WCTC-Biasing: Retraining-free Contextual Biasing ASR with Wildcard CTC-based Keyword Spotting and Inter-layer Biasing

Michael Hentschel; Yu Nakagome

REVIEW 4 major objections 6 minor 43 references

WCTC-Biasing: Retraining-free Contextual Biasing ASR with Wildcard CTC-based Keyword Spotting and Inter-layer Biasing

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A retraining-free method improves rare-word recognition in CTC-based ASR by spotting keywords with wildcard CTC at intermediate layers and biasing subsequent layers.

desk verdict Useful retraining-free biasing idea, but oracle-built keyword lists make the reported F1 gains an unmeasured upper bound. read the letter →

arxiv 2506.01263 v1 pith:D7Q4XGHC submitted 2025-06-02 cs.CL cs.SDeess.AS

Yu Nakagome , Michael Hentschel This is my paper

classification cs.CLcs.SDeess.AS

keywords automaticspeechrecognitioncontextualbiasingconnectionisttemporalclassificationwildcardCTCkeywordspottingself-conditionedJapaneserareword

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

The paper proposes WCTC-Biasing, a way to make CTC-based speech recognizers output rare words they were never trained on, without retraining the model or generating TTS audio. During inference, the method runs a wildcard-CTC keyword spotter on an intermediate encoder layer, treats every non-keyword frame as a wildcard, and when a target keyword is detected it injects that keyword into the conditioning signal of the subsequent encoder layers. In Japanese ASR experiments across one in-domain and three out-of-domain test sets, the method improved F1 for out-of-vocabulary and in-vocabulary keywords compared with the self-conditioned CTC baseline, with the largest gain a 29% relative F1 improvement for unknown words under keyword-boosted beam search, while character error rate stayed essentially unchanged. The appeal is practical: the approach works on existing large-scale CTC models at inference time, avoiding the training cost and TTS dependency of prior inter-layer biasing methods.

What carries the argument

The load-bearing object is the wildcard-CTC path computed at an intermediate encoder layer. A wildcard token '*' is added to the CTC vocabulary and can match any symbol, including blank, at zero cost, which lets the path ignore unlabeled frames; the method sums over all wildcard paths consistent with a target keyword and applies a threshold $\theta$ to the log-probability to decide detection. On detection, a one-hot bias vector for the keyword is interpolated with the intermediate softmax prediction $Z^{(n)}$ via bias weight $\omega$, then linearly projected to encoder dimension and added to the next layer's input, exactly as the self-conditioned CTC framework injects intermediate predictions. This mechanism carries the argument because it is what turns a spotted keyword into a bias on the subsequent layers, causing the keyword to appear more often among the top hypotheses during beam search.

What would settle it

Decode a test set with a keyword list where most entries never occur in the audio, and compare OOV F1 and character error rate against the no-bias baseline; if the F1 gains shrink or reverse, or the CER rises, the claim that the method improves rare-word recognition without hurting overall accuracy is falsified.

Watch

Extended reading notes

Core claim

The central claim is that the recognition of unknown and rare words in CTC-based ASR can be improved by biasing intermediate encoder predictions toward a keyword list, and that wildcard CTC on an intermediate layer's acoustic features can detect those keywords reliably enough for the bias to help. Wildcard CTC augments the ordinary CTC vocabulary with a '*' token that can match any symbol, including blank, at zero cost, so the alignment path can skip non-keyword frames and focus only on whether the target keyword appears in the encoder output. When the log-probability of a wildcard-CTC path for a keyword exceeds a threshold $\theta$ at layer $n$, the method replaces that layer's prediction with a one-hot vector for the keyword, interpolates it with the original softmax output using a bias weight $\omega$, maps the result back to the encoder dimension, and adds it to the input of the next encoder layer. This pushes the target keyword into the beam-search hypotheses, which is exactly what keyword-boosted beam search needs to act on. In the reported Japanese ASR experiments, this yields a 29% relative improvement in OOV F1 on the TEDxJP-10K test set compared with the self-conditioned CTC baseline, and it matches or beats the TTS-dependent InterBiasing approach on two of the four test sets.

Load-bearing premise

The load-bearing premise is that the keyword list is oracle-like: it is built by decoding each test set, comparing hypotheses to reference labels, and keeping exactly the words the model misrecognized, so the wildcard spotter is never tested on lists containing many words absent from the audio.

