Pith. sign in

REVIEW 4 major objections 6 minor 29 references

Transcribing and Translating, Fast and Slow: Joint Speech Translation and Recognition

T0 review · 4 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 JSTAR couples fast and slow encoder paths to stream both speech recognition and translation from a single transducer model.

desk verdict A genuinely new transducer-based streaming MT model and a well-engineered joint ASR/ST system, but the superiority claim needs better statistics and a fairer baseline. read the letter →

arxiv 2412.15415 v1 pith:QJ6GJ3WJ submitted 2024-12-19 eess.AS cs.CL

Niko Moritz , Ruiming Xie , Yashesh Gaur , Ke Li , Simone Merello , Zeeshan Ahmed , Frank Seide , Christian Fuegen This is my paper · ORCID
classification eess.AScs.CL
keywords JSTARspeechtranslationautomaticrecognitionRNN-Tstreamingfast-slowcascadedencodermulti-talkersmartglasses
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper proposes JSTAR, a single streaming model that simultaneously produces an automatic transcript and a translation of bilingual conversational audio, aimed at smart-glasses use. It claims that equipping a transducer with two separate predictor-joiner pairs, one on a fast low-latency encoder path for ASR and one on a slow wider-context encoder path for ST, allows one model to handle both tasks without one degrading the other. It also claims that a transducer trained on text alone can serve as a streaming machine-translation model, and that using it to initialize the translation side improves final BLEU scores. If these claims hold, streaming speech-translation systems can drop the traditional two-stage ASR-then-MT pipeline and still match or beat it in quality while cutting latency.

What carries the argument

The load-bearing mechanism is the fast-slow cascaded encoder architecture, in which a fast encoder (20 streaming Conformer layers, 300 ms chunks, about 210 ms average latency) feeds a slow encoder (10 layers, 600 ms chunks, about 360 ms latency). On top sit two independent predictor-joiner pairs: the ASR pair consumes the fast encoder output, and the ST pair consumes the slow encoder output, trained jointly with the multi-objective loss $L = L_{\text{st}} + 0.5 \cdot L_{\text{asr}}$ under the RNN-T alignment loss. A second mechanism is the multi-channel direction-aware front-end, which uses 13 NLCMV beamformers covering 12 horizontal directions plus the wearer's mouth so the model can attribute speech to SELF and OTHER and output serialized multi-talker transcripts with speaker tokens. A third mechanism is the newly introduced transducer-based streaming MT model, which replaces the audio front-end with character embeddings and is used to initialize the slow encoder, the ST joiner and predictor, and the ASR predictor.

What would settle it

Train the same transducer-based streaming MT model on a language pair with consistently non-monotonic word order, such as English-to-Japanese with verb-final structure, at the same 5+1 character context; if BLEU collapses relative to an attention-based encoder-decoder baseline and cannot be recovered by increasing right context, the claimed alignment-delay capacity fails. A more direct check is to extract the learned alignment paths from the transducer lattice on the Spanish-English test set and verify that reordered tokens appear with non-zero probability rather than being approximated monotonically.

Watch

Extended reading notes

Core claim

The central claim is that a single RNN-Transducer model with a fast-slow cascaded encoder and task-specific joiner and predictor pairs can simultaneously stream ASR and speech translation, outperforming a strong cascaded ASR plus non-streaming MT baseline. On the MC-FLEURS multi-channel test set, JSTAR reports up to 20.52 BLEU for Spanish-to-English and 18.72 for English-to-Spanish, beating the cascade by 0.9 points on English-to-Spanish and trailing by 0.3 points on Spanish-to-English; on the real-world RealConv set it gains roughly 2 BLEU points on average. The paper also reports a first-finalized-token latency of 3.3 seconds versus 7.1 seconds for the cascade, a 3.8-second reduction, with similar last-token latency. A supporting discovery is that a transducer-based streaming MT model (text-only, character input) reaches BLEU scores within 2.1 points of a transformer attention-based model on Spanish-to-English and roughly equal on English-to-Spanish, which the paper interprets as evidence that RNN-T can represent non-monotonic word order by delaying output tokens.

