Right Label Context in End-to-End Training of Time-Synchronous ASR Models

Hermann Ney; Ralf Schl\"uter; Tina Raissi

REVIEW 3 major objections 5 minor 1 cited by

Right Label Context in End-to-End Training of Time-Synchronous ASR Models

T0 review · 3 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read A factored full-sum loss with auxiliary left and right phoneme outputs lets time-synchronous ASR use right label context, improving word error on 300h data and training hybrid HMMs end-to-end with no forced alignment.

desk verdict The full-sum-only factored hybrid result is real, but the 'right label context' mechanism is not isolated from multi-task auxiliary supervision. read the letter →

arxiv 2501.04521 v2 pith:IPN2BR5W submitted 2025-01-08 cs.SD eess.AS

Tina Raissi , Ralf Schlüter , Hermann Ney This is my paper · ORCID

classification cs.SDeess.AS

keywords automaticspeechrecognitionfull-sumtrainingrightlabelcontextfactoredhybridHMMconnectionisttemporalclassificationlow-resourcetime-synchronousmodels

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper asks whether the right (future) label context can be used in full-sum training of time-synchronous speech recognition models. CTC-style discriminative training marginalizes over alignments with locally normalized outputs, so conditioning on future labels is not mathematically well-defined; the authors instead exploit the generative HMM formulation and add auxiliary left and right phoneme prediction outputs to a factored full-sum loss. On Switchboard 300h, adding the right context lowers word error relative to CTC and plain HMM full-sum models under matched training, while on LibriSpeech 960h the gain is small but consistent. The same training criterion builds a factored hybrid HMM from scratch, without forced alignment, reaching word error rates comparable to a multi-stage best-path pipeline. The conclusion is that right label context is most valuable when training data are limited.

What carries the argument

The machinery is a factored frame-level loss for HMM full-sum training. Instead of one softmax over all triphones, Eq. (7b) factorizes the joint posterior $P(a_{t-1}, a_t, a_{t+1}\mid h_t)$ into three context-independent softmax branches for left, center, and right phonemes, and approximates the gradient as a weighted sum of per-factor log posteriors. The weights $\gamma_t$ come from a forward-backward pass over all alignments, and the state prior scale is set to zero during training. This keeps the generative asymmetry of the HMM, so the model can see both neighboring labels, while keeping the output vocabulary linear in the phoneme set. The same outputs are used at decoding time by selecting one branch, usually the center phoneme, for best-path search.

What would settle it

A controlled ablation that removes only the right-label auxiliary branch while keeping the left branch and all other training settings identical; if word error rates stay the same, the paper's attribution of the gain to right label context is falsified. A second check is to train on a mid-size corpus with a right-context output whose labels are randomly permuted; if word error still improves, the effect is not information from the right label.

Watch

Extended reading notes

Core claim

On its own terms, the paper establishes that conditioning acoustic-model training on the right label context is feasible in the full-sum setting and worthwhile. The central claim is that a factored loss with three softmax outputs per frame, one each for the left, center, and right phoneme, can stand in for the joint triphone posterior, and that when trained by summing over all alignments the right-label output reduces word error. The benefit is concentrated in the 300-hour Switchboard task; on 960-hour LibriSpeech the differences are small but the right-context model remains the best zero-order model. A second claim is that this full-sum criterion alone suffices to train hybrid HMM systems: no external alignment is needed, and the resulting factored hybrid model performs comparably to a system trained with an alignment in a multi-stage pipeline. The authors also report that HMM full-sum models outperform CTC under matched conditions.

Load-bearing premise

The argument assumes that the simplified gradient obtained by treating left, center, and right phoneme outputs as independent, and by dropping the state prior during training, still captures the benefit of the right label context. The center output is never actually conditioned on the right phoneme, so the observed gains could come from multi-task regularization rather than true right-context modeling.

Editorial extensions

If this is right

If right-context auxiliary outputs are the cause of the gains, low-resource ASR recipes built on full-sum training should include them rather than relying on CTC or center-only HMM targets.
The full-sum-only factored hybrid HMM removes the forced-alignment stage, so a hybrid pipeline can be trained with a single criterion and still match multi-stage word error rates.
The right-context model offers a 10% training-time speedup and a 16% decoding real-time-factor improvement over the diphone factored model with only a 1.6% relative word-error increase, making it the cheaper deployment choice.
Under matched setups, HMM-based full-sum models outperform CTC, suggesting the HMM topology contributes part of the benefit independently of label context.

