Pith. sign in

REVIEW 2 major objections 4 minor 3 cited by

MERaLiON-SpeechEncoder: Towards a Speech Foundation Model for Singapore and Beyond

T0 review · 2 major / 4 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 MERaLiON-SpeechEncoder is a 630M-parameter self-supervised speech encoder pretrained from scratch on 200,000 hours of audio that improves speech recognition on spontaneous and Singapore English speech while staying competitive across ten…

desk verdict Useful open BEST-RQ model release and SUPERB evaluation, but the Singapore-English gains need an explicit data-split statement before you trust them. read the letter →

arxiv 2412.11538 v3 pith:4EL5ZDPV submitted 2024-12-16 cs.CL cs.AIeess.AS

Muhammad Huzaifah , Geyu Lin , Tianchi Liu , Hardik B. Sailor , Kye Min Tan , Tarun K. Vangani , Qiongqiong Wang , Jeremy H. M. Wong
show 3 more authors
Jinyang Wu Nancy F. Chen Ai Ti Aw
This is my paper · ORCID
classification cs.CLcs.AIeess.AS
keywords self-supervisedspeechrepresentationlearningBEST-RQfoundationmodelSingaporeEnglishSinglishmaskedlanguagemodelingSUPERBbenchmarkautomaticrecognition
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

MERaLiON-SpeechEncoder is a 630M-parameter speech foundation model pretrained from scratch on 200,000 hours of unlabelled audio with BEST-RQ, a BERT-style masked-language objective whose target labels come from frozen random projections rather than iterative clustering. The paper reports improved automatic speech recognition on spontaneous and Singapore English speech: 5.6% word error rate on TEDLIUMv3 test versus 6.2% for an in-house finetuned WavLM large, and a 15.2% average WER on the six-part Singapore English/Singlish National Speech Corpus subset after finetuning on only 420 hours. The same encoder reaches 2.1%/4.3% WER on LibriSpeech test-clean/test-other with 960 hours of finetuning, matching HuBERT large, and scores 82.62 on the ten-task SUPERB benchmark, between HuBERT large (82.25) and WavLM large (84.77). By releasing checkpoints and finetuning recipes, the authors aim to make a large-scale BEST-RQ encoder publicly available and useful for speech applications in Singapore and Southeast Asia.

What carries the argument

The load-bearing mechanism is BEST-RQ (BERT speech pre-training with random-projection quantizer), a masked-language objective that avoids the iterative k-means target computation used by HuBERT and WavLM. A frozen random projection matrix $A_j$ maps each downsampled, segment-level mean/variance-normalized Mel-spectrogram frame to one of 2048 random codewords $c_{ij}$ in each of 32 codebooks, and the label for a frame is $y^{\mathrm{ref}}_j = \arg\min_i (A_j x - c_{ij})^\top(A_j x - c_{ij})$. The 24-layer Conformer encoder is trained to predict these labels for masked frames using cross-entropy over the masked positions only. The paper's reported recipe changes—masking probability 0.4 instead of 0.01, Euclidean distance instead of cosine similarity, no L2 normalization, and 32 codebooks instead of one—are what make the random-projection targets effective at this scale.

What would settle it

Check the NSC evaluation split described in Appendix A.2 against the pretraining corpus in Table 1 by speaker ID or audio hash; if any test utterance appears in pretraining, re-run the Singapore English evaluation on a held-out corpus that was never in the pretraining mix and see whether the 15.2% average WER and the margin over WavLM large survive.

Watch

Extended reading notes

Core claim

The central claim is that a BEST-RQ speech encoder trained at scale—200,000 hours, 630M parameters, 24 Conformer layers—yields representations that transfer especially well to the kinds of English that standard pretraining corpora under-represent: spontaneous talk and Singapore-accented English, including code-switched Singlish. Evidence for the claim is that finetuning on 420 hours of the National Speech Corpus produces a 15.2% average WER across the corpus's six parts, beating a Whisper large v3 finetuned on the full 8,169-hour cleaned corpus (16.9%) and outperforming WavLM large on the named-entity-heavy Part 2 and the spontaneous code-switching Part 4 by relative margins of 36.5% and 5.8%. On spontaneous TED talks, the encoder reaches 5.6% test WER, about 10% relative better than the finetuned WavLM baseline, and on LibriSpeech it ties HuBERT large at 2.1%/4.3%. The authors also show the encoder is not ASR-specific: its 82.62 SUPERB score places it between HuBERT large and WavLM large across phoneme, keyword, intent, slot, speaker, and emotion tasks.

