Transducer-Llama: Integrating LLMs into Streamable Transducer-based Speech Recognition

Jinxi Guo; Keqi Deng; Mike Seltzer; Niko Moritz; Ozlem Kalinli; Philip C. Woodland; Yingyi Ma

REVIEW 3 major objections 6 minor 36 references

Transducer-Llama: Integrating LLMs into Streamable Transducer-based Speech Recognition

T0 review · 3 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read A streaming speech recognizer can carry a frozen Llama inside it, cutting word error by 17 percent relative to a strong factorized-transducer baseline.

desk verdict Solid engineering recipe for streaming ASR with a frozen LLM, but the headline Llama3 result lacks an ablation isolating the weak-to-strong swap from MWER. read the letter →

arxiv 2412.16464 v1 pith:AWPKHAVB submitted 2024-12-21 cs.CL eess.AS

Keqi Deng , Jinxi Guo , Yingyi Ma , Niko Moritz , Philip C. Woodland , Ozlem Kalinli , Mike Seltzer This is my paper · ORCID

classification cs.CLeess.AS

keywords streamingspeechrecognitionfactorizedtransducerlargelanguagemodelsweak-to-strongLMswapminimumworderrorratetrainingvocabularyadaptationLibriRNN-T

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

Transducer-Llama claims that a frozen large language model can act as the internal language model of a streaming factorized transducer, improving speech recognition accuracy without sacrificing online operation. The recipe has three parts: shrink the LLM tokenizer to a 5,000-token ASR vocabulary, train the encoder with a weak bigram-like predictor, then swap in a frozen Llama and fine-tune with a word-error-rate loss. On LibriSpeech this yields 2.47% WER on test-clean and 6.53% on test-other, a 17% relative reduction over the factorized-transducer baseline and 32% over an RNN-T baseline. On the multilingual LibriSpeech corpus the same strategy reduces average WER from 10.73% to 9.24%. The point is that streaming ASR and large text LMs are not incompatible.

What carries the argument

The load-bearing mechanism is the weak-to-strong LM swap, combined with vocabulary adaptation and ILM-aware MWER training. A small stateless non-blank predictor (effectively a bigram token predictor) is used during RNN-T loss training; once the encoder and blank predictor have converged, the small predictor is replaced by a frozen LLM whose probabilities enter the factorized-transducer decoding equation. The vocabulary adaptation step initializes new embedding and output layers for a compact 5,000-token ASR vocabulary from the LLM's original 128,000-token vocabulary, allowing the LLM's transformer layers to stay fixed. MWER training then optimizes the integration of the LLM with the word-level edit distance, which is what makes the swap effective.

What would settle it

Retrain the full pipeline but swap in a randomly initialized Llama3-sized transformer with fixed weights as the non-blank predictor before MWER; if WER stays near the Llama3 result, the gain does not come from the LLM's pre-trained text knowledge, and if WER collapses, the language-model prior is essential to the reported improvement.

Watch

Extended reading notes

Core claim

The paper establishes that a text-only LLM, frozen during ASR training, can serve as the non-blank predictor of a streaming factorized transducer and materially improve accuracy. The critical finding is that simply training the transducer with a strong LLM predictor under the RNN-T loss gives only small gains, whereas training first with a weak predictor and then swapping in the LLM for decoding, followed by minimum word error rate (MWER) fine-tuning, gives large gains. With Llama3 as the swapped-in predictor, the model reaches 2.47% WER on LibriSpeech test-clean and 6.53% on test-other, beating both the strong factorized-transducer baseline and a fully fine-tuned smaller Llama2. The authors interpret this as the encoder being forced to rely on acoustic information during training, so the strong LM is used as a genuine complement rather than a crutch.

Load-bearing premise

The approach assumes that an encoder trained with a weak bigram-like predictor learns acoustic representations that stay compatible with a much stronger language model when that model is swapped in at test time.