Load-bearing premise

The entire translation side rests on the assumption that an RNN-T encoder plus predictor can learn to delay output tokens so that non-monotonic translations can be represented in the transducer alignment, which the paper states but does not prove.

Editorial extensions

If this is right

  • One deployed model can serve both a live transcript and a live translation from a single streaming decoder, removing the separate MT beam-search stage and its token-finalization latency.
  • Because the joint objective optimizes ASR and ST simultaneously, improvements in acoustic modeling or front-end beamforming should benefit translation quality at no additional inference cost.
  • Initializing JSTAR from a text-only streaming transducer MT model offers a viable pre-training recipe when parallel speech-translation data is scarce.
  • The direction-aware multi-channel front-end lets the same joint model handle multi-talker conversations with speaker-attributed output, extending the approach to the smart-glasses use case.

Reading between the lines

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

  • Editorial inference: the paper asserts, but does not prove, that an RNN-T encoder plus predictor can delay outputs to cover non-monotonic word order; a systematic stress test on language pairs with large word-order divergence (such as SOV versus VSO) would reveal whether the restricted 5+1 character context in the MT experiments is the real bottleneck.
  • Editorial inference: if the fast-slow separation is what protects ASR quality while giving ST more context, the same architecture should transfer to other multi-output streaming tasks, such as simultaneous speech-to-speech translation or punctuation-plus-translation, by adding another joiner and predictor on the slow encoder.
  • Editorial inference: the reported 3.8-second latency advantage depends on the cascaded system's sentence-end finalization policy; a cascade tuned for earlier partial translations might narrow the gap, so a full latency-quality curve would be the fairer comparison.
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.

Referee Report

4 major / 6 minor

Summary. The paper proposes JSTAR, a transducer-based joint speech recognition and translation (ASR/ST) model. Audio from a smart-glasses microphone array is processed by 13 NLCMV beamformers, a CNN front-end, and then a fast streaming conformer encoder followed by a slow streaming conformer encoder. A dedicated ASR predictor/joiner operates on the fast encoder and a dedicated ST predictor/joiner operates on the slow encoder. Training minimizes Eq. (2), a weighted sum of ASR and ST RNN-T losses, with serialized output training for overlapping speakers and direction labels for speaker attribution. The authors also propose a text-based RNN-T streaming MT model, use it to initialize part of JSTAR, and evaluate JSTAR against a cascaded ASR+MT baseline on simulated MC-FLEURS and in-house RealConv data, reporting WER, BLEU, and P50 token latency.

Significance. The architectural idea of using separate predictor/joiner pairs on different depths of a fast-slow cascaded encoder is clearly presented and is a plausible way to balance the conflicting latency and context requirements of streaming ASR and ST. The proposed use of a transducer-based MT model for parameter initialization is interesting and, if substantiated, would be a useful transfer-learning result. The paper also benefits from evaluation on real recorded bilingual conversations, which goes beyond purely simulated test sets. However, the headline claim of 'superior performances' in BLEU and latency is not yet supported by the evidence: the BLEU results are mixed across language directions and test sets, no confidence intervals or significance tests are reported, and the latency comparison is against a non-streaming MT baseline with a sentence-end triggering policy that strongly favors the streaming system. The central architecture is sound, but the experimental validation needs to be strengthened before the comparative claims can be accepted.