Load-bearing premise

The load-bearing premise is that the Singapore English test utterances used to report word error rates were not among the audio clips the model saw, unlabelled, during pretraining; the report never states explicitly that this exclusion was made.

Editorial extensions

If this is right

  • Finetuning the encoder on 960 hours of LibriSpeech gives 2.1% WER on test-clean and 4.3% on test-other, matching HuBERT large and beating the in-house finetuned WavLM large (2.5%/4.6%).
  • On TEDLIUMv3 spontaneous speech, the encoder's 5.6% test WER is a relative improvement of about 10% over the in-house finetuned WavLM large's 6.2%.
  • On the Singapore English NSC subset, finetuning with 420 hours yields a 15.2% average WER across the six parts, better than Whisper large v3 finetuned on the full 8,169-hour corpus (16.9%) despite using roughly 5% of the finetuning data.
  • The SUPERB overall score of 82.62 indicates the encoder remains usable on non-ASR tasks such as phoneme recognition, keyword spotting, intent classification, speaker verification, and emotion recognition.
  • Public checkpoints and finetuning recipes let downstream developers adapt the encoder to their own tasks without repeating a 200,000-hour pretraining run.

Reading between the lines

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

  • Editorial inference: the report does not settle whether the NSC test audio was excluded from the 10,000 hours of NSC pretraining; if it was not, the Singapore English gains could partly reflect the encoder having heard the test speakers' unlabelled audio. This is testable by re-evaluating on a fresh Singapore English corpus.
  • Editorial inference: because the random-projection quantizer is language-agnostic and the pretraining mix already includes 30,000 hours of multilingual Common Voice, the same recipe should transfer to the planned Malay, Chinese, Tamil, Indonesian, Thai, and Vietnamese releases once comparable evaluation benchmarks exist.
  • Editorial inference: the 4x downsampling in the feature extractor may limit fine-grained time-sensitive tasks; the SUPERB speaker-verification and diarisation scores sit between HuBERT and WavLM, so future versions with less downsampling or pretraining augmentation could close the gap without losing the ASR gains.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 4 minor

Summary. The paper presents MERaLiON-SpeechEncoder, a 630M-parameter Conformer-based speech encoder pretrained with the BEST-RQ objective on roughly 200k hours of speech, initialized from a Libri-light checkpoint and then continuously pretrained on an expanded corpus that includes 10k hours of Singapore English from the National Speech Corpus (NSC). The authors report ASR results after CTC finetuning on LibriSpeech, TEDLIUMv3, and a 420-hour NSC subset, as well as results on ten SUPERB tasks. They claim improvements over an in-house finetuned WavLM large on spontaneous speech and Singapore English, and competitive performance with Wav2Vec 2.0, HuBERT, and WavLM across SUPERB. The model checkpoints are released on Hugging Face.

Significance. If the reported results are taken at face value, the paper's main contribution is an open, independently implemented, large-scale BEST-RQ encoder with detailed training and evaluation documentation. This fills a real gap, since no public BEST-RQ model at this scale has been released, and the focus on Singapore English/Singlish is valuable for the regional speech community. The detailed account of AMD-GPU pretraining and the SUPERB evaluation beyond ASR are also useful. However, the central Singapore-ASR claim and the SUPERB comparison rest on data-hygiene issues that must be resolved before the results can be interpreted as stated.

