Pith. sign in

REVIEW 2 major objections 5 minor 52 references

FlanEC: Exploring Flan-T5 for Post-ASR Error Correction

T0 review · 2 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read A single Flan-T5 model fine-tuned on all eight HyPoradise domains with full parameter updates achieves the best overall post-ASR correction, reaching an average word error rate of 8.5 percent.

desk verdict A clean, reproducible sweep of Flan-T5 for post-ASR correction; the main rankings are plausible but single-run WERs leave the headline FT-over-LoRA and CD-over-SD claims thinner than they look. read the letter →

arxiv 2501.12979 v1 pith:ZSKGUDQ4 submitted 2025-01-22 cs.CL cs.AIcs.SDeess.AS

Moreno La Quatra , Valerio Mario Salerno , Yu Tsao , Sabato Marco Siniscalchi This is my paper · ORCID
classification cs.CLcs.AIcs.SDeess.AS
keywords Post-ASRerrorcorrectionGenerativespeechFlan-T5HyPoradiseEncoder-decodermodelLoRAFullfine-tuningWordrate
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to establish that a single instruction-tuned encoder-decoder model, Flan-T5, can serve as a general post-ASR error corrector across diverse speech domains. The authors train FlanEC at 250M, 800M, and 3B parameter scales on the HyPoradise benchmark, either on each domain separately or on all eight domains combined. They report that the 3B model trained on the combined data with full fine-tuning reaches an average word error rate of 8.5%, the best overall, and that cumulative training plus full fine-tuning generally beats per-domain training and LoRA-style adapters. The paper also finds that zero-shot Flan-T5 fails while fine-tuned FlanEC beats ChatGPT in-context learning, so task-specific parameter updates carry the result.

What carries the argument

The mechanism is the instruction-tuned encoder-decoder Flan-T5 adapted with a fixed natural-language prompt that maps an n-best list of ASR hypotheses into a single corrected transcription. The training design contrasts per-domain (SD) versus cumulative-dataset (CD) fine-tuning, and LoRA adapters versus full fine-tuning, at three model sizes; the key quantity is the word error rate on the HyPoradise test subsets, especially the proportion of reference tokens absent from all n-best hypotheses (new tokens), which motivates unconstrained decoding.

What would settle it

Train the same FlanEC-CD 3B model on HyPoradise but evaluate on n-best lists produced by a different ASR engine (e.g., a modern Whisper or an RNN-T model) across the same domains; if the cumulative model no longer beats per-domain models, or full fine-tuning no longer beats LoRA, the paper's central ranking fails to generalize.

Watch

Extended reading notes

Core claim

The central discovery is that cumulative training works: a single Flan-T5 model fine-tuned on all eight HyPoradise subsets jointly outperforms models trained on each subset individually, and full fine-tuning consistently beats LoRA once data is pooled. The 3B FlanEC-CD model with full fine-tuning tops 4 of 8 subsets and reaches the lowest average WER of 8.5%, versus 9.0% for its LoRA counterpart, while the 250M SD-LoRA models beat their FT peers on most subsets. The paper interprets this as evidence that large parameter scale and diverse training data let the model exploit shared linguistic knowledge, and that overfitting from full fine-tuning is a small-model, small-data phenomenon.

Load-bearing premise

The ranking assumes the HyPoradise benchmark faithfully represents real post-ASR correction: its fixed n-best lists, references, and splits come from one ASR setup, and if its error patterns or domains are not representative of other ASR systems, the reported advantages of cumulative training and full fine-tuning may not transfer.

Editorial extensions

If this is right

  • A single FlanEC-CD 3B model can serve as a general post-ASR corrector across WSJ, ATIS, CHiME-4, Tedlium-3, CV-accent, SwitchBoard, LRS2, and CORAAL, lowering average WER from the 11.8% baseline to 8.5%.
  • Cumulative training should be preferred over per-domain training for GenSEC when data from multiple domains is available, since CD matches or beats SD on most datasets.
  • Full fine-tuning, not parameter-efficient adapters, is the right choice for larger models and pooled data; LoRA only helps small models trained on single datasets where overfitting threatens.
  • Task-specific fine-tuning is necessary: zero-shot Flan-T5 does not beat the baseline, while fine-tuned FlanEC outperforms ChatGPT with in-context learning.
  • The CORAAL subset remains unsolved; no model reaches the ASR baseline there, and the paper attributes this to reference tokens missing from all n-best hypotheses in over 90% of sentences.

