Pith. sign in

REVIEW 3 major objections 5 minor 110 references

Cross-attention scores in encoder-decoder speech-to-text models capture only about half of the input relevance identified by perturbation-based saliency maps, so the paper argues they cannot stand alone as explanations.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-04 15:45 UTC pith:C66DJQH3

load-bearing objection First systematic test of cross-attention as explanation in speech-to-text; the qualitative conclusion is solid, but the headline percentages overstate the precision and the reference saliency method is not independent. the 3 major comments →

arxiv 2509.18010 v2 pith:C66DJQH3 submitted 2025-09-22 cs.CL cs.AIcs.SD

Cross-Attention is Half Explanation in Speech-to-Text Models

Sara Papi , Dennis Fucci , Marco Gaido , Matteo Negri , Luisa Bentivogli This is my paper
classification cs.CL cs.AIcs.SD
keywords cross-attentionspeech-to-textexplainable AIsaliency mapsfeature attributioncontext mixingautomatic speech recognitionspeech translation
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper tests a routine assumption in speech-to-text models: that the cross-attention weights connecting the decoder to the encoder show which parts of the audio actually drove each generated word. It compares those weights against saliency maps produced by a perturbation-based feature-attribution method, across monolingual and multilingual models, ASR and speech-translation tasks, and three model sizes. It finds that cross-attention correlates moderately to strongly with input saliency, especially when scores are averaged over heads and layers, but still accounts for only about half of the input relevance. Even when compared to saliency computed directly on encoder outputs, the numbers reach only 52–75% of what the saliency reference shows. The conclusion is practical: cross-attention is a cheap, partial signal that can complement attribution methods, but it should not be treated as a stand-alone explanation, especially in applications like timestamp estimation and alignment.

Core claim

Central claim: in encoder-decoder speech-to-text models, cross-attention is only a partial explanation. On the monolingual ASR model the best head/layer aggregation correlates at about 0.59 with input saliency; on multitask multilingual models layer-averaged correlations run 0.49–0.63, which the paper reads as 49–63% of input relevance. Against encoder-output saliency—where context mixing is no longer a confound—correlations rise by 0.03–0.18 but plateau at 52–75%. A deletion check on the base model sharpens the point: cross-attention scores 41.2, frequency-aggregated saliency 52.9, full-resolution saliency 91.3. The paper takes this as evidence that cross-attention is an auxiliary cue, not

What carries the argument

The load-bearing comparison objects are the cross-attention matrix, extracted from each decoder layer and head during generation, and two saliency references: input saliency maps, obtained by perturbing spectrogram regions and measuring KL divergence in the output distribution, and encoder-output saliency maps, computed by perturbing encoder hidden states. The cross-attention matrix is averaged over selected heads and layers; saliency maps are aggregated across frequency and time to the same temporal resolution. The correlations between flattened attention and saliency vectors are measured with Pearson correlation, and the difference between input-level and encoder-output-level correlations

Load-bearing premise

The central numbers depend on treating the perturbation-based saliency maps as a silver reference for what the model truly uses; if that reference is biased, the measured gap could be an artifact of the reference rather than a property of cross-attention, as the paper itself acknowledges in its limitations section.