major comments (2)
  1. [Section 3.3, Table 1, and Appendix A.2/Table 9] The Singapore-English claim depends on whether the NSC evaluation utterances were excluded from the 10k-hour NSC component of the pretraining corpus. The paper states in Table 1 that NSC supplied about 10k hours of unlabelled pretraining data, while Table 9 shows that after filtering there are 8169.48 hours of NSC train data and 37.65 hours of NSC test data. The text never states that the Table 9 test utterances were withheld from pretraining. Appendix A.2's 'Data Splits Consistency' note only prevents identical transcriptions from appearing in different supervised splits; it does not address the unsupervised pretraining manifest. Since BEST-RQ pretraining consumes raw waveforms without labels, any overlap between pretraining audio and the test set would let the CTC finetuning stage exploit utterance- or speaker-specific acoustic structure. Each NSC Part test split is only 4–8 hours, so even modest contamination could materially inflate the reported WERs and the 36.5% relative gain on Part 2. Please state explicitly whether the NSC test subset was excluded from the pretraining corpus, and if the full raw NSC corpus was used, provide the exact manifest or filtering procedure that guarantees exclusion.
  2. [Section 5.2, Table 8, and Table 10] The SUPERB comparison is weakened by test-set selection of hyperparameters. Section 5.2 states that the learning rates and batch sizes in Table 10 'were chosen simply by using the best result on the test set, between either the default SUPERB hyperparameters or those chosen for WavLM large.' Selecting hyperparameters on the test set inflates the reported SUPERB scores relative to baselines whose numbers may come from default settings. Please report which of the two configurations was chosen for each task, present results for both configurations, or select hyperparameters on the development set. This is load-bearing for the claim that the encoder is 'competitive' with other state-of-the-art encoders across ten SUPERB tasks.
minor comments (4)
  1. [Abstract, Section 1, and Section 3.4] The abstract and introduction say the model was 'pre-trained from scratch on 200,000 hours,' but Section 3.4 explains that the full pretraining was initialized from a checkpoint pretrained on Libri-light (60k hours). Please rephrase to avoid the contradictory 'from scratch' description.
  2. [Table 7] The row for 'Whisper large v3 (finetuned in-house)' appears as '81694.4' in the Finetuning column; this should be '8169' followed by the separate WER '4.4'.
  3. [Throughout] There are several typographical and spacing issues, including 'V oxPopuli' and 'V oice' for VoxPopuli and Voice, 'retraining' instead of 'retaining' in the contributions list, and 'non-conclusive' instead of 'non-exhaustive' in Section 6. These should be corrected.
  4. [Appendix A.2] The random selection of 70 hours from each NSC part is not reproducible as described; please provide the random seed or a stable identifier for the chosen subset, including the test subset construction.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the encoder is evaluated against external benchmarks, and the only flagged concern (potential NSC pretraining/evaluation overlap) is a data-splitting issue, not a circular derivation.

full rationale

The claimed results are empirical evaluations, not derivations. Pre-training uses BEST-RQ as an externally published objective, with model checkpoints and finetuning recipes released for reproduction; hyperparameters are selected by ablations and then measured against LibriSpeech, TEDLIUMv3, and NSC test splits, with WavLM large and Whisper as independent baselines. No equation in the paper defines a reported score in terms of a fitted parameter, a self-cited result, or the pre-training objective itself, so there is no self-definitional or fit-renamed-as-prediction step. The only self-citation, MERaLiON-AudioLLM, is for a future integrated system and is not used to justify encoder quality. One data-hygiene caveat exists: Table 1 lists 10K hours of NSC among the pretraining data, while Table 9 defines a 37.65-hour NSC test split, and the paper never states that test audio was excluded from unsupervised pretraining; if overlap occurred, the Singapore-English WER gains would be inflated. That is a potential data-split leakage concern, not a reasoning-circularity reduction, so it does not by itself raise the circularity score.