Reading between the lines

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

  • The success of cumulative training suggests that a single error-correction model could be shipped as a universal ASR post-processor, updated once on pooled data rather than maintained per domain.
  • The paper's comparison to ChatGPT ICL is indirect (no Flan-T5 ICL); a fairer test with an encoder-decoder ICL variant or a decoder-only model of matched scale would clarify whether the gap comes from architecture or fine-tuning.
  • A testable extension is to measure whether cumulative training helps most on domains with overlapping vocabulary or error patterns; the paper's new-token analysis (Table 1) implies CORAAL's failures stem from n-best coverage, which could be probed by injecting oracle hypotheses.
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.

Referee Report

2 major / 5 minor

Summary. The paper presents FlanEC, an encoder-decoder post-ASR error-correction model built on Flan-T5, and evaluates it on Task 1 of the GenSEC Grand Challenge using the HyPoradise benchmark. It compares single-dataset (SD) and cumulative-dataset (CD) training, full fine-tuning (FT) versus LoRA, and model scales from 250M to 3B parameters, reporting WER on eight ASR domains. The headline result is that the 3B FlanEC-CD model with full fine-tuning attains the best average WER (8.5%), that CD training generally helps in the fully fine-tuned setting, and that full fine-tuning generally beats LoRA except at the smallest scale in SD training. The paper also reports zero-shot Flan-T5 and ChatGPT in-context-learning baselines and honestly notes that no model improves over the baseline on the CORAAL subset.

Significance. If the comparative findings hold, the paper provides useful engineering evidence for the GenSEC community: a single instruction-tuned encoder-decoder model trained on combined domains can match or beat per-domain specialist models, and LoRA is not a full substitute for full fine-tuning at larger scales. The work has concrete strengths: it uses a public benchmark, releases code for reproducibility, evaluates multiple model sizes and adaptation schemes, and reports negative results on CORAAL rather than selecting around them. The main weakness is statistical: all comparative conclusions rest on single-run WER values with no error bars, confidence intervals, or significance tests, and several headline gaps are small relative to the resolution of the metric. The stress-test concern about test-set noise therefore lands: the central claim of consistent superiority of CD over SD and FT over LoRA is plausible but not established at the reported resolution.

major comments (2)
  1. [Section 5, 'LoRA vs. Full Fine-tuning'] The sentence 'full fine-tuning (FT) outperforms LoRA across all datasets' is contradicted by Table 2 in the CD setting. For example, at 250M, ATIS LoRA gives 2.7 versus FT 3.0 and LRS2 LoRA gives 10.0 versus FT 10.6; at 800M, LRS2 LoRA gives 9.3 versus FT 9.6 and Tedlium-3 is a tie at 4.3. Please revise this to 'outperforms LoRA on most datasets' or provide a corrected quantitative statement that matches the table.
  2. [Section 5, 'LoRA vs. Full Fine-tuning'] The reported 'average improvement of 0.64 ± 0.46 absolute WER points' is not defined. If the ± value is the standard deviation across the eight subsets, it does not by itself convey statistical significance; if it is a standard error or confidence interval, the computation should be stated. In either case, the paired comparison and its test statistic should be reported so that the reader can assess whether the FT advantage is robust.
minor comments (5)
  1. [Section 4.2] Please report the LoRA rank and alpha; stating that adapters are added to all fully connected layers is not sufficient for reproducibility.
  2. [Section 4.2] Please state how the validation split is constructed and how 'best model based on validation WER' interacts with the different epoch budgets for SD (10) and CD (2); otherwise the SD/CD comparison is not fully specified.
  3. [Table 1] There are spacing typos in the dataset names ('A TIS', 'T edlium-3'), and the caption should state that the 'Overall' row is an unweighted macro-average over the eight subsets.
  4. [Section 3.1] The prompt template does not include the confidence scores that the HyPoradise n-best lists provide, although Section 1 mentions associated confidence scores; please clarify whether confidence scores were used as model input.
  5. [Abstract and Section 5] The abstract claims improvements in grammaticality and linguistic correctness, but the evaluation reports only WER; either add a text-quality or grammaticality metric or soften the language to match the reported evaluation.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: all central claims are empirical evaluations on the fixed, externally defined HyPoradise benchmark.