major comments (4)
  1. [§IV-D, Tables II–III] The comparative claim in the abstract is not backed by any measure of statistical reliability. No confidence intervals, significance tests, or multiple-seed variances are reported for any WER or BLEU value. In Table III, the best JSTAR configuration (#10) is 0.3 BLEU below the cascaded baseline on MC-FLEURS ES→EN (21.3 vs. 21.6) and only 0.9 BLEU above on EN→ES (20.2 vs. 19.3); even the larger RealConv differences (2.3 and 1.7 BLEU) could be within run-to-run variability for a 121M-parameter model. Please report confidence intervals or significance tests, or substantially soften the 'superior performances' claim.
  2. [§IV-E, Table IV] The latency comparison is not an architecture-level comparison. The cascaded baseline uses a non-streaming MT model that is triggered only by ASR sentence-end punctuation (as described by the reference to [5]), so its first finalized token is delayed by the waiting-for-sentence-boundary policy; a streaming transducer is almost guaranteed to emit an earlier first token under this protocol. The 3.8 s first-token advantage is therefore not strong evidence that JSTAR is a superior streaming design. Please compare against a streaming or incremental ST baseline, or control the triggering condition across systems. In addition, please define precisely what 'first/last final token' and 'P50' mean, because the last-token latencies (2.6 vs. 2.5 s) being smaller than the first-token latencies (7.1 vs. 3.3 s) is counterintuitive without an explicit definition.
  3. [§III, Table I] The paper's foundation for transducer-based MT is the claim that an RNN-T encoder plus predictor can 'delay an output such that the model can learn to emit words in the correct order' for non-monotonic translations. This capacity and learnability assumption is not tested directly. Table I reports only BLEU on a single ES↔EN conversational test set, with no analysis of whether the learned alignments actually realize the required reorderings, no comparison with a monotonic variant, and no error analysis for reordering failures. This concern is not necessarily fatal—any finite target permutation can be represented by delaying emissions—but the claimed 'first time' streaming MT capability needs more support. Please add alignment diagnostics or an ablation, and ideally results on a second, reordering-heavy test set or language pair.
  4. [§IV-D, Tables III–IV] The 'strong cascaded ST model' baseline is insufficiently specified. The ASR component, MT architecture and size, training data, decoding/beam-search settings, and whether the baseline uses the same simulated multi-channel front-end are not stated. Without this information, the reader cannot judge whether the comparison is controlled or whether the baseline is indeed strong. Please specify the baseline configuration in detail, including whether it shares the same ASR output or preprocessing, and report the relevant model parameters and training data.
minor comments (6)
  1. [§IV-C, §IV-E] There are typos: 'OHTER' in Section IV-C should be 'OTHER', and 'JSTART' in Section IV-E should be 'JSTAR'.
  2. [Abstract and §V] The abstract claims 'for the first time we propose to train a streaming MT model based on RNN-T' without qualification, while the conclusion says 'to the best of our knowledge'; please use a consistent qualifier and clarify the novelty relative to transducer-based speech translation systems such as LAMASSU [17], which also use RNN-T for translation but from speech rather than text.
  3. [§II-A, Eq. (2)] The loss weight λ=0.5 is chosen without any sensitivity analysis; a small sweep or at least a statement of stability across nearby values would help readers judge the robustness of the multi-objective training.
  4. [§IV-E, Table IV] The table caption says 'P50 latency' while the text in Section IV-E says 'average latency'; please clarify whether the reported quantity is the median, the mean, or another percentile, and how it is computed over the multi-minute conversations.
  5. [§IV-A] The description of the slow encoder gives the chunk size and resulting theoretical latency but does not explicitly state its right-context or left-context; please provide these values so that the fast/slow latency comparison is complete.
  6. [§IV-C] The MT comparison in Table I is between a 110M-parameter AED model and a 70M-parameter RNN-T model; the difference in model capacity is acknowledged but not discussed as a potential confound, so a sentence noting that the comparison is not parameter-matched would help.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: JSTAR's multi-objective losses and held-out BLEU/latency evaluations are externally grounded; the RNN-T representability caveat and mixed baseline comparisons are evidence-quality concerns, not derivation-to-fit circularity.