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

The model is an empirical artifact; the listed free parameters are training hyperparameters tuned via ablations, not constants derived from first principles. The central claim relies on the BEST-RQ assumption and on data split integrity, particularly the NSC pretraining/evaluation overlap.

free parameters (4)
  • masking_probability = 0.4
    Tuned via validation loss and SUPERB ASR results; differs from the original BEST-RQ value of 0.01.
  • number_of_codebooks = 32
    Ablations compare 1 vs 16 codebooks; 32 is used in the final model without a direct ablation, likely inherited from USM.
  • vocabulary_size_per_codebook = 2048
    Ablated across 1024, 2048, and 8192; 2048 gives the best SUPERB ASR result in Table 2.
  • codebook_vector_dimension = 16
    Not ablated in the paper; adopted from prior BEST-RQ/USM work.
assumptions (4)
  • domain assumption Random-projection quantizer targets are informative enough for self-supervised speech representation learning.
    The paper adopts BEST-RQ from Chiu et al. (2022) and relies on this assumption for the pretraining objective (Section 3.1).
  • domain assumption The SUPERB benchmark is a valid measure of generalizability of speech representations.
    The paper uses SUPERB as the primary general-purpose evaluation (Section 5.2).
  • domain assumption The National Speech Corpus (NSC) is representative of Singapore English and Singlish.
    The paper uses NSC as both pretraining data and ASR evaluation data for Singapore English (Section 3.3 and Appendix A.2).
  • domain assumption Continuous pretraining from a Libri-light checkpoint transfers to new domains.
    The full pretraining run is initialized from the Libri-light model based on the experiments in Section 4.1.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MERaLiON-SpeechEncoder: Towards a Speech Foundation Model for Singapore and Beyond." pith.science (2026). https://pith.science/paper/4EL5ZDPV

@misc{pith2026241211538,
  author       = {Pith},
  title        = {Pith review of: MERaLiON-SpeechEncoder: Towards a Speech Foundation Model for Singapore and Beyond},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4EL5ZDPV}},
  note         = {Machine review of arXiv:2412.11538}
}
read the original abstract

This technical report describes the MERaLiON-SpeechEncoder, a foundation model designed to support a wide range of downstream speech applications. Developed as part of Singapore's National Multimodal Large Language Model Programme, the MERaLiON-SpeechEncoder is tailored to address the speech processing needs in Singapore and the surrounding Southeast Asian region. The model currently supports mainly English, including the variety spoken in Singapore. We are actively expanding our datasets to gradually cover other languages in subsequent releases. The MERaLiON-SpeechEncoder was pre-trained from scratch on 200,000 hours of unlabelled speech data using a self-supervised learning approach based on masked language modelling. We describe our training procedure and hyperparameter tuning experiments in detail below. Our evaluation demonstrates improvements to spontaneous and Singapore speech benchmarks for speech recognition, while remaining competitive to other state-of-the-art speech encoders across ten other speech tasks. We commit to releasing our model, supporting broader research endeavours, both in Singapore and beyond.

Figures

Figures reproduced from arXiv: 2412.11538 by the authors.

Figure 1
Figure 1. Different ways of utilising the MERaLiON-SpeechEncoder. (Left) Decoder layers are added [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Overview of the pre-training framework. The random projection quantiser comprises a [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. (Left) Validation loss on the Librispeech dataset when initialising with a checkpoint [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Validation loss over training steps for different types of positional embeddings: relative [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: (Left) Validation loss over steps for different masking probabilities. Approximate masking [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 3 Pith papers

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

  1. Contextual Paralinguistic Data Creation for Multi-Modal Speech-LLM: Data Condensation and Spoken QA Generation

    cs.CL 2025-05 conditional novelty 5.0 of 10

    A pipeline condenses in-the-wild speech with emotion labels and uses ChatGPT to generate contextual paralinguistic QA pairs, released as a 480-sample benchmark.

  2. Singlish, Can or Not? Fine-Tuning and Evaluating Zero-Shot TTS for Singapore English

    eess.AS 2026-07 conditional novelty 4.0 of 10

    Fine-tuning Chatterbox and CosyVoice 3 on 50 Singlish speakers measurably raises accent similarity, and the gain persists on held-out speakers.

  3. MERaLiON-AudioLLM: Bridging Audio and Language with Large Language Models

    cs.CL 2024-12 conditional novelty 4.0 of 10

    A new Singapore-tailored AudioLLM, built by fusing a fine-tuned Whisper encoder with SEA-LION V3, shows strong in-domain speech recognition but mixed task-understanding gains versus a cascaded baseline.