What would settle it

Run the same correlation and deletion pipeline with an independent reference—manually aligned word or phoneme boundaries, or a second feature-attribution method—and check whether cross-attention matches that reference far better than it matches the paper's saliency baseline. In particular, an attention-based deletion score near the full-resolution saliency score (91.3 on the paper's base model, versus 41.2 for attention) would falsify the 'half explanation' conclusion.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Downstream tools that use cross-attention for timestamp prediction, audio-text alignment, or guided simultaneous translation should stop relying on a single head or layer; averaging over heads and, where possible, layers is closer to the saliency reference.
  • Because cross-attention aligns better with encoder-output saliency than with input saliency, attention-based explanations are better read as descriptions of how the decoder uses encoded representations, not of which raw audio regions matter.
  • Cross-attention cannot replace feature-attribution methods for faithful explanations; its upper range of 52–75% against encoder-output saliency leaves too much unexplained for it to stand alone.
  • The last decoder layers are consistently the most aligned with saliency across tasks, languages, and model sizes, so layer choice is consequential if attention is used at all.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • A plausible editorial extension: if the ~50% gap holds, attention-map visualizations shipped in real products (e.g., timestamped transcriptions) should carry a confidence caveat and be cross-checked against forced alignment rather than presented as ground truth.
  • A natural extension the paper mentions but does not test is to train with an auxiliary loss that pulls cross-attention toward saliency maps or monotonic alignment; the results here predict that such regularization would improve both faithfulness and downstream word-level accuracy.
  • The paper deliberately excludes speech-LLM architectures; since those models add a second encoder-decoder boundary, the context-mixing effect measured here could be larger there, making cross-attention an even weaker explanation for that family.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper investigates whether cross-attention scores in encoder-decoder speech-to-text (S2T) models can serve as explanations. It compares cross-attention to SPES-based input saliency maps and to an ad-hoc encoder-output saliency variant across monolingual ASR, multilingual/multitask ASR+ST models, at three scales and two languages. The main findings are that averaged cross-attention correlates moderately with input saliency (up to about 0.63) and more strongly with encoder-output saliency (up to about 0.75), but that cross-attention still 'captures only about 50% of input relevance' and only 52-75% of encoder-output saliency. The paper concludes that cross-attention is an informative but incomplete explanation proxy and should not be used as a stand-alone XAI tool.

Significance. If the headline quantitative claim is reliable, this is a useful and timely contribution: it directly tests a widely made assumption in speech-to-text research and has implications for timestamp estimation, alignment, and simultaneous translation. The experimental design has clear strengths: models are trained from scratch to avoid contamination; the analysis spans multiple tasks, languages, and scales; aggregation choices are ablated; and the deletion metric provides a behavioral check on explanation quality. The paper is also transparent about its limitations. The main concern is that the central percentages rest on a single attribution reference and on an invalid reading of Pearson correlation as a proportion of relevance.

major comments (3)
  1. [Section 5.1.2, Section 5.2, Tables 2-3] The paper converts Pearson r values directly into percentages of 'total input relevance' and 'saliency' (e.g., '49-63%' and '52-75%' in Sections 5.1.2 and 5.2). A Pearson correlation is not a share of relevance; the proportion of shared variance is r^2. Under the reported numbers, r=0.63 corresponds to roughly 40% shared variance and r=0.75 to roughly 56%. Even r^2 would require the additional assumption that shared variance equals captured relevance. The quantitative headline in the abstract and conclusions is therefore not supported by the statistic used. Please reframe these percentages or replace them with a directly interpretable metric such as the deletion AUC already used in Appendix A. The qualitative conclusion may survive the correction, but the exact numbers must be revised.
  2. [Section 3.2, Appendix F] The load-bearing claim that cross-attention 'captures only about 50% of the input relevance' is measured entirely against SPES, and the encoder-output claim is measured against SMH, an ad-hoc variant of SPES. SPES comes from the same research group (Fucci et al., 2025, with overlapping authorship), and SMH is validated only by the dev-set deletion experiment in Appendix D. If SPES or SMH is biased or incomplete, the measured gap between cross-attention and saliency is not a faithful measure of cross-attention's explanatory power. The limitation appendix acknowledges this in principle, but the abstract and conclusions present the gap as a property of cross-attention rather than as a comparison relative to one attribution method. Please add at least one independent attribution baseline (e.g., a gradient-based or perturbation method not developed by the same group) or a sensitivity analysis
  3. [Section 5.2, first paragraph] The 'context mixing' influence is quantified as '6.6-16.7%' from absolute differences in Pearson r between SMX and SMH. An absolute difference in correlation coefficients is not a percentage of explained behavior, and no derivation is provided to link a change in r to a fraction of context mixing. This should either be derived formally or, more conservatively, reported simply as a range of correlation improvements.
minor comments (5)
  1. [Title/Abstract] The title is typeset as 'CROSS-ATTENTION ISHALF EXPLANATION' in the preprint; there should be a space between 'IS' and 'HALF'.
  2. [Table 5] The row for 'Whisper 10.69.0- 0.797 - -' appears malformed; the table columns do not align cleanly for the comparison models.
  3. [Figure 4] Panel (i) is labeled 'CA(enASR)' but appears in the row for it-ASR; this is likely a typo and should be corrected.
  4. [Appendix E] There is a duplicated phrase: 'this pattern this pattern is absent' should read 'this pattern is absent'.
  5. [Tables 2 and 3] The captions read 'Person ρ' instead of 'Pearson ρ'.