full rationale

The paper's central claims are empirical comparisons of fine-tuned Flan-T5 variants on the fixed HyPoradise test sets. The model is trained to map n-best hypotheses to reference transcriptions, and the reported WER numbers in Table 2 are obtained from held-out test subsets, with model selection performed on validation WER rather than on the test data. No parameter is fitted to the test set and no reported 'prediction' is defined in terms of the quantity it is supposed to predict. Table 1 is a descriptive analysis of new tokens and new sentences, used only to motivate unconstrained decoding; it is not a fitted input passed off as a result. The only notable self-citation is the HyPoradise benchmark itself, which is co-authored by one of the current authors, but the dataset is a public, externally fixed benchmark and the paper's results are falsifiable against it; the citation is therefore real evidence rather than load-bearing circularity. The absence of confidence intervals or significance tests is a legitimate correctness and robustness concern, but it is not a circularity concern. Accordingly, the paper receives a score of 0.

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

The central claim is entirely empirical; no fitted physical constants or invented entities are introduced. Training hyperparameters are largely reported, but the LoRA configuration is incomplete and the benchmark itself is adopted as a domain assumption.

free parameters (2)
  • n-best list size n = 5
    Chosen based on prior work [23], not fitted to the test set; included for completeness.
  • LoRA rank and alpha = not reported
    LoRA adapters are applied to all fully connected layers, but rank and alpha, which control adaptation capacity, are not stated; performance may depend on these unreported choices.
assumptions (4)
  • domain assumption HyPoradise is a valid benchmark for post-ASR error correction
    Section 4.1 adopts the eight subsets and their n-best lists as the evaluation basis without independent validation that they represent real-world ASR error patterns.
  • domain assumption WER on the HyPoradise test sets is the correct measure of correction quality
    Section 4.2 defines evaluation solely by WER; other error types or downstream task impact are not measured.
  • domain assumption The fixed ASR n-best lists are representative of the errors a post-corrector will face
    Section 4.1 uses n-best lists generated by one ASR model; transfer to other ASR systems is assumed, not tested.
  • domain assumption Flan-T5's instruction tuning transfers to the GenSEC prompt
    Section 3.1 assumes the natural-language prompt is sufficient to activate the model's correction behavior.

how reviews work

0 comments
Cite this review

Pith. "Pith review of FlanEC: Exploring Flan-T5 for Post-ASR Error Correction." pith.science (2026). https://pith.science/paper/ZSKGUDQ4

@misc{pith2026250112979,
  author       = {Pith},
  title        = {Pith review of: FlanEC: Exploring Flan-T5 for Post-ASR Error Correction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZSKGUDQ4}},
  note         = {Machine review of arXiv:2501.12979}
}
read the original abstract

In this paper, we present an encoder-decoder model leveraging Flan-T5 for post-Automatic Speech Recognition (ASR) Generative Speech Error Correction (GenSEC), and we refer to it as FlanEC. We explore its application within the GenSEC framework to enhance ASR outputs by mapping n-best hypotheses into a single output sentence. By utilizing n-best lists from ASR models, we aim to improve the linguistic correctness, accuracy, and grammaticality of final ASR transcriptions. Specifically, we investigate whether scaling the training data and incorporating diverse datasets can lead to significant improvements in post-ASR error correction. We evaluate FlanEC using the HyPoradise dataset, providing a comprehensive analysis of the model's effectiveness in this domain. Furthermore, we assess the proposed approach under different settings to evaluate model scalability and efficiency, offering valuable insights into the potential of instruction-tuned encoder-decoder models for this task.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

52 extracted references · 48 canonical work pages

  1. [23]

    The CoNLL-2014 shared task on grammatical error correction,

    Hwee Tou Ng, Siew Mei Wu, Ted Briscoe, Christian Hadiwinoto, Raymond Hendy Susanto, and Christopher Bryant, “The CoNLL-2014 shared task on grammatical error correction,” in Proceedings of the Eighteenth Con- ference on Computational Natural Language Learning: Shared T ask, Baltimore, Maryland, June 2014, pp. 1–14, Association for Computational Linguistics

    work page 2014

  2. [1]