Editorial extensions

If this is right

Streaming deployment can use the full predictive power of an 8B-parameter frozen LLM while keeping the 160 ms segment-based online decoding used by the Emformer encoder.
Because the LLM transformer stack stays frozen, a trained Transducer-Llama can be reused for other downstream tasks without retraining the speech model, and the ASR-specific parameters remain compact.
Vocabulary adaptation cuts the output dimension roughly 26 times, giving an 8x faster training step and lower WER than running the LLM's native tokenizer.
The weak-to-strong swap plus MWER beats both training directly with the LLM under RNN-T loss and fully fine-tuning a smaller LLM, as shown in the ablation table.
Multilingual gains replicate on French, Italian, and Dutch in the MLS corpus, so the approach is not English-specific.

Reading between the lines

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

If the swap is general, the same recipe should work with other frozen text LMs and other streaming encoders, since the central assumption is only that the encoder learns acoustic representations that are largely independent of the predictor used during RNN-T training.
The paper's comparison to offline speech-LLM baselines suggests that a streaming Transducer-Llama may already be competitive with non-streaming LLM-prompted ASR; a latency-controlled head-to-head benchmark would make that trade-off explicit.
A stress test on acoustically noisy or out-of-domain audio, where LM priors can mislead, would clarify whether MWER's gains come from better acoustic encoding or from the language-model prior injected at decoding.
The weak-to-strong recipe resembles classical sequence-discriminative training with an external LM, so the same strategy may transfer to other end-to-end architectures that factor out an internal LM component.

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

Colleague,

This paper is a practical engineering contribution, and honestly a good one. The idea is to take the Factorized Transducer from [8], freeze a Llama model as the non-blank predictor, shrink its vocabulary to the ASR token set via embedding/output-layer initialization, train the encoder with a stateless predictor, then swap in the LLM and run MWER. The reported 17% WERR over a strong FT baseline and 32% over RNN-T on LibriSpeech are consistent with the ablations. The vocabulary adaptation table (Table III) is a nice result on its own: 8x training speedup and slightly better WER than the native 128k Llama tokenizer.

What's genuinely new is the combination: vocabulary transfer for LLM-in-transducer, the weak-to-strong swap, and MWER after the swap. Each piece has precursors, but the package is new and the ablations mostly support the claims. The multilingual MLS results also hold up, and the offline comparison against a published speech LLM is a useful sanity check.

The soft spots are real but not disqualifying. First, no code, no data, no significance tests. The tables are clean, but these are single runs, and ASR numbers on LibriSpeech can move by a couple tenths with seed changes. Second, alpha and beta are copied from [8] with no sensitivity analysis. Minor, since they follow the prior work. Third, the weak-to-strong swap is justified heuristically. The paper says it prevents the encoder from relying on the strong LM, but there is no diagnostic—no analysis of encoder representations or training dynamics. That's a gap, but the ablation rows do show the swap helps across several predictor pairs, so the heuristic has empirical support.

The stress-test concern about Table II is worth flagging: the swap ablations use Llama2 and FFT Llama2, not Llama3. For Llama3, the improvement from 2.76 (no swap, no MWER) to 2.47 (swap + MWER) is not broken down, so strictly speaking the headline number could be driven by MWER alone rather than the swap. That doesn't sink the paper—the full recipe works—but it does mean the paper overstates slightly when it credits the swap for the Llama3 result. A careful referee should ask for one extra row: Llama3 with swap but no MWER, or with MWER but no swap.

Who is this for? Engineers building streaming ASR systems, and researchers working on internal LM fusion. It's a serious empirical paper, not a theoretical one. I'd send it to peer review. The missing Llama3 ablation is fixable in revision.

Recommendation: engage with it. If the authors add the missing ablation and release code, this becomes a reference point for frozen-LLM streaming ASR.

Referee Report

3 major / 6 minor

Summary. This paper proposes Transducer-Llama, a factorized transducer (FT) architecture in which a frozen LLM serves as the non-blank predictor, together with a vocabulary adaptation scheme, a weak-to-strong LM swap, and MWER training. On LibriSpeech, the system with Llama3-8B reports 2.47% WER on test-clean and 6.53% on test-other, which the paper translates into a 17% relative WER reduction over the strongest FT baseline and 32% over an RNN-T baseline. The paper also reports multilingual LibriSpeech (MLS) results and an offline comparison against a speech LLM. The central claim is that frozen LLMs can be integrated into a streaming transducer while improving accuracy and preserving streaming capability.