Circularity Check

0 steps flagged

No circular derivation; empirical comparison with a self-cited baseline is a limitation, not a circularity.

full rationale

The paper's central claims are empirical measurements: Pearson correlations between cross-attention matrices (CA) and saliency maps (SMX, SMH). The saliency reference is SPES (Fucci et al., 2025), a method developed by overlapping authors, which the paper explicitly calls a 'silver reference' in Appendix F. This raises a legitimate concern about the independence of the gold standard, but it does not constitute circularity in the technical sense. CA is computed as a softmax of query-key dot products (Section 3.1), while SMX and SMH are produced by perturbation-based KL divergence (Section 3.2). There is no equation that defines one in terms of the other, and no fitted parameter is renamed as a prediction. The choice of aggregation function (Appendix A) and the occlusion probability pH (Appendix D) are tuned on a dev set, and the reported correlations are computed on a held-out test set, so the headline percentages are not forced by construction. The paper also provides deletion-metric validation of the reference explanations (Appendix C, Table 5). The self-citation to SPES is load-bearing only as the choice of reference method; the central derivation—the observed gap between CA and saliency—is empirical and could have produced different values. The acknowledged limitation in Appendix F is a validity threat, not a circularity. Therefore, no significant circularity is present.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 0 invented entities

The central measurement is an empirical comparison, not a derivation. It depends on the truthfulness of the reference saliency maps, the aggregation and normalization choices, and the statistical metric used to quantify agreement. No new physical entities or fitted theoretical constants are introduced.

free parameters (4)
  • p_H (encoder-output occlusion probability) = 0.7
    Tuned on the dev set to maximize deletion score and CA-SMH correlation (Appendix D); directly shapes the SMH reference maps.
  • p_X (input occlusion probability) = 0.5
    Inherited from SPES configuration; probability of masking spectrogram clusters when computing SMX.
  • SMX aggregation function = 2D max pooling
    Selected on dev set via deletion metric among three pooling options (Appendix A); changes the SMX values and thus all input correlations.
  • N_X / N_H perturbation repeats = 20,000
    Fixed following SPES; larger values reduce noise but do not change the method.
axioms (4)
  • domain assumption SPES saliency maps are a valid 'silver' reference for input and encoder-output relevance.
    The paper states no gold reference exists and adopts SPES as the state of the art (Section 3.2, Appendix F). If SPES is biased, the measured gap is not true explanatory power.
  • domain assumption Pearson correlation between flattened CA and SM vectors measures alignment of relevance patterns.
    Section 3.3 justifies Pearson over rank-based metrics, but the results section reads r as a proportion of explained relevance, conflating r with r-squared.
  • domain assumption Zero-perturbation of encoder hidden states or spectrogram patches faithfully estimates each feature's contribution.
    Perturbation-based attribution assumes that removing a feature and measuring output change approximates the feature's causal contribution.
  • domain assumption Cross-attention scores operate over encoder states, so comparing them with SMH (computed on the same states) is the correct isolation of context mixing.
    Section 5.2 interprets the increase from SMX to SMH correlation as quantifying context mixing; this assumes no other confound differs between the two saliency computations.