full rationale

The central derivation chain is self-contained with respect to its empirical claims. The training objective is an explicit sum of two RNN-T losses, L = L_st + λL_asr (Eq. 2), neither of which contains the reported BLEU or latency metrics; BLEU is computed on held-out external data (FLEURS) and on RealConv recordings against independent references, so the reported numbers are not fitted inputs renamed as predictions. The transducer-based MT used for initialization is evaluated on an in-house bilingual conversation test set and compared with a transformer-based MT model; this is a separate supporting experiment, not a circular derivation of the main system's quality. The fast-slow encoder architecture and multi-channel front-end are adopted from prior work by overlapping authors ([18], [20], [21]), but these citations supply reusable components, not the paper's conclusion; the superiority claim rests on the experiments rather than on those citations. The Section III statement that an RNN-T encoder plus predictor can delay emissions to learn non-monotonic word order is an unproved capacity/learnability assumption, which is a legitimate correctness risk but not circularity, since the paper does not define RNN-T success in terms of that property. The latency comparison uses a cascaded baseline whose MT is triggered by ASR sentence-end punctuation, which structurally disadvantages the cascade; this is an evaluation-protocol concern, not a case of the prediction being equivalent to its input by construction. No quoted step in the paper reduces Eq. X to Eq. Y by definition or imports a uniqueness theorem from the authors' own prior work to make the model choice forced. The absence of significance tests and the mixed MC-FLEURS BLEU results weaken the headline claim, but weakening evidence is not the same as circular reasoning. I therefore find no significant circularity.

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

The central empirical claims rest on hand-set hyperparameters and domain assumptions; no new physical or mathematical entities are introduced. The RNN-T reordering and simulation-transfer assumptions are the most load-bearing.

free parameters (4)
  • loss weight lambda for JSTAR = 0.5
    Set by hand in Eq. (2) and Section IV-A; no sensitivity analysis is given, and the ASR and ST balance affects both tasks.
  • loss weight lambda for intermediate MT loss = 0.1
    Chosen for the transducer-based MT pretraining in Section IV-A; no tuning study is described.
  • fast encoder chunk size = 5 frames (300 ms) with 1 frame right context
    Chosen latency and accuracy tradeoff; affects the stated 210 ms fast-encoder latency.
  • slow encoder chunk size = 10 frames (600 ms)
    Chosen latency and accuracy tradeoff; affects translation quality and the stated 360 ms slow-encoder latency.
assumptions (4)
  • domain assumption RNN-T can represent non-monotonic reorderings by delaying output labels
    Section III asserts the predictor can delay outputs so words emit in the correct order, 'which is often the case'; no proof or capacity bound is provided.
  • domain assumption Simulated multi-channel data approximates real smart-glasses recordings
    Section IV-B: all multi-channel training data uses measured RIRs and simulated conversations; real recordings are used only for evaluation.
  • domain assumption Teacher MT translations and attention alignments provide accurate ST training targets and SOT ordering
    Section II-B generates Z from Y using an MT model and text alignments from attention; errors in teacher targets can propagate into training.
  • domain assumption Pseudo-labeled unsupervised audio is accurate enough for training
    Section IV-B: 500k hours per language with teacher-model pseudo-labels; no quality filtering or error analysis is described.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Transcribing and Translating, Fast and Slow: Joint Speech Translation and Recognition." pith.science (2026). https://pith.science/paper/QJ6GJ3WJ

@misc{pith2026241215415,
  author       = {Pith},
  title        = {Pith review of: Transcribing and Translating, Fast and Slow: Joint Speech Translation and Recognition},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QJ6GJ3WJ}},
  note         = {Machine review of arXiv:2412.15415}
}
read the original abstract

We propose the joint speech translation and recognition (JSTAR) model that leverages the fast-slow cascaded encoder architecture for simultaneous end-to-end automatic speech recognition (ASR) and speech translation (ST). The model is transducer-based and uses a multi-objective training strategy that optimizes both ASR and ST objectives simultaneously. This allows JSTAR to produce high-quality streaming ASR and ST results. We apply JSTAR in a bilingual conversational speech setting with smart-glasses, where the model is also trained to distinguish speech from different directions corresponding to the wearer and a conversational partner. Different model pre-training strategies are studied to further improve results, including training of a transducer-based streaming machine translation (MT) model for the first time and applying it for parameter initialization of JSTAR. We demonstrate superior performances of JSTAR compared to a strong cascaded ST model in both BLEU scores and latency.

