Towards Maximum Likelihood Training for Transducer-based Streaming Speech Recognition

Hyeonseung Lee; Ji Won Yoon; Nam Soo Kim; Sungsoo Kim

REVIEW 4 major objections 5 minor 27 references

Towards Maximum Likelihood Training for Transducer-based Streaming Speech Recognition

T0 review · 4 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Training streaming transducers with a learned forward-variable compensation term, FoCCE, estimates the true likelihood and lowers word error rates on LibriSpeech and TED-LIUM3.

desk verdict FoCCE is a genuinely new training heuristic with small but consistent WER gains, but the paper's claim that it recovers the exact likelihood does not survive close reading. read the letter →

arxiv 2411.17537 v1 pith:AA7QQVI2 submitted 2024-11-26 eess.AS cs.LG

Hyeonseung Lee , Ji Won Yoon , Sungsoo Kim , Nam Soo Kim This is my paper · ORCID

classification eess.AScs.LG

keywords streamingspeechrecognitionRNNtransducerdeformedlikelihoodforwardvariablecausalcompensationmaximumtrainingnormalizingflowsencoderworderrorrate

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 argues that streaming transducer ASR models are trained with a 'deformed likelihood': the standard RNN-T forward recursion was designed for encoders that see the whole input, so applying it to chunked causal encoders does not compute the actual likelihood of the streaming model. It quantifies the mismatch as a ratio called forward variable causal compensation (FoCC) and trains a separate FoCCE network to estimate that ratio from next-chunk acoustic densities, then inserts it into the forward recursion during training. With FoCCE, inference is unchanged; only the training objective moves closer to true maximum likelihood. On LibriSpeech and TED-LIUM3, FoCCE training lowers word error rates and narrows the gap to non-streaming transducers, with the gain controlled by a scaling factor $\lambda_\gamma$.

What carries the argument

The key object is the FoCC ratio $\gamma_\theta(t,u)$ from Eq. (10), defined as $P_\theta(y_{0:u}, z_u \le t < z_{u+1} \mid x_{1:e(t+1)}) / P_\theta(y_{0:u}, z_u \le t < z_{u+1} \mid x_{1:e(t)})$, which equals a ratio of next-chunk densities when $t$ crosses a chunk boundary and is $1$ otherwise. The machinery is the modified forward recursion of Eq. (11) (and its estimable version Eqs. (18)--(19)): the compensation multiplies only the blank-transition term, while the label-transition term is unchanged. The FoCCE network estimates the two densities in Eq. (14) using a normalizing flow, and a stop-gradient keeps the compensation from altering the transducer's gradients except through the objective.

What would settle it

Take a small synthetic streaming transducer with a known generative distribution, compute the standard recursion from Eq. (6) with streaming probabilities from Eq. (9) and the FoCC-corrected recursion from Eq. (11) with the exact ratio of Eq. (10), and compare the two log-likelihoods on all sequences; agreement would falsify the deformed-likelihood premise. A softer check is whether the trained FoCCE's $\gamma_\omega(t,u)$ estimates converge to 1, which would suggest the reported WER gains come from the auxiliary density loss rather than likelihood correction.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that the streaming transducer's likelihood can be corrected by a boundary ratio. At timesteps $t$ where the causal encoder's context grows ($e(t) < e(t+1)$), the partial-alignment probability changes by the factor $\gamma_\theta(t,u)$ in Eq. (10), a ratio of conditional densities of the newly observed acoustic chunk with and without the target history. Because this ratio cannot be read off the transducer's softmax outputs, the paper estimates it with a separate normalizing-flow network and defines a modified forward recursion $\hat{\alpha}_{\theta,\omega}$ in Eqs. (18)--(19) whose final value approximates the true likelihood. The paper claims that training with this modified objective, rather than the conventional recursion, is closer to maximum likelihood training for streaming transducers and produces the WER improvements reported in Table I.

Load-bearing premise

The load-bearing premise is that plugging a chunked causal encoder's local probabilities into the standard full-context forward recursion gives the wrong likelihood for the streaming model; if that gap is actually zero, FoCCE has nothing to correct.