Reading between the lines

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

Editorial inference: because the center output is never conditioned on the right label, the same gain may be obtainable by adding a right-phoneme auxiliary head to CTC or transducer models, where full right-context conditioning is not possible; a matched comparison would isolate the mechanism.
Editorial inference: the right-context benefit appears to shrink as data grows, so an explicit test is to sweep training-set size and measure the word-error gap between right-context and center-only models; if the gap narrows monotonically, the effect is a low-resource regularization phenomenon.
Editorial inference: the zero prior scale and context-independent factorization trade statistical fidelity for trainability; a future model that makes the center output depend on a lattice-marginalized right phoneme would test whether true conditioning yields larger gains.

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.

Desk Editor's Note

read the letter

Short version: this is a solid empirical paper from a group that knows the terrain, but the title claim overreaches the experiments. The genuinely new demonstration—a from-scratch, full-sum-trained factored hybrid HMM with no forced alignment and no state prior—works: Table IV matches a Viterbi pipeline with external alignments (6.0/6.4 vs 6.1/6.8 on dev-other/test-other). That is worth something for the hybrid HMM community. The paper is also honest about the simplifying assumptions in Eq. (7b): it factors the joint triphone posterior into independent left/center/right softmax outputs and drops the prior scale during training. The paired comparisons (Exp. 2 vs 3, 6 vs 7) are well-controlled and the gains are consistent on SWB.

The soft spot is the causal story. Exp. 3 and 7 add two auxiliary heads—left and right—versus a single center head. Nothing isolates the right label's contribution. The model never conditions the center label on the right label; the right label enters only as a second auxiliary prediction target. So the observed WER gain could just as well come from multi-task regularization or from the left head alone. The 'right label context is particularly beneficial with limited data' conclusion is also built on a weak comparison: different corpora, different epoch counts, no significance testing. I would ask for a two-head ablation (center+left only), a control with a non-contextual secondary head, and ideally a significance check or repeated runs before the causal claim can stand.

The 'no prior work modeled right label context for full-sum' sentence is also awkward given that [13] is the authors' own earlier investigation of right context; the paper's own introduction says it extends [13], so the novelty claim needs a sharper boundary. Elsewhere the self-citations are appropriate background.

None of this sinks the paper. The from-scratch full-sum factored hybrid is a real, reproducible result with code and configs linked. The approximation is acknowledged, not hidden. For an ASR audience this is useful and should go to peer review, but the revision should add the ablation that would let the title claim actually be tested.

Referee Report

3 major / 5 minor

Summary. The manuscript proposes a factored full-sum training criterion for time-synchronous ASR, in which the HMM emission model is decomposed into three auxiliary softmax outputs for left, center, and right phoneme labels (Eq. (7b)). The authors argue that this formulation provides access to right label context during full-sum training, which locally normalized CTC and RNN-T formulations lack. They report experiments on Switchboard 300h and LibriSpeech 960h with a 12-layer Conformer encoder: a monophone HMM trained with left/center/right auxiliary losses improves WER over CTC and center-only HMM baselines, and a full-sum-trained diphone factored hybrid HMM matches a multi-stage Viterbi-trained system that uses an external alignment. The paper also reports training-time and real-time-factor comparisons showing that the auxiliary-loss model is slightly faster than the diphone model.