Editorial extensions

If this is right

Any existing CTC-based ASR model with intermediate outputs can adopt the biasing without retraining, so the method can be dropped onto deployed large-scale models at inference time.
Because the F1 gains for OOV and IV keywords come with essentially unchanged character error rates, the biasing does not measurably hurt recognition of non-keyword content.
The wildcard token's tolerance for ambiguous matches allows keyword detection even at lower encoder layers where intermediate predictions are unreliable, widening the range of layers available for inter-layer biasing.
The method combines with standard decoding add-ons: under LM shallow fusion with keyword-boosted beam search, the biasing makes target keywords surface among beam hypotheses, yielding the largest reported F1 improvements.
Out-of-domain evaluation on Common Voice, JSUT basic 5000, and TEDxJP-10K shows consistent F1 gains over the baseline, which suggests the approach transfers beyond the training domain's acoustic and lexical conditions.

Reading between the lines

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

A natural stress test the paper does not run is a realistic keyword list containing many entries that never occur in the audio; the wildcard spotter's false-positive rate and its effect on character error rate would determine whether the reported F1 gains survive production conditions.
The threshold $\theta$ was tuned low ($-40$) to detect more candidate keywords; adapting it per domain or per keyword length could trade detection recall against biasing noise and might further improve out-of-domain F1.
Because wildcard-CTC paths are computed per keyword, inference cost scales with the keyword-list size; for very large lists, pruning candidates by acoustic similarity before running the spotter would be a natural next step.
The spotter-and-bias mechanism is not tied to the specific 18-layer encoder used here; any CTC-family encoder with intermediate outputs could host the same idea, including streaming models, provided their intermediate layers are supervised.

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.

Desk Editor's Note

read the letter

First thing to know: this is a practical simplification of InterBiasing, replacing the TTS-based trigger collection with inference-time wildcard CTC keyword spotting on intermediate encoder layers. That is a real and sensible idea — it removes the operational cost of a TTS module and a retraining step, which matters for deployed CTC models. The paper is clearly written, the method is easy to follow, and the F1 gains over the SelfCond baseline are consistent across four test sets, including three out-of-domain ones. The comparison with InterBiasing is honest: they report where their simpler method wins and loses.

The load-bearing weakness is the oracle keyword construction in Section 4.1. The bias list for each test set is built by decoding that test set with the trained model and keeping exactly the proper nouns the model misrecognized. So every keyword in K is guaranteed to occur in the audio and to be difficult for the baseline. The central risk of a real biasing system — a large keyword list full of distractors that are not in the utterance, causing the spotter to false-positive and the bias to distort the internal representations — is excluded by construction. At theta = -40, Eq. 14 will fire on many absent words, and Eq. 16 will push their one-hot posteriors into every third layer. The paper's claim that CER remains unchanged is not evidence against this; the oracle list removes exactly the spurious-bias errors that would show up in CER. So the headline 29% OOV F1 improvement is an upper bound, not a measured property under realistic conditions.

There are smaller issues too: no error bars or significance tests; the hyperparameters (theta, omega, layer cadence, KBBS weight) are given without ablations; and Eq. 14 writes the wildcard CTC path probability as a log of a sum that is not a proper log-marginal, though that is minor. The paper does not ship code or data, which makes the missing ablations harder to forgive but not fatal.

Who should read it: anyone doing contextual biasing for CTC ASR, especially in production settings where retraining and TTS are expensive. The core idea is worth knowing, and the oracle issue is fixable: evaluate with a realistic keyword list that includes distractor words, report spotter precision, and add a small ablation on the threshold. I would accept it for peer review — a good referee could pull it into shape — but the evaluation needs substantial revision before the numbers can be trusted.

Referee Report

4 major / 6 minor

Summary. The paper proposes WCTC-Biasing, a retraining-free contextual biasing method for CTC-based ASR. During inference, wildcard CTC is applied to an intermediate encoder layer to spot keywords from a supplied list, and detected keywords are converted into one-hot bias features that are injected into subsequent encoder layers within the self-conditioned CTC framework. Experiments on four Japanese test sets compare the method with SelfCond, keyword-boosted beam search, and TTS-based InterBiasing, reporting OOV F1 improvements, including a 29% relative improvement on TEDxJP-10K, while CER remains largely unchanged.

