextual speech recognition in end-to-end neural net- work 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 net- work systems using beam search.” inProc. Interspeech, 2018

    2018

  14. [22]

    A neural model for contextual biasing score learning and filtering,

    W. Huang and W. Wang, “A neural model for contextual biasing score learning and filtering,” inProc. ASRU, 2025

    2025

  15. [23]

    Contextual adapters for personalized speech recognition in neural transducers,

    K. M. Sathyendra, T. Muniyappa, F.-J. Chang, J. Liu, J. Su, G. P. Strimel, A. Mouchtaris, and S. Kunzmann, “Contextual adapters for personalized speech recognition in neural transducers,” in ICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal Process...

    2022

  16. [24]

    Retrieve and copy: Scaling asr personalization to large catalogs,

    S. M. Jayanthi, D. Kulshreshtha, S. Dingliwal, S. Ronanki, and S. Bodapati, “Retrieve and copy: Scaling asr personalization to large catalogs,” inProceedings of the 2023 Conference on Em- pirical Methods in Natural Language Processing: Industry Track, 2023, pp. 631–639

    2023

  17. [25]

    Minimising biasing word errors for contextual asr with the tree-constrained pointer gener- ator,

    G. Sun, C. Zhang, and P. C. Woodland, “Minimising biasing word errors for contextual asr with the tree-constrained pointer gener- ator,”IEEE/ACM Transactions on Audio, Speech, and Language Processing, vol. 31, pp. 345–354, 2022

    2022

  18. [26]

    Graph neural networks for contextual asr with the tree-constrained pointer generator,

    G. Sun, C. Zhang, and P. C. Woodland, “Graph neural networks for contextual asr with the tree-constrained pointer generator,” IEEE/ACM Transactions on Audio, Speech, and Language Pro- cessing, 2024

    2024

  19. [27]

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

    V . Panayotov, G. Chen, D. Povey, and S. Khudanpur, “Lib- rispeech: An asr corpus based on public domain audio books,” inProc. ICASSP, 2015

    2015

  20. [28]

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

    D. Le, M. Jain, G. Keren, S. Kim, Y . Shi, J. Mahadeokar, J. Chan, Y . Shangguan, C. Fuegen, O. Kalinliet al., “Contextualized streaming end-to-end speech recognition with trie-based deep bi- asing and shallow fusion,” inProc. Interspeech, 2021

    2021

  21. [29]

    Ctc-assisted llm-based contextual asr,

    G. Yang, Z. Ma, Z. Gao, S. Zhang, and X. Chen, “Ctc-assisted llm-based contextual asr,” inProc. SLT, 2024

    2024

  22. [30]

    Knowledge prompt for whisper: An asr entity correction approach with knowledge base,

    M. Zhang, X. Qiao, Y . Zhao, C. Su, Y . Li, Y . Li, M. Zhu, M. Piao, S. Peng, S. Tao, H. Yang, and Y . Jiang, “Knowledge prompt for whisper: An asr entity correction approach with knowledge base,” inProc. BigData, 2023

    2023

  23. [31]

    Robust speech recognition via large-scale weak su- pervision,

    A. Radford, J. W. Kim, T. Xu, G. Brockman, C. McLeavey, and I. Sutskever, “Robust speech recognition via large-scale weak su- pervision,” inProc. ICML, 2023

    2023

  24. [32]

    Contextual biasing speech recognition in speech-enhanced large language model,

    X. Gong, A. Lv, Z. Wang, and Y . Qian, “Contextual biasing speech recognition in speech-enhanced large language model,” in Proc. Interspeech, 2024

    2024

  25. [33]

    Contextualized end-to-end automatic speech recognition with intermediate bias- ing loss,

    M. Shakeel, Y . Sudo, Y . Peng, and S. Watanabe, “Contextualized end-to-end automatic speech recognition with intermediate bias- ing loss,” inProc. Interspeech, 2024

    2024

  26. [34]

    Hy- brid ctc/attention architecture for end-to-end speech recognition,

    S. Watanabe, T. Hori, S. Kim, J. R. Hershey, and T. Hayashi, “Hy- brid ctc/attention architecture for end-to-end speech recognition,” IEEE Journal of Selected Topics in Signal Processing, 2017

    2017

  27. [35]

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

    Y . Sudo, M. Shakeel, Y . Fukumoto, Y . Peng, and S. Watan- abe, “Contextualized automatic speech recognition with attention- based bias phrase boosted beam search,” inProc. ICASSP, 2024

    2024

Pith tools

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