Significance. If the reported results hold, the paper offers a useful recipe for inserting frozen LLMs into an already-streamable transducer without retraining the encoder with the LLM, which is a practical contribution to streaming ASR. The strengths of the paper are its consistent held-out evaluation on standard LibriSpeech/MLS test sets, the ablation table (Table II) that isolates the weak-to-strong swap and MWER for Llama2-based predictors, and the vocabulary adaptation result showing a large training-speed advantage (Table III). The main empirical claim is large in magnitude (17% WERR), and the architecture is plausible. However, the key mechanism is not directly ablated for the exact model configuration behind the headline number (Llama3-8B), and the paper provides no latency or uncertainty analysis, so the headline result is not yet fully pinned down.

major comments (3)

[Section IV-C2 / Table II and Table III] The weak-to-strong LM swap is not ablated for Llama3-8B, the model used for the headline 17% WERR. Table II covers only LSTM, stateless, Llama2-0.5B, and FFT Llama2 as test-time predictors; Table III reports Llama3 without the swap and without MWER (2.76/7.36), while Table I reports Llama3 with both the swap and MWER (2.47/6.53). There is therefore no Llama3 row that isolates the swap from MWER, and the 10.5% test-clean gain from 2.76 to 2.47 could be wholly attributable to MWER. Since Llama3 is 16x larger than Llama2 and uses a 128k tokenizer, the assumption that an encoder trained with a bigram predictor remains compatible with Llama3 is nontrivial. Please add a Llama3 no-swap row (Llama3 used at both training and test, with and without MWER) or explicitly restrict the swap claim to the Llama2 results.
[Section IV-C1 / Table I] The WER comparisons are based on a single decoding pass with no confidence intervals or significance tests. Several adjacent rows in Table I differ by only 0.07 WER (e.g., FFT Llama2 vs Llama3 on test-clean), and the text's conclusion that 'as more powerful LMs are used, the WER continues to decrease' depends on these small differences. Repeated runs or a test-set significance test would support this monotonicity claim; at minimum, the absence of such statistics should be acknowledged in the experimental section.
[Section IV-B / IV-C] The paper's title and abstract emphasize streaming, but no latency, real-time factor, or memory measurements are reported for Transducer-Llama. The streaming property is architectural (the encoder uses Emformer/Conformer with chunk masks), but with an 8B-parameter LLM as the autoregressive predictor, practical streaming feasibility is not self-evident. Please report an RTF or latency figure for the streaming models, or explicitly state that streaming capability is inherited from the FT architecture and was not evaluated empirically.

minor comments (6)

[Table II] The symbols '%' and '!' in the MWER column are not defined in the caption; please state explicitly that '%' denotes RNN-T loss only and '!' denotes MWER training.
[Eq. (1)] 'hyper-perameters' should be 'hyper-parameters'.
[Section IV-A] '5.3m total batch size' should be '5.3M' for consistency with the model-size notation, and '0.2k hrs' / '1.1k hrs' are awkward units; consider writing the durations in hours or thousands of hours consistently.
[Table III] 'Tokenize' should be 'Tokenizer', and the meaning of the 'Train Speed' value (8.05) is not defined; state what it is relative to and in what units.
[Section IV-C4] The comparison with the offline Speech LLM [14] in Table IV uses a different encoder and offline decoding; the text acknowledges this, but a sentence clarifying that the comparison is indicative rather than a controlled ablation would help.
[General] No code, model checkpoints, or data splits are released, which limits reproducibility; a reproducibility statement would be welcome.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the reported WER gains are held-out evaluations of a trained system, and the cited prior work is independent published building blocks rather than a self-referential derivation.