    In Automatic Speech Recognition (ASR) systems, GenSEC acts as an effective post-processing step to enhance the accuracy [1, 2]

    INTRODUCTION Generative Speech Error Correction (GenSEC) focuses on correcting linguistic errors in transcriptions of spoken l an- guage. In Automatic Speech Recognition (ASR) systems, GenSEC acts as an effective post-processing step to enhance the accuracy [1, 2]. This step is crucial when ASR systems are used as inputs to downstream applications such as...

    work page

  3. [2]

    LLMs have shown remarkable capabilities in understanding and genera t- ing natural language, making them well-suited for GenSEC tasks [10, 11, 5]

    POST -ASR ERROR CORRECTION Correcting linguistic errors is a fundamental challenge in NLP and has traditionally focused on written text, with signific ant advancements documented over the years [9]. LLMs have shown remarkable capabilities in understanding and genera t- ing natural language, making them well-suited for GenSEC tasks [10, 11, 5]. By integrati...

    work page

  4. [3]

    We define a specific instruction template and use it to create FlanEC us - ing HyPoradise dataset and leveraging the n-best lists gener- ated by an ASR model

    FLANEC The proposed FlanEC is an encoder-decoder model built on Flan-T5, specifically instructed for GenSEC tasks. We define a specific instruction template and use it to create FlanEC us - ing HyPoradise dataset and leveraging the n-best lists gener- ated by an ASR model. 3.1. T ask prompt The prompt for FlanEC is designed to instruct the model to generate ...

    work page

  5. [4]

    We adapt the model at various scales, ranging from 250 mil- lion to 3 billion parameters, to assess the impact of scaling model parameters on GenSEC performance

    EXPERIMENTS We evaluate FlanEC on the HyPoradise dataset, focusing on post-ASR error correction across different ASR domains. We adapt the model at various scales, ranging from 250 mil- lion to 3 billion parameters, to assess the impact of scaling model parameters on GenSEC performance. We compare the model’s performance under different training settings ...

    work page

  6. [5]

    WSJ [24]: Business news and financial data readings

    work page

  7. [6]

    ATIS [25]: Airline Travel Information Systems queries

    work page

  8. [7]

    CHiME-4 [26]: Noisy speech data from different envi- ronments, part of the CHiME-4 challenge

    work page

Show all 52 references
  1. [8]

    T edlium-3[27]: Transcripts from TED talks

  2. [9]

    CV-accent [28]: Accented speech data from the English portion of the Common V oice dataset

  3. [10]

    SwitchBoard [29]: Telephone conversation transcripts

  4. [11]

    LRS2 [30]: Audio clips from BBC programs

  5. [12]

    The diversity of the data ensures that models are evaluated across a wide range of ASR domains, providing a compre- hensive assessment of their capabilities

    CORAAL [31]: Interviews containing accented speech from regional varieties of African American English. The diversity of the data ensures that models are evaluated across a wide range of ASR domains, providing a compre- hensive assessment of their capabilities. Each subset con...

  6. [13]

    The results ar e summarized in Table 2, where we compare the WER scores of FlanEC under different training settings, model sizes, a nd fine-tuning strategies

    RESUL TS AND DISCUSSION In this section, we present and analyze the performance of FlanEC on the HyPoradise dataset, focusing on post-ASR er- ror correction across different ASR domains. The results ar e summarized in Table 2, where we compare the WER scores of FlanEC under di...

  7. [14]

    CONCLUSIONS We introduced FlanEC, an encoder-decoder model based on Flan-T5, for post-ASR error correction. Extensive experi- ments on the HyPoradise dataset demonstrate that FlanEC- CD, trained on a combined dataset, generally outperforms single dataset models, especially whe...

  8. [15]

    A comparison of techniques for language model inte- gration in encoder-decoder speech recognition,

    Shubham Toshniwal, Anjuli Kannan, Chung-Cheng Chiu, Y onghui Wu, Tara N Sainath, and Karen Livescu, “A comparison of techniques for language model inte- gration in encoder-decoder speech recognition,” in 2018 IEEE spoken language technology workshop (SLT) . IEEE, 2018, pp. 369–375

    2018

  9. [16]

    An analysis of incorporating an external language model into a sequence-to-sequence model,

    Anjuli Kannan, Y onghui Wu, Patrick Nguyen, Tara N Sainath, Zhijeng Chen, and Rohit Prabhavalkar, “An analysis of incorporating an external language model into a sequence-to-sequence model,” in 2018 IEEE In- ternational Conference on Acoustics, Speech and Signal Processing (IC...