Editorial extensions

If this is right

FoCCE training leaves the deployed streaming transducer architecture and inference cost unchanged, since the compensation network is used only during training.
On LibriSpeech, FoCCE narrows the streaming-to-non-streaming WER gap by 26.3% on test-clean and 12.3% on test-other; on TED-LIUM3 the test-set gap is reduced by 17.7%.
The modified objective contains the conventional one as the limit $\lambda_\gamma \to 0$, so FoCCE is a smooth generalization of standard streaming transducer training.
Because FoCCE works with local-probability transducer models, it avoids the accuracy trade-off that the paper attributes to globally normalized likelihood objectives.

Reading between the lines

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

A direct testable extension is to compare FoCCE against a control that keeps the same auxiliary density loss but sets $\lambda_\gamma=0$; if WER gains persist, part of the benefit comes from the density objective itself rather than from likelihood correction.
If the deformed-likelihood premise is right, the compensation's importance should vary with chunk structure: with a larger chunk size $C$ there are fewer boundary steps where $\gamma$ departs from 1, so the expected gain should shrink; measuring FoCCE gains across chunk sizes would test the mechanism.
The FoCC idea transfers to other sequence models with chunked causal encoders, such as CTC or attention-based streaming decoders, wherever a full-context dynamic program is reused with causally restricted local probabilities.

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

Colleague — quick take: the FoCCE trick is new and the WER table shows consistent small gains (roughly 0.2–0.4 absolute, 12–26% of the streaming gap), so this is a useful empirical note for people who build streaming ASR recipes. But the theoretical frame is much weaker than the abstract implies.

The paper asserts, without proof, that the conventional streaming recursion is a “deformed likelihood.” And the gamma correction in (10)–(12) has a definitional problem: the ratio is written as a probability under the transducer θ, yet θ only models P(y,z|x), not P(x|y,z) or future audio. If you take the streaming model literally, the event in (10) depends only on x_{1:e(t)}, so the ratio is 1 and the correction collapses. If instead the ratio is meant to be a data density ratio, then the modified loss is not a maximum-likelihood objective for the transducer; it is a heuristic reweighting plus an auxiliary future-chunk density task. The paper can’t have it both ways.

What it does well: FoCCE is a clean, simple modification to the forward recursion, inference cost is unchanged, and the experiments use standard LibriSpeech and TED-LIUM3 recipes with a strong Zipformer baseline. The hyperparameters are reported. The idea of using a future-chunk density ratio as a training signal is worth testing further.

Soft spots: there is no ablation separating the auxiliary density objective (17) from the gamma multiplication; a plain future-prediction auxiliary loss might give the same gains. There are no error bars on the WERs, and λγ=0.05 is selected on dev, so the gains could be partly hyperparameter luck. The authors also concede they “could not find any analytic solution” and simply assume γ_ω converges to γ. That is a big assumption for the core mechanism.

Bottom line: this is worth reading for streaming ASR practitioners and as a discussion piece, but as a claim about maximum likelihood it overreaches. I’d send it to a serious referee because the idea is novel and the experiments are real, but the referee should ask for a toy verification or an ablation before accepting the mechanism. My own verdict is skeptical; I would not cite it as evidence for exact-likelihood training.

Referee Report

4 major / 5 minor

Summary. This paper considers streaming RNN-T training. The authors argue that applying the standard non-streaming forward recursion with causally computed local probabilities yields a 'deformed likelihood,' and introduce a correction factor gamma(t,u) in Eq. (10), called FoCC, which they claim recovers the exact likelihood if inserted into a modified recursion (Eqs. (11)-(12)). Since gamma has no analytic form, a separate FoCCE network estimates it via normalizing flows, trained with the auxiliary objective in Eq. (17), and the modified objective in Eqs. (18)-(19) is used to train the transducer. Experiments on LibriSpeech and TED-LIUM3 report consistent but modest WER improvements (Table I).

Significance. If the central claim were correct, the paper would offer a principled way to train streaming transducers closer to maximum likelihood at no inference cost, and the WER gains would be an important empirical confirmation. The paper has strengths: the experimental setup is based on a public recipe (icefall), the transducer inference footprint is held fixed across conditions, and the reported gains are consistent across two corpora. However, the central theoretical mechanism is not supported: the paper concedes in Section III that no analytic solution for gamma is available, and the arguments below show that the 'deformed likelihood' premise is not established and that the estimator does not target the claimed quantity. The empirical gains are therefore not attributable to exact-likelihood correction, and the significance of the contribution as framed is not demonstrated.

major comments (4)