Significance. If the empirical claims hold, the method is practically attractive: it requires no retraining, no TTS module, and no WFST graph reconstruction, and it works with both greedy and shallow-fusion beam-search decoding. The core idea of using wildcard CTC as a fast keyword spotter at intermediate layers is novel and plausible, and the comparison against InterBiasing and KBBS is directly relevant. The manuscript does not provide code or per-keyword results, and the evaluation has a load-bearing oracle-like keyword construction, missing sensitivity analysis, and no statistical error bars; consequently the central quantitative claim is not yet established.

major comments (4)

[Sec. 4.1] The bias keyword list for each test set is constructed by decoding that same test set with the baseline model and keeping only proper nouns/personal names that are misrecognized relative to the reference. This makes the keyword set oracle-like: every keyword is known to occur in the audio and to be difficult for the baseline. The main practical risk of the method is the opposite case, namely false positives from a realistic keyword list containing many words that are absent from the utterance, because Eq. (16) injects a one-hot bias for every detected keyword into selected layers. In the current evaluation this false-positive regime is almost entirely absent, so the reported F1 gains are upper bounds rather than estimates under realistic contextual-biasing conditions. The unchanged CER in Table 2 does not mitigate this concern, since the oracle lists exclude exactly the spurious detections that would degrade CER. Please re-run the evaluation with a keyword list that is not derived from test-set errors, e.g., a fixed list drawn from the training corpus or an external vocabulary including distractor words, and report OOV F1, IV F1, and CER for that setting.
[Sec. 4.2 / Table 2] The method introduces several hyperparameters that are set without supporting ablations: the wildcard threshold θ=-40, the bias weight ω=0.7, the biased layer set S={3,6,...,15}, the KBBS weight 3.0, and the LM weight/length penalty. The only sensitivity discussion is the statement in Sec. 4.3 that a relatively low θ gave better results on the validation set. Since θ directly controls the false-positive rate that the oracle evaluation hides, the robustness of the claimed 29% improvement to θ and ω (and to the layer cadence) must be quantified. Please add ablations or a sensitivity table for these parameters.
[Eq. (14), Sec. 3.1] Equation (14) defines the keyword detection condition as log P_{π*∈B^{-1}(κ)} p(π_t | z_t^{(n)}) > θ, but p(π_t | z_t) is a frame-level posterior, not a path probability, and no summation over alignment paths is shown. For wildcard CTC, the relevant quantity is the collapsed path probability, i.e., the forward-backward sum over alignments π* whose collapsed form is the keyword κ, with the wildcard token assigned probability 1 at the frames it covers. As written, the expression is not a well-defined probability and cannot be implemented as stated. Please rewrite Eq. (14) with the correct path sum and specify how the forward-backward algorithm is applied at layer n.
[Table 2 / Sec. 4.3] No error bars, confidence intervals, or significance tests are reported, and the F1 scores are computed on small keyword sets (Table 1 lists 23–212 OOV keywords). For example, on Common Voice the OOV F1 difference between WCTC-Biasing and SelfCond with LM+BS is 22.7 vs 18.6, which could correspond to a small number of keyword instances. Please report bootstrap confidence intervals or per-utterance significance tests so that the reader can assess whether the claimed improvements are stable.

minor comments (6)

[Eq. (15), Sec. 3.2] Equation (15) uses a wedge symbol over κ∈K without a time index or an explicit definition of how per-frame detections for different keywords are aggregated; please clarify whether the aggregation is a max, a sum, or a logical OR over the one-hot vectors.
[Abstract / Table 2] The abstract and conclusion state a 29% improvement in F1 for unknown words without qualification; Table 2 shows 29% only for OOV F1 on TEDxJP-10K with LM+KBBS decoding, with smaller gains elsewhere. Please state the exact condition in the abstract or weaken the claim.
[Sec. 4.1] The authors note that CSJ eval1 and eval2 were excluded because not enough OOV keywords could be collected; this test-set selection should be stated in the data description and its effect on the generality of the results discussed.
[Sec. 2.3 / Sec. 3.2] Equations (11) and (16) contain an extra closing parenthesis after the Softmax argument; the notation should be cleaned up.
[References] References [9] and [13] are duplicated entries for the same paper; one should be removed or the citation numbers adjusted.
[Figure 2] The caption and axis labels in Figure 2 appear garbled in the manuscript (non-Japanese glyph artifacts); they should be regenerated or provided as clean text.