full rationale

The derivation chain in Transducer-Llama is empirical rather than formal: the central claims are WER numbers on held-out LibriSpeech and MLS test sets (Tables I and IV), produced by training an FT-style encoder with a stateless predictor, swapping in a frozen Llama model, and applying MWER training. No equation in the paper defines the predicted improvement in terms of a fitted parameter: α and β are fixed at 0.6 following the published FT paper [8], the ASR vocabulary (5k tokens) is specified before training, and the test sets are not used in training or hyperparameter selection. The load-bearing components borrowed from same-group prior work—the factorized transducer structure, ILM loss, and MWER loss from [8] and [20]—are externally published (ICASSP/Interspeech) and do not themselves assert the current paper's claimed LLM-integration result; reusing them is standard engineering, not circularity. The weak-to-strong swap and MWER are evaluated by ablations (Table II) on held-out WER, so the comparison is self-contained. The absence of a Llama3-only ablation isolating the swap from MWER is a reporting/attribution weakness, not a circular reduction: the headline number is still a measured test-set result rather than a quantity constructed from its own inputs.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central claim rests on a borrowed transducer factorization, a frozen-LLM adaptation heuristic, and an empirical training recipe. No free parameters are fit to the test set, but alpha/beta and vocabulary size are fixed design choices, and the swap strategy is an unproved training-dynamics assumption. No new physical or conceptual entities are postulated.

free parameters (2)

alpha and beta in Eq. 1 (LM fusion weights) = 0.6
Borrowed from prior FT work [8], fixed for all LLM predictors, and no sensitivity analysis is reported.
ASR vocabulary size = 5000
A design choice for LibriSpeech and MLS; claimed to improve speed and WER versus the LLM vocabulary, but 5000 itself is not optimized.

assumptions (4)

domain assumption The factorized transducer score combination in Eq. 1, with the LLM probability treated as logP_ilm, is a valid way to merge acoustic and LLM scores.
The decoding and MWER training rest on Eq. 1 inherited from [8]; the paper applies it to LLM predictors without modification.
domain assumption Frozen LLM transformer layers retain useful language knowledge after the embedding and output layers are replaced via vocabulary adaptation.
Sec III-A fixes the LLM layers and trains only new embedding and output matrices; the paper provides no analysis of how much knowledge is lost.
domain assumption Averaging LLM subword embeddings is a sufficient initialization for ASR tokens not present in the LLM vocabulary.
Sec III-A initializes out-of-vocabulary tokens as the average of corresponding LLM embeddings, or randomly if undecomposable; this is an unverified heuristic.
ad hoc to paper The weak-to-strong LM swap does not require the encoder to be retrained with the LLM under RNN-T loss.
Sec III-B/C swaps predictors after convergence and uses only MWER to adjust integration; the success depends on this training-dynamics assumption rather than on a proof.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Transducer-Llama: Integrating LLMs into Streamable Transducer-based Speech Recognition." pith.science (2026). https://pith.science/paper/AWPKHAVB

@misc{pith2026241216464,
  author       = {Pith},
  title        = {Pith review of: Transducer-Llama: Integrating LLMs into Streamable Transducer-based Speech Recognition},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AWPKHAVB}},
  note         = {Machine review of arXiv:2412.16464}
}

read the original abstract

While large language models (LLMs) have been applied to automatic speech recognition (ASR), the task of making the model streamable remains a challenge. This paper proposes a novel model architecture, Transducer-Llama, that integrates LLMs into a Factorized Transducer (FT) model, naturally enabling streaming capabilities. Furthermore, given that the large vocabulary of LLMs can cause data sparsity issue and increased training costs for spoken language systems, this paper introduces an efficient vocabulary adaptation technique to align LLMs with speech system vocabularies. The results show that directly optimizing the FT model with a strong pre-trained LLM-based predictor using the RNN-T loss yields some but limited improvements over a smaller pre-trained LM predictor. Therefore, this paper proposes a weak-to-strong LM swap strategy, using a weak LM predictor during RNN-T loss training and then replacing it with a strong LLM. After LM replacement, the minimum word error rate (MWER) loss is employed to finetune the integration of the LLM predictor with the Transducer-Llama model. Experiments on the LibriSpeech and large-scale multi-lingual LibriSpeech corpora show that the proposed streaming Transducer-Llama approach gave a 17% relative WER reduction (WERR) over a strong FT baseline and a 32% WERR over an RNN-T baseline.