[Section II-B, Eqs. (6)-(9)] The claim that substituting streaming local probabilities into the forward recursion 'breaks Bayes' rule' is not demonstrated and is at odds with the standard causal transducer model. In a model with local probabilities conditioned on x_{1:e(s)} at frame s, the recursion (6) is an exact dynamic program: each transition multiplies the accumulated sum by the appropriate local probability, and the final value in (7) is the sum over all complete alignments of the product of local probabilities, which is by definition the conditional likelihood of the causal model. The interpretation of alpha(t,u) in (4) as a full marginal P(y0:u, z_u <= t <= z_{u+1}|x_{1:T}) does not match what the recursion computes; the intermediate alpha values are prefix sums, not marginals over complete alignments. No counterexample or derivation supporting the deformation claim is supplied, and Section III explicitly states that an analytic solution could not be found.
[Eq. (10), Section III] The gamma ratio is not a well-defined likelihood ratio of the streaming transducer. For the causal model in (8)-(9), the probability of the event E_{t,u} = {y0:u, z_u <= t < z_{u+1}} under the full conditioning x_{1:T} is not generally equal to a probability conditioned on x_{1:e(t)} only, because the event constrains future emissions whose local probabilities depend on later inputs. Conditioning on x_{1:e(t)} therefore defines a different, truncated model, and the numerator and denominator in (10) are probabilities under two different models rather than a Bayes factor of one model. The second equality in (10) additionally requires a generative density over acoustic features, which is not part of the transducer model. If alpha is instead read as the standard prefix forward variable, then gamma is identically 1 because the prefix sum up to t depends only on x_{1:e(t)}; either way, the correction is not recovering a missing likelihood. Thus the convergence assumption in Section III that gamma_omega approaches gamma_theta is an unverified postulate about an ill-defined target.
[Section III, Eqs. (14)-(17)] The FoCCE training objective does not estimate the quantity defined in (10). The conditioning event in chi_omega(t,u), namely {z_u <= t < z_{u+1}}, is an alignment-dependent event, but the network input in (15)-(16) is only [f_t^chi; g_u^chi], i.e., a causal acoustic embedding and the target prefix embedding. The objective L_chi in (17) trains a predictive density of the next input chunk given the input prefix and label prefix; nothing in the architecture or the objective uses the alignment variable, so chi_omega cannot learn the alignment-conditioned density in (14). The ratio in (13) is therefore not an estimator of the FoCC ratio.
[Table I] The experiments do not isolate the effect of the gamma reweighting from the effect of the auxiliary density objective L_chi. Setting lambda_gamma=0 in (13) makes the modified recursion (19) coincide with the conventional recursion while retaining L_chi in the total objective (20). A control row with streaming plus FoCCE network but lambda_gamma=0 (or equivalently streaming plus L_chi only) is needed to attribute the WER improvements (e.g., LibriSpeech test-clean 3.27% to 3.13%) to gamma correction rather than to the auxiliary future-chunk prediction task. Without this control, the empirical claim in the abstract is not supported by the table.

minor comments (5)

[Eq. (17)] The objective L_chi is written as a sum of densities, not log-densities; maximizing a sum of probability densities is not a standard maximum-likelihood density-estimation criterion. The authors presumably intend log chi_omega(t,u) and log bar_chi_omega(t), and this should be stated explicitly.
[Table I] The hyperparameter is denoted lambda_gamma in the LibriSpeech block and lambda_r in the TED-LIUM3 block; this should be made consistent, and the column header 'Attention chunk size' should read 'chunk size' to avoid confusion with attention mechanisms.
[Section II-A] The remark that the forward-backward algorithm 'does not obey Bayes' rule' is unsupported and is not used in the rest of the paper; it should be removed or substantiated.
[Table I] The header 'libriSpeech' should be 'LibriSpeech'.
[Figure 1] Figure 1 is dense and the placement of the stop-gradient and the hyperparameters (e.g., lambda_gamma) is hard to read; the caption should explain the red boxes and the meaning of the arrows more explicitly.

Circularity Check

1 steps flagged · score 6.0 of 10

FoCC collapses to identity under the paper's own streaming model, making the 'actual likelihood' the conventional streaming likelihood by construction.

self definitional [Section II-B, Eqs. (10)-(12)]
"To obtain the actual likelihood, we introduce a probability ratio, namely Forward Variable Causal Compensation (FoCC): γθ(t, u) := Pθ(y0:u, zu ≤ t < zu+1|x1:e(t+1)) / Pθ(y0:u, zu ≤ t < zu+1|x1:e(t)) = Pθ(xe(t)+1:e(t+1)|x1:e(t), y0:u, zu ≤ t < zu+1) / Pθ(xe(t)+1:e(t+1)|x1:e(t))."