Circularity Check

1 steps flagged · score 6.0 of 10

Oracle keyword selection: the bias list is built from the baseline's own test-set errors, so reported F1 gains are upper bounds.

fitted input called prediction [Section 4.1 (Data; bias keyword selection) and Table 2 / Abstract (claimed F1 gains)]
"First, we decoded each evaluation set using the CSJ-trained model. By comparing the resulting hypotheses with the corresponding reference labels, we identified misrecognized words. These words were segmented using morphological analysis with MeCab [34], and we retained only proper nouns and personal names consisting of two or more characters based on morphological labels."

The bias keyword list K is constructed by decoding each evaluation set with the baseline model and keeping exactly the words that model misrecognized. The reported OOV/IV F1 scores in Table 2 are then computed over the same K, so the experiment measures the method's ability to correct the baseline's own known errors on each test set. Realistic contextual-biasing lists contain many keywords absent from a given utterance; with the low detection threshold theta=-40 used in Sec. 4.2, Eqs. 14-16 would produce spurious one-hot biases for such distractors, and the resulting CER/F1 damage is never measured because K excludes absent words by construction. The central 29% OOV F1 improvement is therefore an upper bound, not a measured property under realistic keyword conditions.

full rationale

The proposed WCTC-Biasing mechanism itself is not circular: wildcard-CTC keyword spotting (Eqs. 13-14) and inter-layer biasing (Eqs. 15-17) are legitimate inference-time operations built on external prior work (wildcard CTC [22], self-conditioned CTC [21]), with no parameter fit that is renamed as a prediction. The circularity is in the evaluation protocol. Section 4.1 derives the bias list from the baseline's errors on each test set, so the F1 metric is computed over exactly the failures the method is designed to fix, while excluding the false-positive regime that would arise with realistic keyword lists containing absent or easy words. This makes the 29% OOV F1 gain an oracle-conditioned upper bound rather than a fair estimate. There is no load-bearing self-citation chain or definitional equivalence in the equations, but the central empirical claim is partly an artifact of the keyword-selection construction.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new entities beyond the inference-time wildcard-CTC mechanism, which is built from the existing wildcard token of [22]. The main hidden assumptions are that a model not trained with wildcard tokens can be probed with them, and that the oracle keyword lists are a fair proxy for real biasing lists.

free parameters (5)

theta (keyword detection threshold) = -40
Controls whether a keyword's wildcard-CTC log probability counts as detected; set low by tuning, no sensitivity analysis.
omega (bias interpolation weight) = 0.7
Interpolation weight between original softmax and bias one-hot features; tuned.
S (biased layer set) = every 3 layers from 3 to 15
Hand-chosen cadence for applying biasing; not ablated.
KBBS boost weight = 3.0
Decoding hyperparameter for keyword-boosted beam search, tuned on dev; affects all methods equally.
LM weight / length penalty = 0.5 / 0.2
Shallow-fusion decoding hyperparameters tuned on CSJ dev; affects all methods.

assumptions (4)

standard math CTC and self-conditioned CTC as defined in [1,21] provide valid posteriors for inference-time manipulation.
The method relies on pre-trained intermediate posteriors Z^(n) being meaningful for keyword spotting (Sec. 2, Eqs. 3-8).
ad hoc to paper Wildcard CTC paths over a model trained without a wildcard token yield useful keyword detection scores.
The wildcard '*' is introduced only at inference (Eqs. 13-14); the model was never trained to emit or tolerate wildcard frames, so the posterior product assumption is untested.
ad hoc to paper Injecting a one-hot keyword vector at every frame of selected layers can steer the encoder toward that keyword without retraining.
Eq. 14 sets Z_bias,kappa,t = OneHot_kappa for all t when a keyword is detected anywhere, and Eqs. 16-17 feed this through a pre-trained linear layer; no temporal localization is used.
domain assumption The test-derived oracle keyword lists represent realistic contextual biasing conditions.
Sec. 4.1 constructs K by decoding each test set and keeping misrecognized proper nouns; this assumes such a list would be available in deployment and that distractors would not hurt.