Figures

Figures reproduced from arXiv: 2412.15415 by the authors.

Figure 2
Figure 2. RNN-T lattice for an MT task example. B. Multi-talker Training JSTAR is applied for simultaneously recognizing and trans￾lating speech of a bilingual conversation with smart glasses in a streaming fashion. In order to recognize multiple speakers with an RNN-T based system without explicit speaker sepa￾ration, we are using SOT. Let’s consider the following case, where the wearer of the smart glasses (SELF) is speakin… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

29 extracted references · 23 canonical work pages

  1. [5]

    Navigating the minefield of mt beam search in cascaded streaming speech translation,

    Rastislav Rabatin, Frank Seide, and Ernie Chang, “Navigating the minefield of mt beam search in cascaded streaming speech translation,” arXiv preprint arXiv:2407.11010 , 2024

    arXiv 2024

  2. [1]

    Seamless: Multilingual expressive and streaming speech translation,

    Seamless Communication, Lo ¨ıc Barrault, Yu-An Chung, Mariano Coria Meglioli, David Dale, Ning Dong, Mark Duppenthaler, Paul-Ambroise Duquenne, Brian Ellis, Hady Elsahar, Justin Haaheim, John Hoffman, Min-Jae Hwang, Hirofumi Inaguma, Christopher Klaiber, Ilia Kulikov, Pengwei Li, Daniel Licht, Jean Maillard, Ruslan Mavlyutov, Alice Rako- toarison, Kaushik...

    arXiv 2023

  3. [2]

    Strea- mAtt: Direct streaming speech-to-text translation with attention-based audio history selection,

    Sara Papi, Marco Gaido, Matteo Negri, and Luisa Bentivogli, “Strea- mAtt: Direct streaming speech-to-text translation with attention-based audio history selection,” in Proc. Association for Computational Lin- guistic, Aug. 2024, pp. 3692–3707

    work page 2024

  4. [3]

    Streamspeech: Simultaneous speech-to-speech transla- tion with multi-task learning,

    Shaolei Zhang, Qingkai Fang, Shoutao Guo, Zhengrui Ma, Min Zhang, and Yang Feng, “Streamspeech: Simultaneous speech-to-speech transla- tion with multi-task learning,” arXiv preprint arXiv:2406.03049 , 2024

    arXiv 2024

  5. [4]

    A comparative study on end-to-end speech to text translation,

    Parnia Bahar, Tobias Bieschke, and Hermann Ney, “A comparative study on end-to-end speech to text translation,” in Proc. ASRU, 2019, pp. 792– 799

    work page 2019

  6. [6]

    End-to-end speech translation with knowledge distillation,

    Yuchen Liu, Hao Xiong, Jiajun Zhang, Zhongjun He, Hua Wu, Haifeng Wang, and Chengqing Zong, “End-to-end speech translation with knowledge distillation,” in Proc. Interspeech, 2019, pp. 1128–1132

    work page 2019

  7. [7]

    Multilingual speech translation from efficient finetuning of pretrained models,

    Xian Li, Changhan Wang, Yun Tang, Chau Tran, Yuqing Tang, Juan Pino, Alexei Baevski, Alexis Conneau, and Michael Auli, “Multilingual speech translation from efficient finetuning of pretrained models,” in Proc. ACL IJCNLP , Aug. 2021, pp. 827–838

    work page 2021

  8. [8]

    ComSl: A composite speech-language model for end-to-end speech-to-text translation,

    Chenyang Le, Yao Qian, Long Zhou, Shujie LIU, Yanmin Qian, Michael Zeng, and Xuedong Huang, “ComSl: A composite speech-language model for end-to-end speech-to-text translation,” in Proc. NeurIPS , A. Oh, T. Naumann, A. Globerson, K. Saenko, M. Hardt, and S. Levine, Eds., 2023, vol. 36, pp. 58312–58323

    work page 2023

Show all 29 references
  1. [9]