Figures

Figures reproduced from arXiv: 2412.16464 by the authors.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

36 extracted references · 28 canonical work pages

[8]

Effective internal language model training and fusion for factorized transducer model,

J. Guo, N. Moritz, Y . Ma, F. Seide, C. Wu, J. Mahadeokar, O. Kalinli, C. Fuegen, and M. Seltzer, “Effective internal language model training and fusion for factorized transducer model,” in Proc. ICASSP, 2024

work page 2024
[1]

Connection- ist temporal classification: Labelling unsegmented sequence data with recurrent neural networks,

A. Graves, S. Fern ´andez, F. J. Gomez, and J. Schmidhuber, “Connection- ist temporal classification: Labelling unsegmented sequence data with recurrent neural networks,” in Proc. ICML, 2006

work page 2006
[2]

Sequence transduction with recurrent neural networks,

A. Graves, “Sequence transduction with recurrent neural networks,” ArXiv, vol. abs/1211.3711, 2012

arXiv 2012
[3]

Label-synchronous neural transducer for adaptable online E2E speech recognition,

K. Deng and P. C. Woodland, “Label-synchronous neural transducer for adaptable online E2E speech recognition,” IEEE/ACM Transactions on Audio, Speech, and Language Processing , vol. 32, pp. 3507–3516, 2024

work page 2024
[4]

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. ICASSP, 2020

work page 2020
[5]

Factorized neural transducer for efficient language model adaptation,

X. Chen, Z. Meng, S. Parthasarathy, and J. Li, “Factorized neural transducer for efficient language model adaptation,” in Proc. ICASSP , 2022

work page 2022
[6]

Decoupled structure for improved adaptability of end-to-end models,

K. Deng and P. C. Woodland, “Decoupled structure for improved adaptability of end-to-end models,” Speech Communication , vol. 163, p. 103109, 2024

work page 2024
[7]

Modular hybrid autoregressive transducer,

Z. Meng, T. Chen, R. Prabhavalkar, Y . Zhang, G. Wang, K. Audhkhasi, J. Emond, T. Strohman, B. Ramabhadran, W. R. Huang, et al., “Modular hybrid autoregressive transducer,” in Proc. SLT, 2023

work page 2023

Show all 36 references

[9]

Fast and accurate factorized neural transducer for text adaption of end-to-end speech recognition models,

R. Zhao, J. Xue, P. Parthasarathy, V . Miljanic, and J. Li, “Fast and accurate factorized neural transducer for text adaption of end-to-end speech recognition models,” in Proc. ICASSP, 2023

2023
[10]

Language models are few-shot learners,

T. Brown, B. Mann, N. Ryder, M. Subbiah, J. D. Kaplan, P. Dhariwal, A. Neelakantan, P. Shyam, G. Sastry, A. Askell, et al. , “Language models are few-shot learners,” in Proc. NeurIPS, 2020

2020
[11]

LLaMa 2: Open foundation and fine-tuned chat models,

H. Touvron, L. Martin, K. Stone, P. Albert, A. Almahairi, Y . Babaei, N. Bashlykov, S. Batra, P. Bhargava, S. Bhosale, et al., “LLaMa 2: Open foundation and fine-tuned chat models,” 2023

2023
[12]

The Llama 3 herd of models,

A. Dubey, A. Jauhri, A. Pandey, A. Kadian, A. Al-Dahle, A. Letman, A. Mathur, A. Schelten, A. Yang, et al., “The Llama 3 herd of models,” arXiv preprint arXiv:2407.21783 , 2024

2024 arXiv
[13]

Training language models to follow instructions with human feedback,

L. Ouyang, J. Wu, X. Jiang, D. Almeida, C. Wainwright, P. Mishkin, C. Zhang, S. Agarwal, K. Slama, A. Ray, et al. , “Training language models to follow instructions with human feedback,” Proc. NeurIPS , 2022

2022
[14]

Prompting large language models with speech recognition abilities,

Y . Fathullah, C. Wu, E. Lakomkin, J. Jia, Y . Shangguan, K. Li, J. Guo, W. Xiong, J. Mahadeokar, O. Kalinli, C. Fuegen, and M. Seltzer, “Prompting large language models with speech recognition abilities,” in Proc. ICASSP, 2024