Under the streaming model defined by (8)-(9), the local probabilities at frame s condition only on x1:e(s). The event {y0:u, z_u ≤ t < z_{u+1}} is a prefix event over decisions up to frame t, so its probability is a product of local factors for s ≤ t, each identical whether the overall conditioning is x1:e(t) or x1:e(t+1). Hence γθ(t,u) ≡ 1 for every t,u. Substituting γ=1 into the modified recursion (11) gives exactly the conventional recursion (6) with streaming locals substituted, and the 'actual likelihood' (12) coincides with the 'deformed likelihood' from (7). The claimed gap is therefore zero by construction; the paper's assertion that the naive objective 'breaks Bayes rule' is never demonstrated, and FoCC is a definitional placeholder rather than an independent quantification.

full rationale

The paper's experimental evaluation on LibriSpeech and TED-LIUM3 is externally benchmarked, so the WER improvements are not themselves fitted inputs. However, the central theoretical contribution — FoCC as a mathematical quantification of the gap between deformed and actual likelihood — is self-definitional under the paper's own equations. The streaming model in (8)-(9) defines local probabilities at each frame using only the causal context x1:e(t). Because the forward-variable event in γ(t,u) depends only on decisions made through frame t, conditioning on the next chunk x1:e(t+1) cannot change its probability; γ is identically 1. The 'modified recursion' then reduces to the conventional streaming recursion, and the 'actual likelihood' equals the conventional streaming likelihood by construction. The paper asserts without proof that the conventional objective 'breaks Bayes rule' and provides no counterexample. The FoCCE network is trained on an independent density objective, so the method is not a simple fit of the target metric, but the claimed exact-likelihood correction is vacuous as derived. Score 6 reflects that a central prediction (the gap and its compensation) reduces by construction to the input, while the empirical results retain independent content.

Assumptions & free parameters 4 free parameters · 4 assumptions · 2 invented entities

The central claim depends on the existence of a nontrivial FoCC gap and on the ability of FoCCE to estimate it. Both are introduced by this paper and neither is independently evidenced; the only support is downstream WER changes.

free parameters (4)

lambda_gamma = 0.05 (selected among 0.01, 0.05, 0.25 for best WER)
Scaling factor for the FoCCE ratio in Eq. (13); tuned on validation WER, not derived from first principles.
lambda_chi = 0.01
Weight for the FoCCE density objective in the total loss (20); hand-set.
lambda_mod = 1
Weight for the modified likelihood term in the total loss (20); fixed.
FoCCE network parameters = 29.1M parameters for the best configuration
All FoCCE weights are fitted to the density objective (17); convergence of gamma_omega to true gamma is assumed, not demonstrated.

assumptions (4)

standard math Bayes' rule decomposition of the FoCC ratio in Eq. (10) treats acoustic chunks as random variables.
Used to rewrite P(A|x1:e(t+1))/P(A|x1:e(t)) as a density ratio; mathematically valid but requires a probabilistic model of future acoustic frames.
domain assumption The standard streaming forward recursion with the local probabilities (9) is not the exact likelihood (deformed likelihood premise).
Section II-B asserts this without proof; no counterexample is given. If false, gamma is unnecessary.
domain assumption With enough capacity, the FoCCE network omega converges to the true FoCC ratio gamma.
Unverified; the estimator is trained on a density objective, not validated against any ground-truth gamma.
domain assumption Normalizing flows with a standard Gaussian prior can represent the required audio feature densities.
Used in (15)-(17) to model arbitrary densities; no evidence of sufficient flow capacity is provided.

invented entities (2)

Forward Variable Causal Compensation (FoCC) gamma(t,u)
purpose: Ratio that compensates for the context change between e(t) and e(t+1) in the streaming forward recursion.
A new mathematical quantity introduced to define the 'actual likelihood'. It is never computed or measured; only its estimator is trained.
FoCCE network omega
purpose: Neural network estimator of FoCC using normalizing flows, trained with the objective in (17).
The estimator's convergence to the true ratio is assumed, not demonstrated, and its outputs are not validated.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Towards Maximum Likelihood Training for Transducer-based Streaming Speech Recognition." pith.science (2026). https://pith.science/paper/AA7QQVI2