pith-pipeline@v1.3.0-alltime-deepseek · 25928 in / 10704 out tokens · 83788 ms · 2026-08-04T15:45:58.856757+00:00 · methodology

0 comments
read the original abstract

Cross-attention is a core mechanism in encoder-decoder architectures, widespread in many fields, including speech-to-text (S2T) processing. Its scores have been repurposed for various downstream applications--such as timestamp estimation and audio-text alignment--under the assumption that they reflect the dependencies between input speech representation and the generated text. While the explanatory nature of attention mechanisms has been widely debated in the broader NLP literature, this assumption remains largely unexplored within the speech domain. To address this gap, we assess the explanatory power of cross-attention in S2T models by comparing its scores to input saliency maps derived from feature attribution. Our analysis spans monolingual and multilingual, single-task and multi-task models at multiple scales, and shows that attention scores moderately to strongly align with saliency-based explanations, particularly when aggregated across heads and layers. However, it also shows that cross-attention captures only about 50% of the input relevance and, in the best case, only partially reflects how the decoder attends to the encoder's representations--accounting for just 52-75% of the saliency. These findings uncover fundamental limitations in interpreting cross-attention as an explanatory proxy, suggesting that it offers an informative yet incomplete view of the factors driving predictions in S2T models.

Figures

Figures reproduced from arXiv: 2509.18010 by Dennis Fucci, Luisa Bentivogli, Marco Gaido, Matteo Negri, Sara Papi.

Figure 1
Figure 1. Figure 1: Visual representation of which part of the model is covered by [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Input (a) and encoder output (b) saliency maps and cross-attention matrix (c) extracted from [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: An example of SMX maps for the predicted sentence “What is important are the options, not quantity”. The frequency axis is represented in Hertz on a logarithmic scale. These results indicate that global averaging over time and frequency may obscure localized salient regions, and this is particularly impactful in the frequency dimension, where preserving saliency seems to play a crucial role. This is due to… view at source ↗
Figure 4
Figure 4. Figure 4: Example of input (first column) and encoder output (second column) saliency maps and [PITH_FULL_IMAGE:figures/full_fig_p025_4.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

110 extracted references · 22 canonical work pages

  1. [1]

    online" 'onlinestring :=

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block STRING...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    Uncovering syllable constituents in the self-attention-based speech representations of whisper

    Erfan A Shams, Iona Gessinger, and Julie Carson-Berndsen. Uncovering syllable constituents in the self-attention-based speech representations of whisper. In Proceedings of the 7th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP, pp.\ 238--247, Miami, Florida, US, November 2024. doi:10.18653/v1/2024.blackboxnlp-1.16. URL https://ac...

    doi:10.18653/v1/2024.blackboxnlp-1.16 2024

  4. [4]

    Quantifying attention flow in transformers

    Samira Abnar and Willem Zuidema. Quantifying attention flow in transformers. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pp.\ 4190--4197, Online, July 2020. doi:10.18653/v1/2020.acl-main.385. URL https://aclanthology.org/2020.acl-main.385/

    doi:10.18653/v1/2020.acl-main.385 2020

  5. [5]

    Slic superpixels compared to state-of-the-art superpixel methods

    Radhakrishna Achanta, Appu Shaji, Kevin Smith, Aurelien Lucchi, Pascal Fua, and Sabine Süsstrunk. Slic superpixels compared to state-of-the-art superpixel methods. IEEE Transactions on Pattern Analysis and Machine Intelligence, 34 0 (11): 0 2274--2282, 2012. doi:10.1109/TPAMI.2012.120

    doi:10.1109/tpami.2012.120 2012

  6. [6]

    Common voice: A massively-multilingual speech corpus

    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 corpus. In Proceedings of the 12th Language Resources and Evaluation Conference, pp.\ 4218--4222, Marseille, France, May 2020. ISBN 979-10-95546-34-4. URL h...