2024
[15]

Wav2Prompt: End-to-end speech prompt generation and tuning for llm in zero and few-shot learning,

K. Deng, G. Sun, and P. C. Woodland, “Wav2Prompt: End-to-end speech prompt generation and tuning for llm in zero and few-shot learning,” arXiv preprint arXiv:2406.00522 , 2024

2024 arXiv
[16]

BESTOW: Efficient and streamable speech language model with the best of two worlds in GPT and T5,

Z. Chen, H. Huang, O. Hrinchuk, K. C. Puvvada, N. R. Koluguri, P. ˙Zelasko, J. Balam, and B. Ginsburg, “BESTOW: Efficient and streamable speech language model with the best of two worlds in GPT and T5,” arXiv preprint arXiv:2406.19954 , 2024

2024 arXiv
[17]

Zero-shot tokenizer transfer,

B. Minixhofer, E. M. Ponti, and I. Vuli ´c, “Zero-shot tokenizer transfer,” arXiv preprint arXiv:2405.07883 , 2024

2024
[18]

Improving non-autoregressive end-to-end speech recognition with pre- trained acoustic and language models,

K. Deng, Z. Yang, S. Watanabe, Y . Higuchi, G. Cheng, and P. Zhang, “Improving non-autoregressive end-to-end speech recognition with pre- trained acoustic and language models,” in Proc. ICASSP, 2022

2022
[19]

Getting the most out of your tokenizer for pre-training and domain adaptation,

G. Dagan, G. Synnaeve, and B. Rozi `ere, “Getting the most out of your tokenizer for pre-training and domain adaptation,” arXiv preprint arXiv:2402.01035, 2024

2024 arXiv
[20]

Efficient minimum word error rate training of RNN- transducer for end-to-end speech recognition,

J. Guo, G. Tiwari, J. Droppo, M. V . Segbroeck, C.-W. Huang, A. Stolcke, and R. Maas, “Efficient minimum word error rate training of RNN- transducer for end-to-end speech recognition,” in Proc. Interspeech , 2020

2020
[21]

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 Proc. ICASSP , 2015

2015
[22]

MLS: A large-scale multilingual dataset for speech research,

V . Pratap, Q. Xu, A. Sriram, G. Synnaeve, and R. Collobert, “MLS: A large-scale multilingual dataset for speech research,” in Proc. Inter- speech, 2020

2020
[23]

Speech ReaLLM – real-time streaming speech recognition with multimodal llms by teaching the flow of time,

F. Seide, M. Doulaty, Y . Shi, Y . Gaur, J. Jia, and C. Wu, “Speech ReaLLM – real-time streaming speech recognition with multimodal llms by teaching the flow of time,” arXiv preprint arXiv:2406.09569 , 2024

2024 arXiv
[24]

Decoder-only architecture for streaming end-to-end speech recogni- tion,

E. Tsunoo, H. Futami, Y . Kashiwagi, S. Arora, and S. Watanabe, “Decoder-only architecture for streaming end-to-end speech recogni- tion,” arXiv preprint arXiv:2406.16107 , 2024

2024 arXiv
[25]

SimulMT to SimulST: Adapting simultaneous text translation to end-to-end simultaneous speech trans- lation,

X. Ma, J. M. Pino, and P. Koehn, “SimulMT to SimulST: Adapting simultaneous text translation to end-to-end simultaneous speech trans- lation,” in Proc. AACL/IJCNLP, 2020

2020
[26]

Label-synchronous neural transducer for E2E simultaneous speech translation,

K. Deng and P. C. Woodland, “Label-synchronous neural transducer for E2E simultaneous speech translation,” in Proc. ACL, 2024

2024
[27]

FOCUS: effective embedding initializa- tion for monolingual specialization of multilingual models,

K. Dobler and G. de Melo, “FOCUS: effective embedding initializa- tion for monolingual specialization of multilingual models,” in Proc. EMNLP, 2023

2023
[28]

Fast vocabulary transfer for language model compression,

L. Gee, A. Zugarini, L. Rigutini, and P. Torroni, “Fast vocabulary transfer for language model compression,” in Proc. EMNLP (Industry Track), 2022