Significance. Should the causal claim survive scrutiny, the paper would make a useful contribution: it offers a practical way to incorporate future-label information in alignment-free full-sum training and shows that hybrid HMM systems can be trained end-to-end without external alignments. The experimental setup is careful in several respects: all models share the same encoder, features, SpecAugment, learning-rate schedule, and 4-gram LM; comparisons are made under fixed epoch counts within each corpus; and example configurations are released. The RTF and training-time measurements add practical value. The main weakness is that the evidence does not isolate the effect of right label context from the effect of adding auxiliary prediction heads in general, so the title's causal claim is currently stronger than the experiments support.

major comments (3)

[Section II-D / Eq. (7b); Table I Exp. 2 vs Exp. 3; Table II Exp. 6 vs Exp. 7] The central comparison for the title's claim contrasts a center-only monophone HMM with a model trained using left, center, and right auxiliary outputs. In the actual objective of Eq. (7b), the joint triphone posterior is factored into three independent softmax outputs P(l|h), P(c|h), and P(r|h); the center label is never conditioned on the right label, and the right label appears only as an additional prediction target. The observed gain therefore conflates the right-context head with the left-context head and with the general regularizing effect of any auxiliary output. No ablation with only (l|h) and (c|h), and no control with a non-contextual or random auxiliary head, is reported. Without such controls, the results support the weaker conclusion that multi-task auxiliary supervision helps full-sum HMM training, not specifically that right label context is beneficial.
[Section III-A / Section III-B] The claim that right label context is "particularly beneficial when training data resources are limited" rests on comparing SWB 300h (50 epochs; gains of 0.8/0.6 WER in Table I) with LBS 960h (25 epochs; gains of 0.3/0.2 WER). These corpora differ jointly in data size, domain, feature dimension, and number of training epochs, so the larger SWB gain cannot be attributed to data scarcity alone. No significance testing or multi-seed variance is reported, and several of the absolute differences are only 0.1-0.4% WER. To support the low-resource conclusion, the authors should hold epoch count and optimization schedule fixed across corpora, or systematically vary the training-set size within a single corpus, and report confidence intervals or seed variability.
[Section II-D / Eq. (7a)-(8)] The approximation in Eq. (7b) is presented as a simplification of Eq. (7a), but it is not derived and the marginal posterior gamma_t(a) is not defined. Equation (8) defines gamma for the joint triphone state; using it in Eq. (7b) requires marginalizing over the other two context labels, and this marginal is never written. Replacing a properly normalized triphone posterior with a sum of three independent factor gradients, and setting the prior scale beta to zero, changes the training objective substantially. The paper acknowledges the simplification in Section II-F1, but the acknowledgement does not bridge the gap between the claimed "right label context" mechanism and the objective actually optimized. A formal statement of the approximation, or at minimum the missing ablations, is needed before the mechanism in the title can be accepted.

minor comments (5)

[Section II-F1] The sentence describing the gradient as "the weighted sum of each partial log probabilities of left, center, and left phonemes" should say "left, center, and right phonemes."
[Equation (7b)] The notation gamma_t(a) is used without definition; please specify explicitly that it is the marginal of the joint posterior in Eq. (8) over the other two context labels.
[Table II] The column header "dev-othertest-other" is missing a space, and the test-set column naming is inconsistent across Tables I, II, and IV; please unify the notation.
[Section III-A] The description that Exps. 9 and 10 in Table II are initialized with "the seed model" is ambiguous; please state that they are initialized from the models of Exps. 2 and 3, respectively.
[References] Citations [4] and [17] appear to be two versions of the same paper (SLT 2023 and the 2022 preprint); please consolidate or clarify the distinction.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the paper's empirical comparisons and full-sum-only hybrid HMM results are self-contained; self-citations provide background but do not force the conclusions.

full rationale

The central claims are supported by genuine experimental comparisons (Tables I, II, and IV) between models that share the same encoder, data, training epochs, and decoding configuration. The full-sum-only hybrid HMM result is measured against a multi-stage external-alignment baseline rather than derived from it. The gradient simplification in Eq. (7b) is explicitly presented as an approximation, and the paper does not claim that the approximate gradient is a derivation of the observed WER gains. Self-citations to [8], [13], and [15] supply the factored-hybrid modeling framework and prior results, but the new contribution—full-sum training with auxiliary left/right label factors—is evaluated independently. No uniqueness theorem is invoked to forbid alternatives, and no fitted parameter is renamed as a prediction. A possible empirical confound exists: Experiments 3 and 7 add both left and right auxiliary heads relative to the monophone HMM baseline, so the isolated contribution of the right label is not directly ablated. That is a validity concern, not a circularity, because the conclusion is not embedded in the model definition or in the cited prior work. Accordingly, no circular step is scored.