    2018

  10. [17]

    Mondegreen: A post-processing so- lution to speech recognition error correction for voice search queries,

    Sukhdeep S Sodhi, Ellie Ka-In Chio, Ambarish Jash, Santiago Onta˜ n´ on, Ajit Apte, Ankit Kumar, A y- ooluwakunmi Jeje, Dima Kuzmin, Harry Fung, Heng- Tze Cheng, et al., “Mondegreen: A post-processing so- lution to speech recognition error correction for voice search queries,”...

    2021

  11. [18]

    Robust speech recognition via large-scale weak supervision,

    Alec Radford, Jong Wook Kim, Tao Xu, Greg Brock- man, Christine McLeavey, and Ilya Sutskever, “Robust speech recognition via large-scale weak supervision,” in International Conference on Machine Learning. PMLR, 2023, pp. 28492–28518

    2023

  12. [19]

    Scaling instruction-finetuned language models,

    Hyung Won Chung, Le Hou, Shayne Longpre, Bar- ret Zoph, Yi Tay, William Fedus, Y unxuan Li, Xuezhi Wang, Mostafa Dehghani, Siddhartha Brahma, et al., “Scaling instruction-finetuned language models,” Jour- nal of Machine Learning Research , vol. 25, no. 70, pp. 1–53, 2024

    2024

  13. [20]

    CoEdIT: Text editing by task-specific instruc- tion tuning,

    Vipul Raheja, Dhruv Kumar, Ryan Koo, and Dongyeop Kang, “CoEdIT: Text editing by task-specific instruc- tion tuning,” in Findings of the Association for Com- putational Linguistics: EMNLP 2023 , Singapore, Dec. 2023, pp. 5274–5291, Association for Computational Linguistics

    2023

  14. [21]

    Hyporadise: An open baseline for generative speech recognition with large language models,

    Chen Chen, Y uchen Hu, Chao-Han Huck Y ang, Sabato Marco Siniscalchi, Pin-Y u Chen, and Eng-Siong Chng, “Hyporadise: An open baseline for generative speech recognition with large language models,” Ad- vances in Neural Information Processing Systems , vol. 36, 2024

    2024

  15. [22]

    Lora: Low-rank adaptation of large lan- guage models,

    Edward J Hu, Y elong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Y uanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen, “Lora: Low-rank adaptation of large lan- guage models,” arXiv preprint arXiv:2106.09685, 2021

    2021 arXiv

  16. [24]

    Llama: Open and efficient foundation lan- guage models,

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timoth´ ee Lacroix, Baptiste Rozi` ere, Naman Goyal, Eric Hambro, Faisal Azhar, et al., “Llama: Open and efficient foundation lan- guage models,” arXiv preprint arXiv:2302.13971, 2023

    2023 arXiv

  17. [25]

    Language models are few-shot learners,

    Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al., “Language models are few-shot learners,” Advances in neural information processing systems, vol. 33, pp. 1877–1901, 2020

    1901

  18. [26]

    Genera- tive speech recognition error correction with large lan- guage models and task-activating prompting,

    Chao-Han Huck Y ang, Yile Gu, Yi-Chieh Liu, Shalini Ghosh, Ivan Bulyko, and Andreas Stolcke, “Genera- tive speech recognition error correction with large lan- guage models and task-activating prompting,” in 2023 IEEE Automatic Speech Recognition and Understand- ing W orkshop (...

    2023

  19. [27]

    Cross-task generalization via natural language crowdsourcing instructions,

    Swaroop Mishra, Daniel Khashabi, Chitta Baral, and Hannaneh Hajishirzi, “Cross-task generalization via natural language crowdsourcing instructions,” in Pro- ceedings of the 60th Annual Meeting of the Association for Computational Linguistics (V olume 1: Long Papers), Dublin, I...