how reviews work

0 comments

Cite this review

Pith. "Pith review of WCTC-Biasing: Retraining-free Contextual Biasing ASR with Wildcard CTC-based Keyword Spotting and Inter-layer Biasing." pith.science (2026). https://pith.science/paper/D7Q4XGHC

@misc{pith2026250601263,
  author       = {Pith},
  title        = {Pith review of: WCTC-Biasing: Retraining-free Contextual Biasing ASR with Wildcard CTC-based Keyword Spotting and Inter-layer Biasing},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/D7Q4XGHC}},
  note         = {Machine review of arXiv:2506.01263}
}

read the original abstract

Despite recent advances in end-to-end speech recognition methods, the output tends to be biased to the training data's vocabulary, resulting in inaccurate recognition of proper nouns and other unknown terms. To address this issue, we propose a method to improve recognition accuracy of such rare words in CTC-based models without additional training or text-to-speech systems. Specifically, keyword spotting is performed using acoustic features of intermediate layers during inference, and a bias is applied to the subsequent layers of the acoustic model for detected keywords. For keyword detection, we adopt a wildcard CTC that is both fast and tolerant of ambiguous matches, allowing flexible handling of words that are difficult to match strictly. Since this method does not require retraining of existing models, it can be easily applied to even large-scale models. In experiments on Japanese speech recognition, the proposed method achieved a 29% improvement in the F1 score for unknown words.

Figures

Figures reproduced from arXiv: 2506.01263 by the authors.

**Figure 1.** Overview of the proposed WCTC-Biasing. During inference, wildcard CTC is applied to an intermediate acoustic encoder layer to detect the target keyword, which is then injected into subsequent layers. or irrelevant parts. Formally, let V ′ = V ∪ {ϵ} be the extended CTC vocabulary, and let V ′ ∗ = V ′ ∪ {∗} include the wildcard token. A wildcard CTC alignment path π∗ is given by π∗ = (πt ∈ V′ ∗ | t = 1, . . . , T). … view at source ↗

**Figure 2.** Wildcard CTC path for target keyword. Bias keyword (OOV): “忠犬ハチ公”(Chuken Hachiko), ground truth: “忠犬ハチ公の像は渋谷駅前に立っている”, SelfCond outputs: “中堅八号の像は渋谷駅前に建っている”. “ ” is blank symbol. also found that using a relatively low threshold θ yielded better results. This suggests that detecting and biasing a larger number of candidate keywords increases the likelihood of capturing partially matched or weakly pronounced keyword… view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

43 extracted references · 35 canonical work pages

[1]

WCTC-Biasing: Retraining-free Contextual Biasing ASR with Wildcard CTC-based Keyword Spotting and Inter-layer Biasing

Introduction In recent years, the rapid progress of deep neural networks has brought about a dramatic improvement in the performance of end-to-end (E2E) automatic speech recognition (ASR) models, such as connectionist temporal classification (CTC) [1], recur- rent neural network transducers [2], attention-based encoder- decoders [3, 4], and decoder-only a...

work page Pith review arXiv 2025
[2]

Background This section describes CTC [1], self-conditioned CTC [21] and InterBiasing [20], which are the backbone of WCTC-Biasing. 2.1. Connectionist Temporal Classification E2E ASR aims to model the probability distribution of a token sequenceY= (y l ∈ V |l= 1, . . . , L)given a sequence ofD- dimensional audio featuresX= (x t ∈R D |t= 1, . . . , T), whe...

work page
[3]

Keyword spotting is performed using intermediate outputs from the acoustic en- coder

WCTC-Biasing: Wildcard CTC-based keyword spotting and Inter-layer biasing Figure 1 illustrates the proposed framework. Keyword spotting is performed using intermediate outputs from the acoustic en- coder. The detected keywords are converted into frame-level bias features and injected into the subsequent layers. 3.1. Wildcard CTC-based Keyword Spotting Alt...

work page
[4]

The models were evaluated based on character error rates (CERs) and F1 scores

Experiments To evaluate the proposed method, we conducted Japanese ASR experiments using the NeMo toolkit 1 [26]. The models were evaluated based on character error rates (CERs) and F1 scores. Following previous studies [18], we used the F1 score as evalu- ation metric for keyword recognition. Table 1:Summary of keyword set sizes and average character len...

work page
[5]

During inference, wild- card CTC is applied to an intermediate layer of the acoustic encoder to efficiently search CTC paths corresponding to tar- get keywords