@misc{pith2026241117537,
  author       = {Pith},
  title        = {Pith review of: Towards Maximum Likelihood Training for Transducer-based Streaming Speech Recognition},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AA7QQVI2}},
  note         = {Machine review of arXiv:2411.17537}
}

read the original abstract

Transducer neural networks have emerged as the mainstream approach for streaming automatic speech recognition (ASR), offering state-of-the-art performance in balancing accuracy and latency. In the conventional framework, streaming transducer models are trained to maximize the likelihood function based on non-streaming recursion rules. However, this approach leads to a mismatch between training and inference, resulting in the issue of deformed likelihood and consequently suboptimal ASR accuracy. We introduce a mathematical quantification of the gap between the actual likelihood and the deformed likelihood, namely forward variable causal compensation (FoCC). We also present its estimator, FoCCE, as a solution to estimate the exact likelihood. Through experiments on the LibriSpeech dataset, we show that FoCCE training improves the accuracy of the streaming transducers.

Figures

Figures reproduced from arXiv: 2411.17537 by the authors.

**Figure 1.** An illustration of the proposed FoCCE training. The streaming transducer network (left) and the FoCCE network (middle) respectively estimate the [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

27 extracted references · 22 canonical work pages

[1]

Improving proper noun recog- nition in end-to-end asr by customization of the mwer loss criterion,

C. Peyser, T. N. Sainath, and G. Pundak, “Improving proper noun recog- nition in end-to-end asr by customization of the mwer loss criterion,” in Proceedings of IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) . IEEE, 2020, pp. 7789–7793

work page 2020
[2]

Personalization of end-to-end speech recognition on mobile devices for named entities,

K. C. Sim, F. Beaufays, A. Benard, D. Guliani, A. Kabel, N. Khare, T. Lucassen, P. Zadrazil, H. Zhang, L. Johnsonet al., “Personalization of end-to-end speech recognition on mobile devices for named entities,” in 2019 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU). IEEE, 2019, pp. 23–30

work page 2019
[3]

Google usm: Scaling automatic speech recognition beyond 100 languages,

Y . Zhang, W. Han, J. Qin, Y . Wang, A. Bapna, Z. Chen, N. Chen, B. Li, V . Axelrod, G. Wang et al. , “Google usm: Scaling automatic speech recognition beyond 100 languages,” arXiv preprint arXiv:2303.01037 , 2023

arXiv 2023
[4]

Scaling speech technology to 1,000+ languages,

V . Pratap, A. Tjandra, B. Shi, P. Tomasello, A. Babu, S. Kundu, A. Elkahky, Z. Ni, A. Vyas, M. Fazel-Zarandi et al. , “Scaling speech technology to 1,000+ languages,” arXiv preprint arXiv:2305.13516 , 2023

arXiv 2023
[5]

A better and faster end-to- end model for streaming asr,

B. Li, A. Gulati, J. Yu, T. N. Sainath, C.-C. Chiu, A. Narayanan, S.- Y . Chang, R. Pang, Y . He, J. Qin et al. , “A better and faster end-to- end model for streaming asr,” in Proceedings of IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) . IEEE, 2021, pp. 5634–5638

work page 2021
[6]

Cascaded encoders for unifying streaming and non-streaming asr,

A. Narayanan, T. N. Sainath, R. Pang, J. Yu, C.-C. Chiu, R. Prabhavalkar, E. Variani, and T. Strohman, “Cascaded encoders for unifying streaming and non-streaming asr,” in Proceedings of IEEE International Confer- ence on Acoustics, Speech and Signal Processing (ICASSP) . IEEE, 2021, pp. 5629–5633

work page 2021
[7]

Dual-mode asr: Unify and improve streaming asr with full-context modeling,

J. Yu, W. Han, A. Gulati, C.-C. Chiu, B. Li, T. N. Sainath, Y . Wu, and R. Pang, “Dual-mode asr: Unify and improve streaming asr with full-context modeling,” in Proceedings of International Conference on Learning Representations (ICLR) , 2021

work page 2021
[8]

Deep neural networks for acoustic modeling in speech recognition: The shared views of four research groups,