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

The central claim depends on the factorization approximation of the full-sum gradient and on the choice to omit the state prior. The decoding scales and epoch counts are additional tuning choices. No fundamentally new entities are postulated.

free parameters (3)

Decoding scales (prior beta, transition eta, LM lambda) = not reported (beta=0 during training)
Eq. (6) uses scaled log-linear decoding with prior, transition, and LM scales; values are not reported, so the exact operating point is not reproducible from the text.
Training epochs = 50 for SWB, 25 for LBS (doubled for Table II)
The 'limited data' comparison uses different epoch counts per corpus, confounding dataset size with training duration.
Learning rates = peak 6e-4 OCLR, 5e-5 for fine-tuning
Standard hyperparameters chosen without a reported tuning procedure; they affect all models equally in paired comparisons.

assumptions (4)

domain assumption First-order Markov transition independence in the HMM
Eq. (4a) assumes P(s_t|s_{t-1}) and conditional independence of emissions; standard for hybrid HMM.
domain assumption Pronunciation lexicon with a single pronunciation per word
Section II-A states 'We consider a single pronunciation for each word', simplifying the label topology.
ad hoc to paper Factorization of the joint posterior into independent left, center, right outputs (Eq. 7b)
Section II-F1 introduces this approximation without a formal justification; the paper's central claim depends on it.
ad hoc to paper Prior scale beta=0 during training
Section II-F2 states this is a special case to avoid convergence issues; it changes the objective from the strict generative model.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Right Label Context in End-to-End Training of Time-Synchronous ASR Models." pith.science (2026). https://pith.science/paper/IPN2BR5W

@misc{pith2026250104521,
  author       = {Pith},
  title        = {Pith review of: Right Label Context in End-to-End Training of Time-Synchronous ASR Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IPN2BR5W}},
  note         = {Machine review of arXiv:2501.04521}
}

read the original abstract

Current time-synchronous sequence-to-sequence automatic speech recognition (ASR) models are trained by using sequence level cross-entropy that sums over all alignments. Due to the discriminative formulation, incorporating the right label context into the training criterion's gradient causes normalization problems and is not mathematically well-defined. The classic hybrid neural network hidden Markov model (NN-HMM) with its inherent generative formulation enables conditioning on the right label context. However, due to the HMM state-tying the identity of the right label context is never modeled explicitly. In this work, we propose a factored loss with auxiliary left and right label contexts that sums over all alignments. We show that the inclusion of the right label context is particularly beneficial when training data resources are limited. Moreover, we also show that it is possible to build a factored hybrid HMM system by relying exclusively on the full-sum criterion. Experiments were conducted on Switchboard 300h and LibriSpeech 960h.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

Gradient-Based Speech-to-Text Alignment for Any ASR Model: From CTC to Speech LLMs
cs.CL 2026-07 accept novelty 6.0 of 10

Gradient saliency of teacher-forced token log-probabilities, decoded by dynamic programming, yields usable word alignments for every ASR family including speech LLMs.

Reference graph

Works this paper leans on

34 extracted references · 33 canonical work pages · cited by 1 Pith paper

[13]

Towards Consistent Hybrid HMM Acoustic Modeling

T. Raissi, E. Beck, R. Schl¨ uter, and H. Ney, “Towards Co nsistent Hybrid HMM Acoustic Modeling,” Apr. 2021, arXiv:2104.02387

work page Pith review arXiv 2021
[1]

Connection- ist temporal classiﬁcation: labelling unsegmented sequen ce data with recurrent neural networks,

A. Graves, S. Fern´ andez, F. Gomez, and J. Schmidhuber, “ Connection- ist temporal classiﬁcation: labelling unsegmented sequen ce data with recurrent neural networks,” in Proc. ICML, 2006, pp. 369–376

work page 2006
[2]

Sequence transduction with recurrent neura l networks,

A. Graves, “Sequence transduction with recurrent neura l networks,” in Proc. ICML, Edinburgh, Scotland, Jun. 2012, Workshop on Representa- tion Learning, arXiv:1211.3711

arXiv 2012
[3]