Conclusions In this paper, we proposed a method to improve the recogni- tion of unknown words and proper nouns in existing CTC-based models without requiring retraining. During inference, wild- card CTC is applied to an intermediate layer of the acoustic encoder to efficiently search CTC paths corresponding to tar- get keywords. These paths are then used ...

work page
[6]

End-to-end speech recognition contextualization with large language models,

E. Lakomkin, C. Wu, Y . Fathullah, O. Kalinli, M. L. Seltzer, and C. Fuegen, “End-to-end speech recognition contextualization with large language models,” inProc. ICASSP, 2024, pp. 12 406– 12 410

work page 2024
[7]

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, p. 369–376

work page 2006
[8]

Sequence transduction with recurrent neural net- works,

A. Graves, “Sequence transduction with recurrent neural net- works,” inProc. ICML: Representation Learning Workshop, 2012

work page 2012

Show all 43 references

[9]

Attention-based models for speech recognition,

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

2015
[10]

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

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

2016
[11]

Audiopalm: A large language model that can speak and listen,

P. K. Rubenstein, C. Asawaroengchai, D. D. Nguyen, A. Bapna, Z. Borsos, F. d. C. Quitry, P. Chen, D. E. Badawy, W. Han, E. Kharitonovet al., “Audiopalm: A large language model that can speak and listen,”arXiv preprint arXiv:2306.12925, 2023

2023 arXiv
[12]

Context-aware transformer trans- ducer for speech recognition,

F.-J. Chang, J. Liu, M. Radfar, A. Mouchtaris, M. Omologo, A. Rastrow, and S. Kunzmann, “Context-aware transformer trans- ducer for speech recognition,” inProc. ASRU, 2021, pp. 503–510

2021
[13]

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,” inProc. INTERSPEECH, 2019, pp. 1418–1422

2019
[14]

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,” in Proc. INTERSPEECH, 2020, pp. 4348–4351

2020
[15]

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, J. Chan, Y . Shangguan, C. Fuegen, O. Kalinli, Y . Saraf, and M. Seltzer, “Contextualized streaming end-to-end speech recognition with trie-based deep biasing and shallow fusion,” inProc. INTER- SPEECH, 2021, pp. 1772–1776

2021
[16]

Deep context: End-to-end contextual speech recogni- tion,

G. Pundak, T. N. Sainath, R. Prabhavalkar, A. Kannan, and D. Zhao, “Deep context: End-to-end contextual speech recogni- tion,” inProc. SLT, 2018, pp. 418–425

2018
[17]

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,” inProc. INTER- SPEECH, 2020, pp. 11–15

2020
[18]

Spell my name: Keyword boosted speech recognition,

N. Jung, G. Kim, and J. S. Chung, “Spell my name: Keyword boosted speech recognition,” inProc. ICASSP, 2022, pp. 6642– 6646

2022
[19]

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, J. Chan, Y . Shangguan, C. Fuegen, O. Kalinli, Y . Saraf, and M. L. Seltzer, “Contextualized streaming end-to-end speech recognition with trie-based deep biasing and shallow fusion,” inProc. INTER- SPEECH, 2021, pp. 1772–1776

2021
[20]

Contextual Adapters for Personalized Speech Recognition in Neural Transducers,

K. M. Sathyendra, T. Muniyappa, F.-J. Chang, J. Liu, J. Suet al., “Contextual Adapters for Personalized Speech Recognition in Neural Transducers,” inProc. ICASSP, 2022, pp. 8537–8541

2022
[21]

Contextualized End-to-End Speech Recognition with Contextual Phrase Prediction Network,

K. Huang, A. Zhang, Z. Yang, P. Guo, B. Mu, T. Xu, and L. Xie, “Contextualized End-to-End Speech Recognition with Contextual Phrase Prediction Network,” inProc. INTERSPEECH, 2023, pp. 4933–4937

2023
[22]

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, pp. 10 896–10 900

2024
[23]

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, pp. 257–261

2024
[24]

Intermediate loss regularization for ctc- based speech recognition,

J. Lee and S. Watanabe, “Intermediate loss regularization for ctc- based speech recognition,” inProc. ICASSP, 2021, pp. 6224– 6228

2021
[25]

Why does ctc result in peaky behavior?