    2020

  7. [7]

    Kartik Audhkhasi, Yinghui Huang, Bhuvana Ramabhadran, and Pedro J. Moreno. Analysis of self-attention head diversity for conformer-based automatic speech recognition. In Interspeech 2022, pp.\ 1026--1030, 2022. doi:10.21437/Interspeech.2022-10560

    work page doi:10.21437/interspeech.2022-10560 2022

  8. [8]

    Neural machine translation by jointly learning to align and translate

    Dzmitry Bahdanau, Kyung Hyun Cho, and Yoshua Bengio. Neural machine translation by jointly learning to align and translate. In 3rd International Conference on Learning Representations, ICLR 2015, 2015

    2015

  9. [9]

    Seamlessm4t: Massively multilingual & multimodal machine translation

    Lo \" c Barrault, Yu-An Chung, Mariano Cora Meglioli, David Dale, Ning Dong, Paul-Ambroise Duquenne, Hady Elsahar, Hongyu Gong, Kevin Heffernan, John Hoffman, et al. Seamlessm4t: Massively multilingual & multimodal machine translation. arXiv preprint arXiv:2308.11596, 2023

    Pith/arXiv arXiv 2023

  10. [10]

    Explainable artificial intelligence (xai): Concepts, taxonomies, opportunities and challenges toward responsible ai

    Alejandro Barredo Arrieta , Natalia Díaz-Rodríguez, Javier Del Ser , Adrien Bennetot, Siham Tabik, Alberto Barbado, Salvador Garcia, Sergio Gil-Lopez, Daniel Molina, Richard Benjamins, Raja Chatila, and Francisco Herrera. Explainable artificial intelligence (xai): Concepts, taxonomies, opportunities and challenges toward responsible ai. Information Fusion...

    doi:10.1016/j.inffus.2019.12.012 2020

  11. [11]

    Jasmijn Bastings and Katja Filippova. The elephant in the interpretability room: Why use attention as explanation when we have saliency methods? In Proceedings of the Third BlackboxNLP Workshop on Analyzing and Interpreting Neural Networks for NLP, pp.\ 149--155, Online, November 2020. doi:10.18653/v1/2020.blackboxnlp-1.14. URL https://aclanthology.org/20...

    doi:10.18653/v1/2020.blackboxnlp-1.14 2020

  12. [12]

    Audiomnist: Exploring explainable artificial intelligence for audio analysis on a simple benchmark

    Sören Becker, Johanna Vielhaben, Marcel Ackermann, Klaus-Robert Müller, Sebastian Lapuschkin, and Wojciech Samek. Audiomnist: Exploring explainable artificial intelligence for audio analysis on a simple benchmark. Journal of the Franklin Institute, 361 0 (1): 0 418--428, 2024. ISSN 0016-0032. doi:https://doi.org/10.1016/j.jfranklin.2023.11.038. URL https:...

    doi:10.1016/j.jfranklin.2023.11.038 2024

  13. [13]

    Generic attention-model explainability for interpreting bi-modal and encoder-decoder transformers

    Hila Chefer, Shir Gur, and Lior Wolf. Generic attention-model explainability for interpreting bi-modal and encoder-decoder transformers. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pp.\ 397--406, October 2021

    2021

  14. [14]

    Accurate word alignment induction from neural machine translation

    Yun Chen, Yang Liu, Guanhua Chen, Xin Jiang, and Qun Liu. Accurate word alignment induction from neural machine translation. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pp.\ 566--576, Online, November 2020. doi:10.18653/v1/2020.emnlp-main.42. URL https://aclanthology.org/2020.emnlp-main.42/

    work page doi:10.18653/v1/2020.emnlp-main.42 2020

  15. [15]