    2022

  20. [28]

    A survey for in-context learning,

    Qingxiu Dong, Lei Li, Damai Dai, Ce Zheng, Zhiyong Wu, Baobao Chang, Xu Sun, Jingjing Xu, Lei Li, and Zhifang Sui, “A survey for in-context learning,” 2022

    2022

  21. [29]

    Chain-of-thought prompting elicits reasoning in large language models,

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, et al., “Chain-of-thought prompting elicits reasoning in large language models,” Advances in neural information pro- cessing systems, vol. 35, pp. 24824–24837, 2022

    2022

  22. [30]

    Can whisper perform speech-based in-context learn- ing?,

    Siyin Wang, Chao-Han Y ang, Ji Wu, and Chao Zhang, “Can whisper perform speech-based in-context learn- ing?,” in ICASSP 2024-2024 IEEE International Con- ference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2024, pp. 13421–13425

    2024

  23. [31]

    Emergent abilities of large language models,

    Jason Wei, Yi Tay, Rishi Bommasani, Colin Raffel, Bar- ret Zoph, Sebastian Borgeaud, Dani Y ogatama, Maarten Bosma, Denny Zhou, Donald Metzler, Ed H. Chi, Tat- sunori Hashimoto, Oriol Vinyals, Percy Liang, Jeff Dean, and William Fedus, “Emergent abilities of large language mod...

    2022

  24. [32]

    Large language models are efficient learners of noise-robust speech recognition,

    Y uchen Hu, CHEN CHEN, Chao-Han Huck Y ang, Ruizhe Li, Chao Zhang, Pin-Y u Chen, and EngSiong Chng, “Large language models are efficient learners of noise-robust speech recognition,” in The Twelfth International Conference on Learning Representations , 2024

    2024

  25. [33]

    Towards asr robust spoken language understand- ing through in-context learning with word confusion networks,

    Kevin Everson, Yile Gu, Huck Y ang, Prashanth Gu- runath Shivakumar, Guan-Ting Lin, Jari Kolehmainen, Ivan Bulyko, Ankur Gandhe, Shalini Ghosh, Wael Hamza, Hung-Yi Lee, Anya Rastrow, and Andreas Stol- cke, “Towards asr robust spoken language understand- ing through in-context ...

    2024

  26. [34]

    Few-shot fine-tuning vs. in-context learning: A fair comparison and evalua- tion,

    Marius Mosbach, Tiago Pimentel, Shauli Ravfogel, Di- etrich Klakow, and Y anai Elazar, “Few-shot fine-tuning vs. in-context learning: A fair comparison and evalua- tion,” in Findings of the Association for Computational Linguistics: ACL 2023 , Toronto, Canada, July 2023, pp. 12...

    2023

  27. [35]

    Exploring the limits of transfer learning with a unified text-to-text transformer,

    Adam Roberts, Colin Raffel, Katherine Lee, Michael Matena, Noam Shazeer, Peter J Liu, Sharan Narang, Wei Li, and Y anqi Zhou, “Exploring the limits of transfer learning with a unified text-to-text transformer, ” Google, T ech. Rep., 2019

    2019

  28. [36]

    InstructEval: Towards holistic evaluation of instruction-tuned large language models,

    Y ew Ken Chia, Pengfei Hong, Lidong Bing, and Sou- janya Poria, “InstructEval: Towards holistic evaluation of instruction-tuned large language models,” in Pro- ceedings of the First edition of the W orkshop on the Scaling Behavior of Large Language Models (SCALE- LLM 2024) , S...

    2024

  29. [37]

    N-best T5: Robust ASR Error Correction using Multiple Input Hypotheses and Constrained Decoding Space,

    Rao Ma, Mark J. F. Gales, Kate M. Knill, and Mengjie Qian, “N-best T5: Robust ASR Error Correction using Multiple Input Hypotheses and Constrained Decoding Space,” in Proc. INTERSPEECH 2023, 2023, pp. 3267– 3271

    2023

  30. [38]

    The design for the Wall Street Journal-based CSR corpus,

    Douglas B. Paul and Janet M. Baker, “The design for the Wall Street Journal-based CSR corpus,” in Speech and Natural Language: Proceedings of a W orkshop Held at Harriman, New Y ork, February 23-26, 1992, 1992

    1992

  31. [39]

    The A TIS spoken language systems pi- lot corpus,

    Charles T. Hemphill, John J. Godfrey, and George R. Doddington, “The A TIS spoken language systems pi- lot corpus,” in Speech and Natural Language: Proceed- ings of a W orkshop Held at Hidden V alley, Pennsylvania, June 24-27,1990, 1990