A. Zeyer, R. Schluter, and H. Ney, “Why does ctc result in peaky behavior?”arXiv preprint arXiv:2105.14849, 2021

2021 arXiv
[26]

Interbiasing: Boost unseen word recognition through biasing intermediate predictions,

Y . Nakagome and M. Hentschel, “Interbiasing: Boost unseen word recognition through biasing intermediate predictions,” in Proc. INTERSPEECH, 2024, pp. 207–211

2024
[27]

Relaxing the conditional indepen- dence assumption of CTC-based ASR by conditioning on inter- mediate predictions,

J. Nozaki and T. Komatsu, “Relaxing the conditional indepen- dence assumption of CTC-based ASR by conditioning on inter- mediate predictions,” inProc. INTERSPEECH, 2021

2021
[28]

LM + BS” indicates LM shallow fusion + beam search and “LM + KBBS

and speed perturbation [29] were also applied with the ES- PNet recipe [30]. For evaluation, we tested on one in-domain testset (CSJ eval3 [27]) and three out-of-domain testsets (JSUT-basic 5000 [31], Common V oice v8.0 [32], and TEDxJP-10K [33]). The out-of-domain sets reflec...
[29]

W- CTC: a connectionist temporal classification loss with wild cards,

X. Cai, J. Yuan, Y . Bian, G. Xun, J. Huang, and K. Church, “W- CTC: a connectionist temporal classification loss with wild cards,” inProc. ICLR, 2022

2022
[30]

Conformer: Convolution-augmented Transformer for Speech Recognition,

A. Gulati, J. Qin, C.-C. Chiu, N. Parmar, Y . Zhang, J. Yu, W. Han, S. Wang, Z. Zhang, Y . Wu, and R. Pang, “Conformer: Convolution-augmented Transformer for Speech Recognition,” in Proc. INTERSPEECH, 2020, pp. 5036–5040

2020
[31]

A comparative study on non-autoregressive modelings for speech-to-text generation,

Y . Higuchi, N. Chen, Y . Fujita, H. Inaguma, T. Komatsu, J. Lee, J. Nozaki, T. Wang, and S. Watanabe, “A comparative study on non-autoregressive modelings for speech-to-text generation,” in Proc. ASRU, 2021, pp. 47–54

2021
[32]

Nemo: a toolkit for building ai applications using neural modules,

O. Kuchaiev, J. Li, H. Nguyen, O. Hrinchuk, R. Leary, B. Gins- burg, S. Kriman, S. Beliaev, V . Lavrukhin, J. Cooket al., “Nemo: a toolkit for building ai applications using neural modules,”arXiv preprint arXiv:1909.09577, 2019

1909 arXiv
[33]

Corpus of spontaneous japanese: Its design and evaluation,

K. Maekawa, “Corpus of spontaneous japanese: Its design and evaluation,” inISCA & IEEE Workshop on Spontaneous Speech Processing and Recognition, 2003

2003
[34]

SpecAugment: A Simple Data Augmen- tation Method for Automatic Speech Recognition,

D. S. Park, W. Chan, Y . Zhang, C.-C. Chiu, B. Zoph, E. D. Cubuk, and Q. V . Le, “SpecAugment: A Simple Data Augmen- tation Method for Automatic Speech Recognition,” inProc. IN- TERSPEECH, 2019, pp. 2613–2617

2019
[35]

Audio augmen- tation for speech recognition,

T. Ko, V . Peddinti, D. Povey, and S. Khudanpur, “Audio augmen- tation for speech recognition,” inProc. INTERSPEECH, 2015, pp. 3586–3589

2015
[36]

ESPnet: End-to-End Speech Processing Toolkit,

S. Watanabe, T. Hori, S. Karita, T. Hayashi, J. Nishitoba, Y . Unno, N. Enrique Yalta Soplin, J. Heymann, M. Wiesner, N. Chen, A. Renduchintala, and T. Ochiai, “ESPnet: End-to-End Speech Processing Toolkit,” inProc. INTERSPEECH, 2018, pp. 2207– 2211

2018
[37]

Jsut corpus: free large-scale japanese speech corpus for end-to-end speech synthe- sis,

R. Sonobe, S. Takamichi, and H. Saruwatari, “Jsut corpus: free large-scale japanese speech corpus for end-to-end speech synthe- sis,”ArXiv, vol. abs/1711.00354, 2017