    Kevin Clark, Urvashi Khandelwal, Omer Levy, and Christopher D. Manning. What does BERT look at? an analysis of BERT `s attention. In Proceedings of the 2019 ACL Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP, pp.\ 276--286, Florence, Italy, August 2019. doi:10.18653/v1/W19-4828. URL https://aclanthology.org/W19-4828/

    doi:10.18653/v1/w19-4828 2019

  16. [17]

    Explaining by removing: A unified framework for model explanation

    Ian Covert, Scott Lundberg, and Su-In Lee. Explaining by removing: A unified framework for model explanation. Journal of Machine Learning Research, 22 0 (209): 0 1--90, 2021 a . URL http://jmlr.org/papers/v22/20-1316.html

    2021

  17. [18]

    Covert, Scott Lundberg, and Su-In Lee

    Ian C. Covert, Scott Lundberg, and Su-In Lee. Explaining by removing: a unified framework for model explanation . The Journal of Machine Learning Research, 22 0 (1): 0 9477--9566, jan 2021 b . ISSN 1532-4435

    2021

  18. [19]

    Di Gangi, Roldano Cattoni, Luisa Bentivogli, Matteo Negri, and Marco Turchi

    Mattia A. Di Gangi, Roldano Cattoni, Luisa Bentivogli, Matteo Negri, and Marco Turchi. M u ST - C : a M ultilingual S peech T ranslation C orpus. In Jill Burstein, Christy Doran, and Thamar Solorio (eds.), Proceedings of the 2019 Conference of the North A merican Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume...

    work page doi:10.18653/v1/n19-1202 2019

  19. [20]

    Rather a nurse than a physician - contrastive explanations under investigation

    Oliver Eberle, Ilias Chalkidis, Laura Cabello, and Stephanie Brandl. Rather a nurse than a physician - contrastive explanations under investigation. In Houda Bouamor, Juan Pino, and Kalika Bali (eds.), Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pp.\ 6907--6920, Singapore, December 2023. Association for Computat...

    doi:10.18653/v1/2023.emnlp-main.427 2023

  20. [21]

    G \'a llego, Belen Alastruey, Carlos Escolano, and Marta R

    Javier Ferrando, Gerard I. G \'a llego, Belen Alastruey, Carlos Escolano, and Marta R. Costa-juss \`a . Towards opening the black box of neural machine translation: Source and target interpretations of the transformer. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pp.\ 8756--8769, Abu Dhabi, United Arab Emirate...

    work page doi:10.18653/v1/2022.emnlp-main.599 2022

  21. [22]

    G \'a llego, and Marta R

    Javier Ferrando, Gerard I. G \'a llego, and Marta R. Costa-juss \`a . Measuring the mixing of contextual information in the transformer. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pp.\ 8698--8714, Abu Dhabi, United Arab Emirates, December 2022 b . doi:10.18653/v1/2022.emnlp-main.595. URL https://aclanthology...

    doi:10.18653/v1/2022.emnlp-main.595 2022

  22. [23]

    A primer on the inner workings of transformer-based language models

    Javier Ferrando, Gabriele Sarti, Arianna Bisazza, and Marta R Costa-juss \`a . A primer on the inner workings of transformer-based language models. arXiv preprint arXiv:2405.00208, 2024

    Pith/arXiv arXiv 2024

  23. [24]

    Spes: Spectrogram perturbation for explainable speech-to-text generation, 2025

    Dennis Fucci, Marco Gaido, Beatrice Savoldi, Matteo Negri, Mauro Cettolo, and Luisa Bentivogli. Spes: Spectrogram perturbation for explainable speech-to-text generation, 2025. URL https://arxiv.org/abs/2411.01710

    Pith/arXiv arXiv 2025

  24. [25]

    MOSEL : 950,000 hours of speech data for open-source speech foundation model training on EU languages