    1990

  32. [40]

    The 4th chime speech separation and recognition challenge,

    Emmanuel Vincent, Shinji Watanabe, Jon Barker, and Ricard Marxer, “The 4th chime speech separation and recognition challenge,” URL: http://spandh. dcs. shef. ac. uk/chime challenge {Last Accessed on 1 August, 2018}, 2016

    2018

  33. [41]

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

    Franc ¸ois Hernandez, Vincent Nguyen, Sahar Ghannay, Natalia Tomashenko, and Y annick Esteve, “Ted-lium 3: Twice as much data and corpus repartition for ex- periments on speaker adaptation,” in Speech and Com- puter: 20th International Conference, SPECOM 2018, Leipzig, Germany...

    2018

  34. [42]

    Springer, 2018, pp. 198–208

    2018

  35. [43]

    Common voice: A massively-multilingual speech cor- pus,

    Rosana Ardila, Megan Branson, Kelly Davis, Michael Kohler, Josh Meyer, Michael Henretty, Reuben Morais, Lindsay Saunders, Francis Tyers, and Gregor Weber, “Common voice: A massively-multilingual speech cor- pus,” in Proceedings of the Twelfth Language Resources and Evaluation ...

    2020

  36. [44]

    Switchboard: Telephone speech corpus for re- search and development,

    John J Godfrey, Edward C Holliman, and Jane Mc- Daniel, “Switchboard: Telephone speech corpus for re- search and development,” in Acoustics, speech, and sig- nal processing, ieee international conference on . IEEE Computer Society, 1992, vol. 1, pp. 517–520

    1992

  37. [45]

    Lip reading sentences in the wild,

    Joon Son Chung, Andrew Senior, Oriol Vinyals, and Andrew Zisserman, “Lip reading sentences in the wild,” in Proceedings of the IEEE conference on computer vi- sion and pattern recognition , 2017, pp. 6447–6456

    2017

  38. [46]

    The corpus of regional african american language,

    Tyler Kendall and Charlie Farrington, “The corpus of regional african american language,” V ersion, vol. 6, pp. 1, 2018

    2018

  39. [47]

    Decoupled weight decay regularization,

    Ilya Loshchilov and Frank Hutter, “Decoupled weight decay regularization,” in International Conference on Learning Representations, 2018

    2018

  40. [48]

    Transformers: State-of-the-art natu- ral language processing,

    Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pierric Cistac, Tim Rault, Remi Louf, Morgan Funtowicz, Joe Davison, Sam Shleifer, Patrick von Platen, Clara Ma, Y acine Jernite, Julien Plu, Canwen Xu, Teven Le Scao, Sylvain Gugger, Mar...

    2020

  41. [49]

    Few-shot parameter-efficient fine-tuning is bet- ter and cheaper than in-context learning,

    Haokun Liu, Derek Tam, Mohammed Muqeeth, Jay Mohta, Tenghao Huang, Mohit Bansal, and Colin A Raffel, “Few-shot parameter-efficient fine-tuning is bet- ter and cheaper than in-context learning,” in Advances in Neural Information Processing Systems , S. Koyejo, S. Mohamed, A. Agar...

    2022

  42. [50]

    Racial disparities in automated speech recognition,

    Allison Koenecke, Andrew Nam, Emily Lake, Joe Nudell, Minnie Quartey, Zion Mengesha, Connor Toups, John R Rickford, Dan Jurafsky, and Sharad Goel, “Racial disparities in automated speech recognition,” Proceedings of the national academy of sciences , vol. 117, no. 14, pp. 7684...

    2020

  43. [51]

    Uneven success: automatic speech recognition and ethnicity-related dialects,

    Alicia Beckford Wassink, Cady Gansen, and Isabel Bartholomew, “Uneven success: automatic speech recognition and ethnicity-related dialects,” Speech Com- munication, vol. 140, pp. 50–70, 2022

    2022

  44. [52]

    Quantifying the role of textual predictabil- ity in automatic speech recognition,

    Sean Robertson, Gerald Penn, and Ewan Dun- bar, “Quantifying the role of textual predictabil- ity in automatic speech recognition,” arXiv preprint arXiv:2407.16537, 2024

    2024 arXiv

Pith tools

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