2022
[29]

WECHSEL: effective initialization of subword embeddings for cross-lingual transfer of mono- lingual language models,

B. Minixhofer, F. Paischer, and N. Rekabsaz, “WECHSEL: effective initialization of subword embeddings for cross-lingual transfer of mono- lingual language models,” in Proc. NAACL-HLT, 2022

2022
[30]

Model reprogramming: Resource-efficient cross-domain machine learning,

P.-Y . Chen, “Model reprogramming: Resource-efficient cross-domain machine learning,” in Proc. AAAI, 2024

2024
[31]

Sequence- discriminative training of deep neural networks,

K. Vesel `y, A. Ghoshal, L. Burget, and D. Povey, “Sequence- discriminative training of deep neural networks,” in Proc. Interspeech, 2013

2013
[32]

Minimum phone error and I-smoothing for improved discriminative training,

D. Povey and P. Woodland, “Minimum phone error and I-smoothing for improved discriminative training,” in Proc. ICASSP, 2002

2002
[33]

Interdependence of language models and discriminative training,

R. Schl ¨uter, B. M ¨uller, F. Wessel, and H. Ney, “Interdependence of language models and discriminative training,” in Proc. ASRU, 1999

1999
[34]

Component fusion: Learning replaceable language model component for end-to-end speech recognition system,

C. Shan, C. Weng, G. Wang, D. Su, M. Luo, D. Yu, and L. Xie, “Component fusion: Learning replaceable language model component for end-to-end speech recognition system,” in Proc. ICASSP, 2019

2019
[35]

Emformer: Efficient memory transformer based acoustic model for low latency streaming speech recognition,

Y . Shi, Y . Wang, C. Wu, C.-F. Yeh, J. Chan, F. Zhang, D. Le, and M. Seltzer, “Emformer: Efficient memory transformer based acoustic model for low latency streaming speech recognition,” in Proc. ICASSP, 2021

2021
[36]

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

2020

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.

[1] [8]

Effective internal language model training and fusion for factorized transducer model,

J. Guo, N. Moritz, Y . Ma, F. Seide, C. Wu, J. Mahadeokar, O. Kalinli, C. Fuegen, and M. Seltzer, “Effective internal language model training and fusion for factorized transducer model,” in Proc. ICASSP, 2024

work page 2024

[2] [1]

Connection- ist temporal classification: Labelling unsegmented sequence data with recurrent neural networks,

A. Graves, S. Fern ´andez, F. J. Gomez, and J. Schmidhuber, “Connection- ist temporal classification: Labelling unsegmented sequence data with recurrent neural networks,” in Proc. ICML, 2006

work page 2006

[3] [2]

Sequence transduction with recurrent neural networks,

A. Graves, “Sequence transduction with recurrent neural networks,” ArXiv, vol. abs/1211.3711, 2012

arXiv 2012

[4] [3]

Label-synchronous neural transducer for adaptable online E2E speech recognition,

K. Deng and P. C. Woodland, “Label-synchronous neural transducer for adaptable online E2E speech recognition,” IEEE/ACM Transactions on Audio, Speech, and Language Processing , vol. 32, pp. 3507–3516, 2024

work page 2024

[5] [4]

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. ICASSP, 2020

work page 2020

[6] [5]

Factorized neural transducer for efficient language model adaptation,

X. Chen, Z. Meng, S. Parthasarathy, and J. Li, “Factorized neural transducer for efficient language model adaptation,” in Proc. ICASSP , 2022

work page 2022

[7] [6]

Decoupled structure for improved adaptability of end-to-end models,

K. Deng and P. C. Woodland, “Decoupled structure for improved adaptability of end-to-end models,” Speech Communication , vol. 163, p. 103109, 2024

work page 2024

[8] [7]

Modular hybrid autoregressive transducer,

Z. Meng, T. Chen, R. Prabhavalkar, Y . Zhang, G. Wang, K. Audhkhasi, J. Emond, T. Strohman, B. Ramabhadran, W. R. Huang, et al., “Modular hybrid autoregressive transducer,” in Proc. SLT, 2023

work page 2023