G. Hinton, L. Deng, D. Yu, G. E. Dahl, A.-r. Mohamed, N. Jaitly, A. Senior, V . Vanhoucke, P. Nguyen, T. N. Sainath et al., “Deep neural networks for acoustic modeling in speech recognition: The shared views of four research groups,” IEEE Signal processing magazine , vol. 29, no. 6, pp. 82–97, 2012

2012

Show all 27 references

[9]

Connectionist temporal classification: labelling unsegmented sequence data with recur- rent neural networks,

A. Graves, S. Fern ´andez, F. Gomez, and J. Schmidhuber, “Connectionist temporal classification: labelling unsegmented sequence data with recur- rent neural networks,” in Proceedings of the International Conference on Machine learning (ICML) , 2006, pp. 369–376

2006
[10]

Sequence transduction with recurrent neural networks,

A. Graves, “Sequence transduction with recurrent neural networks,” in Representation Learning Workshop in International Coneference on Machine Learning (ICML) , 2012

2012
[11]

End-to-end attention-based large vocabulary speech recognition,

D. Bahdanau, J. Chorowski, D. Serdyuk, P. Brakel, and Y . Bengio, “End-to-end attention-based large vocabulary speech recognition,” in Proceedings of IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , 2016, pp. 4945–4949

2016
[12]

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,” in Proceedings of IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , 2016, pp. 4960–4964

2016
[13]

Cif: Continuous integrate-and-fire for end-to-end speech recognition,

L. Dong and B. Xu, “Cif: Continuous integrate-and-fire for end-to-end speech recognition,” arXiv preprint arXiv:1905.11235 , 2019

1905 arXiv
[14]

Crf-based single-stage acoustic modeling with ctc topology,

H. Xiang and Z. Ou, “Crf-based single-stage acoustic modeling with ctc topology,” in Proceedings of IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) . IEEE, 2019, pp. 5676–5680

2019
[15]

Imputer: Sequence modelling via imputation and dynamic programming,

W. Chan, C. Saharia, G. Hinton, M. Norouzi, and N. Jaitly, “Imputer: Sequence modelling via imputation and dynamic programming,” in Pro- ceedings of the International Conference on Machine learning (ICML) . PMLR, 2020, pp. 1403–1413

2020
[16]

Global normalization for streaming speech recognition in a modular framework,

E. Variani, K. Wu, M. D. Riley, D. Rybach, M. Shannon, and C. Al- lauzen, “Global normalization for streaming speech recognition in a modular framework,” Proceedings of Advances in Neural Information Processing Systems (NeurIPS) , vol. 35, pp. 4257–4269, 2022

2022
[17]

An unsupervised autoregressive model for speech representation learning,

Y .-A. Chung, W.-N. Hsu, H. Tang, and J. Glass, “An unsupervised autoregressive model for speech representation learning,” in Proceedings of Interspeech. ISCA, 2019, pp. 146–150

2019
[18]

Variational inference with normalizing flows,

D. Rezende and S. Mohamed, “Variational inference with normalizing flows,” in Proceedings of the International Conference on Machine learning (ICML). PMLR, 2015, pp. 1530–1538

2015
[19]

https://github.com/k2-fsa/icefall

k2-fsa (2021), icefall [Source Code]. https://github.com/k2-fsa/icefall

2021
[20]

Librispeech: an asr corpus based on public domain audio books,

V . Panayotov, G. Chen, D. Povey, and S. Khudanpur, “Librispeech: an asr corpus based on public domain audio books,” in Proceedings of IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2015, pp. 5206–5210

2015
[21]

Ted-lium 3: Twice as much data and corpus repartition for experiments on speaker adaptation,

F. Hernandez, V . Nguyen, S. Ghannay, N. Tomashenko, and Y . Esteve, “Ted-lium 3: Twice as much data and corpus repartition for experiments on speaker adaptation,” in Speech and Computer: 20th International Conference, SPECOM 2018, Leipzig, Germany, September 18–22, 2018, Proc...