Reference graph

Works this paper leans on

36 extracted references · 28 canonical work pages · cited by 3 Pith papers

  1. [1]

    Ardila, M

    R. Ardila, M. Branson, K. Davis, M. Kohler, J. Meyer, M. Henretty, R. Morais, L. Saunders, F. Tyers, and G. Weber. Common V oice: A massively-multilingual speech corpus. In N. Calzolari, F. B \'e chet, P. Blache, K. Choukri, C. Cieri, T. Declerck, S. Goggi, H. Isahara, B. Maegaard, J. Mariani, H. Mazo, A. Moreno, J. Odijk, and S. Piperidis, editors, LREC,...

    work page 2020

  2. [2]

    Press release on S ingapore’s national multimodal large language model programme, 2023

    A*STAR. Press release on S ingapore’s national multimodal large language model programme, 2023. URL https://www.a-star.edu.sg/i2r/news-accolades/news-accolades/press-releases/NewsNAccolades/press-releases/SEA-LLM. Accessed: 2024-12-10

    work page 2023

  3. [3]

    Baevski, H

    A. Baevski, H. Zhou, A. Mohamed, and M. Auli. wav2vec 2.0: A framework for self-supervised learning of speech representations. In NeurIPS, Red Hook, NY, USA, 2020. Curran Associates Inc. ISBN 9781713829546

    work page 2020

  4. [4]

    Baevski, W.-N

    A. Baevski, W.-N. Hsu, Q. Xu, A. Babu, J. Gu, and M. Auli. data2vec: A general framework for self-supervised learning in speech, vision and language. In ICML, volume 162, pages 1298--1312. PMLR, 17--23 Jul 2022

    work page 2022

  5. [5]

    G. Chen, S. Chai, G.-B. Wang, J. Du, W.-Q. Zhang, C. Weng, D. Su, D. Povey, J. Trmal, J. Zhang, M. Jin, S. Khudanpur, S. Watanabe, S. Zhao, W. Zou, X. Li, X. Yao, Y. Wang, Z. You, and Z. Yan. GigaSpeech : An evolving, multi-domain ASR corpus with 10,000 hours of transcribed audio. In Interspeech, pages 3670--3674, 2021 a

    work page 2021

  6. [6]

    S. Chen, C. Wang, Z. Chen, Y. Wu, S. Liu, Z. Chen, J. Li, N. Kanda, T. Yoshioka, X. Xiao, J. Wu, L. Zhou, S. Ren, Y. Qian, Y. Qian, M. Zeng, and F. Wei. Wav LM : Large-scale self-supervised pre-training for full stack speech processing. IEEE Journal of Selected Topics in Signal Proc., 16: 0 1505--1518, 2021 b

    work page 2021

  7. [7]

    C.-C. Chiu, J. Qin, Y. Zhang, J. Yu, and Y. Wu. Self-supervised learning with random-projection quantizer for speech recognition. In ICML, volume 162, pages 3915--3924. PMLR, 17--23 Jul 2022

    work page 2022

  8. [8]

    Devlin, M.-W

    J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova. BERT : Pre-training of deep bidirectional transformers for language understanding. In NAACL, 2019

    work page 2019

Show all 36 references
  1. [9]

    Galvez, G

    D. Galvez, G. Diamos, J. Torres, K. Achorn, J. Cer\' o n, A. Gopi, D. Kanter, M. Lam, M. Mazumder, and V. Janapa Reddi. The People’s Speech : A large-scale diverse E nglish speech recognition dataset for commercial usage. In NeurIPS Track on Datasets and Benchmarks, volume 1, ...