    Sequence transduction with recurrent neural networks,

    Alex Graves, “Sequence transduction with recurrent neural networks,” arXiv preprint arXiv:1211.3711 , 2012

    2012 arXiv

  2. [10]

    Streaming end-to-end speech recognition for mobile devices,

    Y . He, T. N. Sainath, R. Prabhavalkar, I. McGraw, R. Alvarez, D. Zhao, D. Rybach, A. Kannan, Y . Wu, R. Pang, Q. Liang, D. Bhatia, Y . Shang- guan, B. Li, G. Pundak, K. C. Sim, T. Bagby, S. Chang, K. Rao, and A. Gruenstein, “Streaming end-to-end speech recognition for mobile ...

    2019

  3. [11]

    Dual causal/non- causal self-attention for streaming end-to-end speech recognition,

    Niko Moritz, Takaaki Hori, and Jonathan Le Roux, “Dual causal/non- causal self-attention for streaming end-to-end speech recognition,” in Proc. Interspeech, 2021, pp. 1822–1826

    2021

  4. [12]

    Token-level serialized output training for joint streaming asr and st leveraging textual alignments,

    Sara Papi, Peidong Wan, Junkun Chen, Jian Xue, Jinyu Li, and Yashesh Gaur, “Token-level serialized output training for joint streaming asr and st leveraging textual alignments,” in Proc. ASRU, 2023

    2023

  5. [13]

    Extended graph temporal classification for multi- speaker end-to-end ASR,

    Xuankai Chang, Niko Moritz, Takaaki Hori, Shinji Watanabe, and Jonathan Le Roux, “Extended graph temporal classification for multi- speaker end-to-end ASR,” in Proc. ICASSP, 2022, pp. 7322–7326

    2022

  6. [14]

    Streaming multi-talker ASR with token-level serialized output training,

    Naoyuki Kanda, Jian Wu, Yu Wu, Xiong Xiao, Zhong Meng, Xiaofei Wang, Yashesh Gaur, Zhuo Chen, Jinyu Li, and Takuya Yoshioka, “Streaming multi-talker ASR with token-level serialized output training,” arXiv preprint arXiv:2202.00842 , 2022

    2022 arXiv

  7. [15]

    Cross attention augmented transducer networks for simultaneous translation,

    Dan Liu, Mengge Du, Xiaoxi Li, Ya Li, and Enhong Chen, “Cross attention augmented transducer networks for simultaneous translation,” in Proc. EMNLP, 2021, pp. 39–55

    2021

  8. [16]

    Large- scale streaming end-to-end speech translation with neural transducers,

    Jian Xue, Peidong Wang, Jinyu Li, Matt Post, and Yashesh Gaur, “Large- scale streaming end-to-end speech translation with neural transducers,” in Proc. Interspeech, 2022

    2022

  9. [17]

    Lamassu: Streaming language-agnostic mul- tilingual speech recognition and translation using neural transducers,

    Peidong Wang, Eric Sun, Jian Xue, Yu Wu, Long Zhou, Yashesh Gaur, Shujie Liu, and Jinyu Li, “Lamassu: Streaming language-agnostic mul- tilingual speech recognition and translation using neural transducers,” in Proc. Interspeech, 2023

    2023

  10. [18]

    Streaming parallel transducer beam search with fast-slow cascaded encoders,

    Jay Mahadeokar, Yangyang Shi, Ke Li, Duc Le, Jiedan Zhu, Vikas Chandra, Ozlem Kalinli, and Michael L. Seltzer, “Streaming parallel transducer beam search with fast-slow cascaded encoders,” in Proc. Interspeech, 2022

    2022

  11. [19]

    Streaming transformer transducer based speech recognition using non-causal convolution,

    Yangyang Shi, Chunyang Wu, Dilin Wang, Alex Xiao, Jay Mahadeokar, Xiaohui Zhang, Chunxi Liu, Ke Li, Yuan Shangguan, Varun Nagaraja, Ozlem Kalinli, and Mike Seltzer, “Streaming transformer transducer based speech recognition using non-causal convolution,” in Proc. ICASSP, 2022,...