    Marco Gaido, Sara Papi, Luisa Bentivogli, Alessio Brutti, Mauro Cettolo, Roberto Gretter, Marco Matassoni, Mohamed Nabih, and Matteo Negri. MOSEL : 950,000 hours of speech data for open-source speech foundation model training on EU languages. In Yaser Al-Onaizan, Mohit Bansal, and Yun-Nung Chen (eds.), Proceedings of the 2024 Conference on Empirical Metho...

    work page doi:10.18653/v1/2024.emnlp-main.771 2024

  25. [26]

    Marco Gaido, Sara Papi, Matteo Negri, and Luisa Bentivogli. Speech translation with speech foundation models and large language models: What is there and what is missing? In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pp.\ 14760--14778, Bangkok, Thailand, August 2024 b . Association for ...

    doi:10.18653/v1/2024.acl-long.789 2024

  26. [27]

    Attention in natural language processing

    Andrea Galassi, Marco Lippi, and Paolo Torroni. Attention in natural language processing. IEEE Transactions on Neural Networks and Learning Systems, 32 0 (10): 0 4291--4308, 2021. doi:10.1109/TNNLS.2020.3019893

    arXiv 2021

  27. [28]

    Jointly learning to align and translate with transformer models

    Sarthak Garg, Stephan Peitz, Udhyakumar Nallasamy, and Matthias Paulik. Jointly learning to align and translate with transformer models. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pp.\ 4453--4462, Hong Kong, China, Novem...

    work page doi:10.18653/v1/d19-1453 2019

  28. [29]

    Connectionist temporal classification: labelling unsegmented sequence data with recurrent neural networks

    Alex Graves, Santiago Fern\' a ndez, Faustino Gomez, and J\" u rgen Schmidhuber. Connectionist temporal classification: labelling unsegmented sequence data with recurrent neural networks. In Proceedings of the 23rd International Conference on Machine Learning, ICML '06, pp.\ 369–376, New York, NY, USA, 2006. ISBN 1595933832. doi:10.1145/1143844.1143891. U...

    arXiv 2006

  29. [30]

    Neural turing machines

    Alex Graves, Greg Wayne, and Ivo Danihelka. Neural turing machines. arXiv preprint arXiv:1410.5401, 2014

    Pith/arXiv arXiv 2014

  30. [31]

    Conformer: Convolution-augmented Transformer for Speech Recognition

    Anmol Gulati, James Qin, Chung-Cheng Chiu, Niki Parmar, Yu Zhang, et al. Conformer: Convolution-augmented Transformer for Speech Recognition . In Proc. Interspeech, 2020. doi:10.21437/Interspeech.2020-3015

    doi:10.21437/interspeech.2020-3015 2020

  31. [32]

    Recent developments on espnet toolkit boosted by conformer

    Pengcheng Guo, Florian Boyer, Xuankai Chang, Tomoki Hayashi, Yosuke Higuchi, Hirofumi Inaguma, Naoyuki Kamo, Chenda Li, Daniel Garcia-Romero, Jiatong Shi, Jing Shi, Shinji Watanabe, Kun Wei, Wangyou Zhang, and Yuekai Zhang. Recent developments on espnet toolkit boosted by conformer. In ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech ...

    arXiv 2021

  32. [33]

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

    Fran c ois Hernandez, Vincent Nguyen, Sahar Ghannay, Natalia Tomashenko, and Yannick Est \`e ve. Ted-lium 3: Twice as much data and corpus repartition for experiments on speaker adaptation. In Alexey Karpov, Oliver Jokisch, and Rodmonga Potapova (eds.), Speech and Computer, pp.\ 198--208, Cham, 2018. ISBN 978-3-319-99579-3

    2018

  33. [34]

    An introductory survey on attention mechanisms in nlp problems

    Dichao Hu. An introductory survey on attention mechanisms in nlp problems. In Intelligent Systems and Applications: Proceedings of the 2019 Intelligent Systems Conference (IntelliSys) Volume 2, pp.\ 432--448. Springer, 2020

    2019

  34. [35]