2017 arXiv
[38]

Common voice: A massively-multilingual speech corpus,

R. Ardila, M. Branson, K. Davis, M. Henretty, M. Kohler, J. Meyer, R. Morais, L. Saunders, F. M. Tyers, and G. Weber, “Common voice: A massively-multilingual speech corpus,” in Proc. LREC, 2020, pp. 4211–4215

2020
[39]

Construction of a large-scale japanese asr corpus on tv recordings,

S. Ando and H. Fujihara, “Construction of a large-scale japanese asr corpus on tv recordings,” inProc. ICASSP, 2021, pp. 6948– 6952

2021
[40]

Applying condi- tional random fields to Japanese morphological analysis,

T. Kudo, K. Yamamoto, and Y . Matsumoto, “Applying condi- tional random fields to Japanese morphological analysis,” inProc. EMNLP, 2004, pp. 230–237

2004
[41]

Adam: A Method for Stochastic Opti- mization,

D. P. Kingma and J. Ba, “Adam: A Method for Stochastic Opti- mization,” inProc. ICLR, 2015

2015
[42]

Attention is all you need,

A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin, “Attention is all you need,” inProc. NeurlPS, 2017, p. 6000–6010

2017
[43]

KenLM: Faster and smaller language model queries,

K. Heafield, “KenLM: Faster and smaller language model queries,” inProceedings of the Sixth Workshop on Statistical Ma- chine Translation, Jul. 2011, pp. 187–197

2011

Pith tools

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

[1] [1]

WCTC-Biasing: Retraining-free Contextual Biasing ASR with Wildcard CTC-based Keyword Spotting and Inter-layer Biasing

Introduction In recent years, the rapid progress of deep neural networks has brought about a dramatic improvement in the performance of end-to-end (E2E) automatic speech recognition (ASR) models, such as connectionist temporal classification (CTC) [1], recur- rent neural network transducers [2], attention-based encoder- decoders [3, 4], and decoder-only a...

work page Pith review arXiv 2025

[2] [2]

Background This section describes CTC [1], self-conditioned CTC [21] and InterBiasing [20], which are the backbone of WCTC-Biasing. 2.1. Connectionist Temporal Classification E2E ASR aims to model the probability distribution of a token sequenceY= (y l ∈ V |l= 1, . . . , L)given a sequence ofD- dimensional audio featuresX= (x t ∈R D |t= 1, . . . , T), whe...

work page

[3] [3]

Keyword spotting is performed using intermediate outputs from the acoustic en- coder

WCTC-Biasing: Wildcard CTC-based keyword spotting and Inter-layer biasing Figure 1 illustrates the proposed framework. Keyword spotting is performed using intermediate outputs from the acoustic en- coder. The detected keywords are converted into frame-level bias features and injected into the subsequent layers. 3.1. Wildcard CTC-based Keyword Spotting Alt...

work page

[4] [4]

The models were evaluated based on character error rates (CERs) and F1 scores

Experiments To evaluate the proposed method, we conducted Japanese ASR experiments using the NeMo toolkit 1 [26]. The models were evaluated based on character error rates (CERs) and F1 scores. Following previous studies [18], we used the F1 score as evalu- ation metric for keyword recognition. Table 1:Summary of keyword set sizes and average character len...

work page

[5] [5]

During inference, wild- card CTC is applied to an intermediate layer of the acoustic encoder to efficiently search CTC paths corresponding to tar- get keywords

Conclusions In this paper, we proposed a method to improve the recogni- tion of unknown words and proper nouns in existing CTC-based models without requiring retraining. During inference, wild- card CTC is applied to an intermediate layer of the acoustic encoder to efficiently search CTC paths corresponding to tar- get keywords. These paths are then used ...

work page

[6] [6]

End-to-end speech recognition contextualization with large language models,

E. Lakomkin, C. Wu, Y . Fathullah, O. Kalinli, M. L. Seltzer, and C. Fuegen, “End-to-end speech recognition contextualization with large language models,” inProc. ICASSP, 2024, pp. 12 406– 12 410

work page 2024

[7] [7]

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, p. 369–376

work page 2006

[8] [8]

Sequence transduction with recurrent neural net- works,

A. Graves, “Sequence transduction with recurrent neural net- works,” inProc. ICML: Representation Learning Workshop, 2012

work page 2012