Monotonic recu rrent neural network transducer and decoding strategies,

A. Tripathi, H. Lu, H. Sak, and H. Soltau, “Monotonic recu rrent neural network transducer and decoding strategies,” in Proc. IEEE ASRU, 2019, pp. 944–948

work page 2019
[4]

H MM vs. CTC for Automatic Speech Recognition: Comparison Based on F ull- Sum Training from Scratch,

T. Raissi, W. Zhou, S. Berger, R. Schl¨ uter, and H. Ney, “H MM vs. CTC for Automatic Speech Recognition: Comparison Based on F ull- Sum Training from Scratch,” in Proc. IEEE SLT , 2023

work page 2023
[5]

H. A. Bourlard and N. Morgan, Connectionist Speech Recognition: a Hybrid Approach. Norwell, MA: Kluwer Academic Publishers, 1993

work page 1993
[6]

Neural machine tra nslation of rare words with subword units,

R. Sennrich, B. Haddow, and A. Birch, “Neural machine tra nslation of rare words with subword units,” in Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics ( V olume 1: Long Papers) , K. Erk and N. A. Smith, Eds. Berlin, Germany: Association for Computational Linguistics, Aug. 2016, pp. 1715–1725. [Online]. Available: ...

work page 2016
[7]

Multitask lear ning of context- dependent targets in deep neural network acoustic models,

P . Bell, P . Swietojanski, and S. Renals, “Multitask lear ning of context- dependent targets in deep neural network acoustic models,” IEEE/ACM Transactions on Audio, Speech, and Language Process. , vol. 25, no. 2, pp. 238–247, 2016

work page 2016

Show all 34 references

[8]

Context-de pendent acoustic modeling without explicit phone clustering,

T. Raissi, E. Beck, R. Schl¨ uter, and H. Ney, “Context-de pendent acoustic modeling without explicit phone clustering,” in Proc. IEEE ICASSP , 2020

2020
[9]

RNN- transducer with stateless prediction network,

M. Ghodsi, X. Liu, J. Apfel, R. Cabrera, and E. Weinstein, “RNN- transducer with stateless prediction network,” in Proc. IEEE ICASSP , 2020, pp. 7049–7053

2020
[10]

Zipformer: A faster and better encoder for automatic speech recognition,

Z. Y ao, L. Guo, X. Y ang, W. Kang, F. Kuang, Y . Y ang, Z. Jin, L. Lin, and D. Povey, “Zipformer: A faster and better encoder for automatic speech recognition,” in The Twelfth International Conference on Learning Representations , 2023

2023
[11]

Phoneme ba sed neural transducer for large vocabulary speech recognition,

W. Zhou, S. Berger, R. Schl¨ uter, and H. Ney, “Phoneme ba sed neural transducer for large vocabulary speech recognition,” in Proc. IEEE ICASSP, Jun. 2021, pp. 5644–5648

2021
[12]

Global normalization for streaming speech recogn ition in a modular framework,

E. V ariani, K. Wu, M. D. Riley, D. Rybach, M. Shannon, and C. Al- lauzen, “Global normalization for streaming speech recogn ition in a modular framework,” Proc. NeurIPS, vol. 35, pp. 4257–4269, 2022

2022
[14]

Conformer: Convolutio n- Augmented Transformer for Speech Recognition,

A. Gulati, J. Qin, C.-C. Chiu, N. Parmar, Y . Zhang, J. Y u, W. Han, S. Wang, Z. Zhang, Y . Wu, and R. Pang, “Conformer: Convolutio n- Augmented Transformer for Speech Recognition,” in Proc. Interspeech, Shanghai, China, Oct. 2020, pp. 5036–5040

2020
[15]

Competitive and resource efﬁcient factored hybrid HMM systems are simpl er than you think,

T. Raissi, C. L ¨ uscher, M. Gunz, R. Schl¨ uter, and H. Ney , “Competitive and resource efﬁcient factored hybrid HMM systems are simpl er than you think,” Proc. Interspeech, 2023

2023
[16]

CDNN : A context dependent neural network for continuous speech recognitio n,

H. Bourlard, N. Morgan, C. Wooters, and S. Renals, “CDNN : A context dependent neural network for continuous speech recognitio n,” in Proc. IEEE ICASSP , 1992