    2021

  2. [10]

    Graves, S

    A. Graves, S. Fern\' a ndez, F. Gomez, and J. Schmidhuber. Connectionist temporal classification: labelling unsegmented sequence data with recurrent neural networks. In ICML, page 369–376, New York, NY, USA, 2006

    2006

  3. [11]

    Gulati, J

    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 Interspeech, pages 5036--5040, 2020

    2020

  4. [12]

    Hernandez, V

    F. Hernandez, V. Nguyen, S. Ghannay, N. A. Tomashenko, and Y. Est \`e ve. TED-LIUM 3 : Twice as much data and corpus repartition for experiments on speaker adaptation. In SPECOM, volume abs/1805.04699, 2018

    2018 arXiv

  5. [13]

    W.-N. Hsu, B. Bolte, Y.-H. H. Tsai, K. Lakhotia, R. Salakhutdinov, and A. Mohamed. Hu BERT : Self-supervised speech representation learning by masked prediction of hidden units. IEEE/ACM Trans. Audio, Speech and Lang. Proc., 29: 0 3451–3460, Oct 2021. ISSN 2329-9290

    2021

  6. [14]

    Kahn , M

    J. Kahn , M. Rivière , W. Zheng , E. Kharitonov , Q. Xu , P. E. Mazaré , J. Karadayi , V. Liptchinsky , R. Collobert , C. Fuegen , T. Likhomanenko , G. Synnaeve , A. Joulin , A. Mohamed , and E. Dupoux . Libri-light: A benchmark for ASR with limited or no supervision. In IEEE ...

    2020

  7. [15]

    J. X. Koh, A. Mislan, K. Khoo, B. Ang, W. Ang, C. Ng, and Y.-Y. Tan. Building the S ingapore E nglish national speech corpus. In Interspeech, pages 321--325, 2019

    2019

  8. [16]

    Kudo and J

    T. Kudo and J. Richardson. S entence P iece: A simple and language independent subword tokenizer and detokenizer for neural text processing. In E. Blanco and W. Lu, editors, EMNLP, pages 66--71, Brussels, Belgium, Nov. 2018

    2018

  9. [17]

    MERaLiON-AudioLLM : Technical report

    MERaLiON Team . MERaLiON-AudioLLM : Technical report. arXiv preprint arXiv:2412.09818, 2024

    2024 arXiv

  10. [18]

    Mohamed, H.-y

    A. Mohamed, H.-y. Lee, L. Borgholt, J. D. Havtorn, J. Edin, C. Igel, K. Kirchhoff, S.-W. Li, K. Livescu, L. Maaløe, T. N. Sainath, and S. Watanabe. Self-supervised speech representation learning: A review. IEEE Journal of Selected Topics in Signal Processing, 16 0 (6): 0 1179-...

    2022

  11. [19]

    Whisper text normalization, 2024

    OpenAI. Whisper text normalization, 2024. URL https://github.com/openai/whisper/tree/main/whisper/normalizers. Accessed: 2024-12-10

    2024

  12. [20]

    M. Ott, S. Edunov, A. Baevski, A. Fan, S. Gross, N. Ng, D. Grangier, and M. Auli. Fairseq: A fast, extensible toolkit for sequence modeling. In Proceedings of NAACL-HLT 2019: Demonstrations, 2019

    2019

  13. [21]

    Panayotov, G

    V. Panayotov, G. Chen, D. Povey, and S. Khudanpur. Librispeech: An ASR corpus based on public domain audio books. In IEEE ICASSP, pages 5206--5210, Brisbane, Australia, 2015

    2015

  14. [22]

    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. In Interspeech, pages 2613--2617, 2019

    2019

  15. [23]

    D. Povey. TEDLIUM ASR training receipe using Kaldi speech recognition toolkit, 2024. URL https://github.com/kaldi-asr/kaldi/tree/master/egs/tedlium/s5_r3. Accessed: 2024-12-10

    2024

  16. [24]

    Pratap, Q

    V. Pratap, Q. Xu, A. Sriram, G. Synnaeve, and R. Collobert. MLS : A large-scale multilingual dataset for speech research. In Interspeech, pages 2757--2761, Shanghai, China, 2020

    2020

  17. [25]

    Radford, J

    A. Radford, J. W. Kim, T. Xu, G. Brockman, C. McLeavey, and I. Sutskever. Robust speech recognition via large-scale weak supervision. In ICML, pages 28492--28518. PMLR, 2023

    2023

  18. [26]

    F. M. Ramirez, L. Chkhetiani, A. Ehrenberg, R. McHardy, R. Botros, Y. Khare, A. Vanzo, T. Peyash, G. Oexle, M. Liang, I. Sklyar, E. Fakhan, A. Efty, D. McCrystal, S. Flamini, D. Donato, and T. Yoshioka. Anatomy of industrial scale multilingual ASR . arXiv preprint arXiv:2404.0...