2018
[22]

A new algorithm for data compression,

P. Gage, “A new algorithm for data compression,” C Users Journal , vol. 12, no. 2, pp. 23–38, 1994

1994
[23]

Neural machine translation of rare words with subword units,

R. Sennrich, B. Haddow, and A. Birch, “Neural machine translation of rare words with subword units,” arXiv preprint arXiv:1508.07909, 2015

2015 arXiv
[24]

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

Z. Yao, L. Guo, X. Yang, W. Kang, F. Kuang, Y . Yang, Z. Jin, L. Lin, and D. Povey, “Zipformer: A faster and better encoder for automatic speech recognition,” arXiv preprint arXiv:2310.11230 , 2023

2023 arXiv
[25]

Rnn- transducer with stateless prediction network,

M. Ghodsi, X. Liu, J. Apfel, R. Cabrera, and E. Weinstein, “Rnn- transducer with stateless prediction network,” in Proceedings of IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2020, pp. 7049–7053

2020
[26]

Made: Masked autoencoder for distribution estimation,

M. Germain, K. Gregor, I. Murray, and H. Larochelle, “Made: Masked autoencoder for distribution estimation,” in Proceedings of the Interna- tional Conference on Machine learning (ICML) . PMLR, 2015, pp. 881–889

2015
[27]

Specaugment: A simple data augmentation 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 augmentation method for automatic speech recognition,” Proceedings of Interspeech , 2019

2019

Pith tools

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

[1] [1]

Improving proper noun recog- nition in end-to-end asr by customization of the mwer loss criterion,

C. Peyser, T. N. Sainath, and G. Pundak, “Improving proper noun recog- nition in end-to-end asr by customization of the mwer loss criterion,” in Proceedings of IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) . IEEE, 2020, pp. 7789–7793

work page 2020

[2] [2]

Personalization of end-to-end speech recognition on mobile devices for named entities,

K. C. Sim, F. Beaufays, A. Benard, D. Guliani, A. Kabel, N. Khare, T. Lucassen, P. Zadrazil, H. Zhang, L. Johnsonet al., “Personalization of end-to-end speech recognition on mobile devices for named entities,” in 2019 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU). IEEE, 2019, pp. 23–30

work page 2019

[3] [3]

Google usm: Scaling automatic speech recognition beyond 100 languages,

Y . Zhang, W. Han, J. Qin, Y . Wang, A. Bapna, Z. Chen, N. Chen, B. Li, V . Axelrod, G. Wang et al. , “Google usm: Scaling automatic speech recognition beyond 100 languages,” arXiv preprint arXiv:2303.01037 , 2023

arXiv 2023

[4] [4]

Scaling speech technology to 1,000+ languages,

V . Pratap, A. Tjandra, B. Shi, P. Tomasello, A. Babu, S. Kundu, A. Elkahky, Z. Ni, A. Vyas, M. Fazel-Zarandi et al. , “Scaling speech technology to 1,000+ languages,” arXiv preprint arXiv:2305.13516 , 2023

arXiv 2023

[5] [5]

A better and faster end-to- end model for streaming asr,

B. Li, A. Gulati, J. Yu, T. N. Sainath, C.-C. Chiu, A. Narayanan, S.- Y . Chang, R. Pang, Y . He, J. Qin et al. , “A better and faster end-to- end model for streaming asr,” in Proceedings of IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) . IEEE, 2021, pp. 5634–5638

work page 2021

[6] [6]

Cascaded encoders for unifying streaming and non-streaming asr,

A. Narayanan, T. N. Sainath, R. Pang, J. Yu, C.-C. Chiu, R. Prabhavalkar, E. Variani, and T. Strohman, “Cascaded encoders for unifying streaming and non-streaming asr,” in Proceedings of IEEE International Confer- ence on Acoustics, Speech and Signal Processing (ICASSP) . IEEE, 2021, pp. 5629–5633

work page 2021

[7] [7]

Dual-mode asr: Unify and improve streaming asr with full-context modeling,

J. Yu, W. Han, A. Gulati, C.-C. Chiu, B. Li, T. N. Sainath, Y . Wu, and R. Pang, “Dual-mode asr: Unify and improve streaming asr with full-context modeling,” in Proceedings of International Conference on Learning Representations (ICLR) , 2021

work page 2021

[8] [8]

Deep neural networks for acoustic modeling in speech recognition: The shared views of four research groups,

G. Hinton, L. Deng, D. Yu, G. E. Dahl, A.-r. Mohamed, N. Jaitly, A. Senior, V . Vanhoucke, P. Nguyen, T. N. Sainath et al., “Deep neural networks for acoustic modeling in speech recognition: The shared views of four research groups,” IEEE Signal processing magazine , vol. 29, no. 6, pp. 82–97, 2012

2012