1992
[17]

HMM vs. CTC for automatic speech recognition: Comparison based on f ull-sum training from scratch,

T. Raissi, W. Zhou, S. Berger, R. Schl¨ uter, and H. Ney, “ HMM vs. CTC for automatic speech recognition: Comparison based on f ull-sum training from scratch,” 2022

2022
[18]

Weighted ﬁnite-sta te transducers in speech recognition,

M. Mohri, F. Pereira, and M. Riley, “Weighted ﬁnite-sta te transducers in speech recognition,” Computer Speech & Language , vol. 16, no. 1, pp. 69–88, 2002

2002
[19]

Regarding topology and variant fra me rates for differentiable WFST-based end-to-end asr,

Z. Zhao and P . Bell, “Regarding topology and variant fra me rates for differentiable WFST-based end-to-end asr,” in Proc. Interspeech, 2023

2023
[20]

CTC in the co ntext of generalized full-sum HMM training,

A. Zeyer, E. Beck, R. Schl¨ uter, and H. Ney, “CTC in the co ntext of generalized full-sum HMM training,” in Proc. Interspeech , 2017, pp. 944–948

2017
[21]

Investigating the effect of label topology and training criterion on asr pe rformance and alignment quality,

T. Raissi, C. L ¨ uscher, S. Berger, R. Schl¨ uter, and H. Ney, “Investigating the effect of label topology and training criterion on asr pe rformance and alignment quality,” Proc. Interspeech, 2024

2024
[22]

Tree-based state tying for high accuracy acoustic modelling,

S. J. Y oung, J. J. Odell, and P . C. Woodland, “Tree-based state tying for high accuracy acoustic modelling,” in Proc. ACL, 1994, pp. 307–312

1994
[23]

Switchboard : Telephone Speech Corpus for Research and Development,

J. Godfrey, E. Holliman, and J. McDaniel, “Switchboard : Telephone Speech Corpus for Research and Development,” in Proc. IEEE ICASSP , 1992

1992
[24]

Libr iSpeech: An ASR Corpus Based on Public Domain Audio Books,

V . Panayotov, G. Chen, D. Povey, and S. Khudanpur, “Libr iSpeech: An ASR Corpus Based on Public Domain Audio Books,” in Proc. IEEE ICASSP, 2015

2015
[25]

RETURNN: The RWTH extensible training framework for unive rsal recurrent neural networks,

P . Doetsch, A. Zeyer, P . V oigtlaender, I. Kulikov, R. Schl¨ uter, and H. Ney, “RETURNN: The RWTH extensible training framework for unive rsal recurrent neural networks,” in Proc. IEEE ICASSP , 2017, pp. 5345– 5349

2017
[26]

RASR-the RWTH Aachen u ni- versity open source speech recognition toolkit,

D. Rybach, S. Hahn, P . Lehnen, D. Nolden, M. Sundermeyer , Z. T ¨ uske, S. Wiesler, R. Schl¨ uter, and H. Ney, “RASR-the RWTH Aachen u ni- versity open source speech recognition toolkit,” in Proc. ieee automatic speech recognition and understanding workshop , 2011

2011
[27]

RA SR2: The RWTH ASR toolkit for generic sequence-to-sequence speech r ecogni- tion,

W. Zhou, E. Beck, S. Berger, R. Schl¨ uter, and H. Ney, “RA SR2: The RWTH ASR toolkit for generic sequence-to-sequence speech r ecogni- tion,” 2023

2023
[28]

Sisyphus, a workﬂow ma nager designed for machine translation and automatic speech reco gnition,

J.-T. Peter, E. Beck, and H. Ney, “Sisyphus, a workﬂow ma nager designed for machine translation and automatic speech reco gnition,” in Proc. EMNLP , 2018, pp. 84–89

2018
[29]

Gamma tone features and feature combination for large vocabulary speech recogn ition,

R. Schl¨ uter, I. Bezrukov, H. Wagner, and H. Ney, “Gamma tone features and feature combination for large vocabulary speech recogn ition,” in Proc. IEEE ICASSP , 2007

2007
[30]