    2022

  12. [20]

    AGADIR: Towards array-geometry agnostic directional speech recognition,

    Ju Lin, Niko Moritz, Yiteng Huang, Ruiming Xie, Ming Sun, Christian Fuegen, and Frank Seide, “AGADIR: Towards array-geometry agnostic directional speech recognition,” in Proc. ICASSP , 2024, pp. 11951– 11955

    2024

  13. [21]

    Directional speech recognition for speaker disambiguation and cross-talk suppression,

    Ju Lin, Niko Moritz, Ruiming Xie, Kaustubh Kalgaonkar, Christian Fuegen, and Frank Seide, “Directional speech recognition for speaker disambiguation and cross-talk suppression,” Proc. Interspeech , pp. 3522–3526, 2023

    2023

  14. [22]

    FLEURS: Few-shot learning evaluation of universal representations of speech,

    Alexis Conneau, Min Ma, Simran Khanuja, Yu Zhang, Vera Axel- rod, Siddharth Dalmia, Jason Riesa, Clara Rivera, and Ankur Bapna, “FLEURS: Few-shot learning evaluation of universal representations of speech,” in Proc. SLT, 2023, pp. 798–805

    2023

  15. [23]

    Word alignment by fine-tuning embeddings on parallel corpora,

    Zi-Yi Dou and Graham Neubig, “Word alignment by fine-tuning embeddings on parallel corpora,” in Proc. EACL, 2021

    2021

  16. [24]

    The CHiME- 8 MMCSG Challenge: Multi-modal conversations in smart glasses,

    Kate ˇrina ˇZmol´ıkov´a, Simone Merello, Kaustubh Kalgaonkar, Ju Lin, Niko Moritz, Pingchuan Ma, Ming Sun, Honglie Chen, Antoine Saliou, Stavros Petridis, Christian Fuegen, and Michael Mandel, “The CHiME- 8 MMCSG Challenge: Multi-modal conversations in smart glasses,” in Proc....

    2024

  17. [25]

    CCMatrix: Mining billions of high-quality parallel sentences on the web,

    Holger Schwenk, Guillaume Wenzek, Sergey Edunov, Edouard Grave, and Armand Joulin, “CCMatrix: Mining billions of high-quality parallel sentences on the web,” arXiv preprint arXiv:1911.04944 , 2019

    1911 arXiv

  18. [26]

    Beyond english-centric multilingual machine translation,

    Angela Fan, Shruti Bhosale, Holger Schwenk, Zhiyi Ma, Ahmed El- Kishky, Siddharth Goyal, Mandeep Baines, Onur Celebi, Guillaume Wenzek, Vishrav Chaudhary, et al., “Beyond english-centric multilingual machine translation,” Journal of Machine Learning Research , vol. 22, no. 107...

    2021

  19. [27]

    Opensubtitles2016: Extracting large parallel corpora from movie and tv subtitles,

    Pierre Lison and J ¨org Tiedemann, “Opensubtitles2016: Extracting large parallel corpora from movie and tv subtitles,” 2016

    2016

  20. [28]

    Building subject-aligned comparable corpora and mining it for truly parallel sentence pairs,

    Krzysztof Wołk and Krzysztof Marasek, “Building subject-aligned comparable corpora and mining it for truly parallel sentence pairs,” Procedia Technology, vol. 18, pp. 126–132, 2014

    2014

  21. [29]

    Chime 8 task 3: Multi-talker word error rate,

    Michael Mandel, Christian Fuegen, Pingchuan Ma, Honglie Chen, Antoine Saliou, Stavros Petridis, Katerina Zmolikova, Simone Merello, Niko Moritz, Kaustubh Kalgaonkar, Ju Lin, and Ming Sun, “Chime 8 task 3: Multi-talker word error rate,” https://www.chimechallenge.org/ current/t...

Pith tools

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