    2024 arXiv

  19. [27]

    Schneider, A

    S. Schneider, A. Baevski, R. Collobert, and M. Auli. wav2vec: Unsupervised pre-training for speech recognition. In Interspeech, pages 3465--3469, 2019. doi:10.21437/Interspeech.2019-1873

    2019 doi

  20. [28]

    Seed- ASR : Understanding diverse speech and contexts with LLM -based speech recognition

    Seed Team . Seed- ASR : Understanding diverse speech and contexts with LLM -based speech recognition. arXiv preprint arXiv:2407.04675, 2024

    2024 arXiv

  21. [29]

    J. Shi, D. Berrebbi, W. Chen, E.-P. Hu, W.-P. Huang, H.-L. Chung, X. Chang, S.-W. Li, A. Mohamed, H. yi Lee, and S. Watanabe. ML-SUPERB: Multilingual Speech Universal PERformance Benchmark . In Interspeech, pages 884--888, 2023. doi:10.21437/Interspeech.2023-1316

    2023 doi

  22. [30]

    Tsai, H.-J

    H.-S. Tsai, H.-J. Chang, W.-C. Huang, Z. Huang, K. Lakhotia, S.-w. Yang, S. Dong, A. Liu, C.-I. Lai, J. Shi, X. Chang, P. Hall, H.-J. Chen, S.-W. Li, S. Watanabe, A. Mohamed, and H.-y. Lee. SUPERB - SG : Enhanced Speech processing Universal PER formance Benchmark for Semantic ...

    2022

  23. [31]

    C. Wang, M. Riviere, A. Lee, A. Wu, C. Talnikar, D. Haziza, M. Williamson, J. Pino, and E. Dupoux. V ox P opuli: A large-scale multilingual speech corpus for representation learning, semi-supervised learning and interpretation. In ACL and IJCNLP (Volume 1: Long Papers), pages ...

    2021

  24. [32]

    C. Wang, Y. Wu, S. Chen, S. Liu, J. Li, Y. Qian, and Z. Yang. Improving self-supervised learning for speech recognition with intermediate layer supervision. In IEEE ICASSP, pages 7092--7096, Singapore, 2022

    2022

  25. [33]

    Watanabe, T

    S. Watanabe, T. Hori, S. Karita, T. Hayashi, J. Nishitoba, Y. Unno, N. Enrique Yalta Soplin , J. Heymann, M. Wiesner, N. Chen, A. Renduchintala, and T. Ochiai. ESP net: End-to-end speech processing toolkit. In Interspeech, pages 2207--2211, 2018. doi:10.21437/Interspeech.2018-1456

    2018 doi

  26. [34]

    wen Yang, P.-H

    S. wen Yang, P.-H. Chi, Y.-S. Chuang, C.-I. J. Lai, K. Lakhotia, Y. Y. Lin, A. T. Liu, J. Shi, X. Chang, G.-T. Lin, T.-H. Huang, W.-C. Tseng, K. tik Lee, D.-R. Liu, Z. Huang, S. Dong, S.-W. Li, S. Watanabe, A. Mohamed, and H. yi Lee. SUPERB: Speech Processing Universal PERform...

    2021

  27. [35]

    Y.-Y. Yang, M. Hira, Z. Ni, A. Chourdia, A. Astafurov, C. Chen, C.-F. Yeh, C. Puhrsch, D. Pollack, D. Genzel, D. Greenberg, E. Z. Yang, J. Lian, J. Mahadeokar, J. Hwang, J. Chen, P. Goldsborough, P. Roy, S. Narenthiran, S. Watanabe, S. Chintala, V. Quenneville-Bélair, and Y. S...

    2022

  28. [36]

    Zhang, W

    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

    2023 arXiv

Pith tools

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