SpecAugment on Large Scale Datasets,

D. S. Park, Y . Zhang, C.-C. Chiu, Y . Chen, B. Li, W. Chan, Q . V . Le, and Y . Wu, “SpecAugment on Large Scale Datasets,” in Proc. IEEE ICASSP, Brighton, UK, May 2019, pp. 6879–6883

2019
[31]

Efﬁcient T raining of Neural Transducer for Speech Recognition,

W. Zhou, W. Michel, R. Schl¨ uter, and H. Ney, “Efﬁcient T raining of Neural Transducer for Speech Recognition,” in Proc. Interspeech, Sep. 2022, arXiv:2204.10586

2022 arXiv
[32]

Conformer: Convolution-augme nted Transformer for Speech Recognition,

A. Gulati, J. Qin, C. Chiu, N. Parmar, Y . Zhang, J. Y u, W. H an, S. Wang, Z. Zhang, Y . Wu, and R. Pang, “Conformer: Convolution-augme nted Transformer for Speech Recognition,” in Proc. Interspeech, 2020

2020
[33]

Super-convergence: V ery f ast training of neural networks using large learning rates,

L. N. Smith and T. Nicholay, “Super-convergence: V ery f ast training of neural networks using large learning rates,” in Artiﬁcial intelligence and machine learning for multi-domain operations application s, 2019

2019
[34]

Incorporating Nesterov Momentum into Adam,

T. Dozat, “Incorporating Nesterov Momentum into Adam, ” in Proc. ICLR, 2016

2016

Pith tools

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

[1] [13]

Towards Consistent Hybrid HMM Acoustic Modeling

T. Raissi, E. Beck, R. Schl¨ uter, and H. Ney, “Towards Co nsistent Hybrid HMM Acoustic Modeling,” Apr. 2021, arXiv:2104.02387

work page Pith review arXiv 2021

[2] [1]

Connection- ist temporal classiﬁcation: labelling unsegmented sequen ce data with recurrent neural networks,

A. Graves, S. Fern´ andez, F. Gomez, and J. Schmidhuber, “ Connection- ist temporal classiﬁcation: labelling unsegmented sequen ce data with recurrent neural networks,” in Proc. ICML, 2006, pp. 369–376

work page 2006

[3] [2]

Sequence transduction with recurrent neura l networks,

A. Graves, “Sequence transduction with recurrent neura l networks,” in Proc. ICML, Edinburgh, Scotland, Jun. 2012, Workshop on Representa- tion Learning, arXiv:1211.3711

arXiv 2012

[4] [3]

Monotonic recu rrent neural network transducer and decoding strategies,

A. Tripathi, H. Lu, H. Sak, and H. Soltau, “Monotonic recu rrent neural network transducer and decoding strategies,” in Proc. IEEE ASRU, 2019, pp. 944–948

work page 2019

[5] [4]

H MM vs. CTC for Automatic Speech Recognition: Comparison Based on F ull- Sum Training from Scratch,

T. Raissi, W. Zhou, S. Berger, R. Schl¨ uter, and H. Ney, “H MM vs. CTC for Automatic Speech Recognition: Comparison Based on F ull- Sum Training from Scratch,” in Proc. IEEE SLT , 2023

work page 2023

[6] [5]

H. A. Bourlard and N. Morgan, Connectionist Speech Recognition: a Hybrid Approach. Norwell, MA: Kluwer Academic Publishers, 1993

work page 1993

[7] [6]

Neural machine tra nslation of rare words with subword units,

R. Sennrich, B. Haddow, and A. Birch, “Neural machine tra nslation of rare words with subword units,” in Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics ( V olume 1: Long Papers) , K. Erk and N. A. Smith, Eds. Berlin, Germany: Association for Computational Linguistics, Aug. 2016, pp. 1715–1725. [Online]. Available: ...

work page 2016

[8] [7]

Multitask lear ning of context- dependent targets in deep neural network acoustic models,

P . Bell, P . Swietojanski, and S. Renals, “Multitask lear ning of context- dependent targets in deep neural network acoustic models,” IEEE/ACM Transactions on Audio, Speech, and Language Process. , vol. 25, no. 2, pp. 238–247, 2016

work page 2016