    Europarl-st: A multilingual corpus for speech translation of parliamentary debates

    Javier Iranzo-Sánchez, Joan Albert Silvestre-Cerdà, Javier Jorge, Nahuel Roselló, Adrià Giménez, Albert Sanchis, Jorge Civera, and Alfons Juan. Europarl-st: A multilingual corpus for speech translation of parliamentary debates. In ICASSP 2020 - 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp.\ 8229--8233, 2020. d...

    arXiv 2020

  35. [36]

    Perturbation-based methods for explaining deep neural networks: A survey

    Maksims Ivanovs, Roberts Kadikis, and Kaspars Ozols. Perturbation-based methods for explaining deep neural networks: A survey. Pattern Recognition Letters, 150: 0 228--234, 2021. ISSN 0167-8655. doi:https://doi.org/10.1016/j.patrec.2021.06.030. URL https://www.sciencedirect.com/science/article/pii/S0167865521002440

    work page doi:10.1016/j.patrec.2021.06.030 2021

  36. [37]

    Sarthak Jain and Byron C. Wallace. A ttention is not E xplanation. In Proceedings of the 2019 Conference of the North A merican Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers) , pp.\ 3543--3556, Minneapolis, Minnesota, June 2019. doi:10.18653/v1/N19-1357. URL https://aclanthology.org/...

    doi:10.18653/v1/n19-1357 2019

  37. [38]

    The state and fate of linguistic diversity and inclusion in the NLP world

    Pratik Joshi, Sebastin Santy, Amar Budhiraja, Kalika Bali, and Monojit Choudhury. The state and fate of linguistic diversity and inclusion in the NLP world. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pp.\ 6282--6293, Online, July 2020. doi:10.18653/v1/2020.acl-main.560. URL https://aclanthology.org/2020.acl...

    doi:10.18653/v1/2020.acl-main.560 2020

  38. [39]

    Identifying important time-frequency locations in continuous speech utterances

    Hassan Salami Kavaki and Michael I Mandel. Identifying important time-frequency locations in continuous speech utterances. In Proceedings of Interspeech, 2020

    2020

  39. [40]

    Cross-attention end-to-end asr for two-party conversations

    Suyoun Kim, Siddharth Dalmia, and Florian Metze. Cross-attention end-to-end asr for two-party conversations. In Interspeech 2019, pp.\ 4380--4384, 2019. doi:10.21437/Interspeech.2019-3173

    work page doi:10.21437/interspeech.2019-3173 2019

  40. [41]

    Attention is not only a weight: Analyzing transformers with vector norms

    Goro Kobayashi, Tatsuki Kuribayashi, Sho Yokoi, and Kentaro Inui. Attention is not only a weight: Analyzing transformers with vector norms. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pp.\ 7057--7075, Online, November 2020. doi:10.18653/v1/2020.emnlp-main.574. URL https://aclanthology.org/2020.emnlp-main.574/

    doi:10.18653/v1/2020.emnlp-main.574 2020

  41. [42]

    I ncorporating R esidual and N ormalization L ayers into A nalysis of M asked L anguage M odels

    Goro Kobayashi, Tatsuki Kuribayashi, Sho Yokoi, and Kentaro Inui. I ncorporating R esidual and N ormalization L ayers into A nalysis of M asked L anguage M odels. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pp.\ 4547--4568, Online and Punta Cana, Dominican Republic, November 2021. doi:10.18653/v1/2021.emnlp-m...

    doi:10.18653/v1/2021.emnlp-main.373 2021

  42. [43]

    Analyzing feed-forward blocks in transformers through the lens of attention maps

    Goro Kobayashi, Tatsuki Kuribayashi, Sho Yokoi, and Kentaro Inui. Analyzing feed-forward blocks in transformers through the lens of attention maps. In The Twelfth International Conference on Learning Representations, 2024. URL https://openreview.net/forum?id=mYWsyTuiRp

    2024

  43. [44]

    S entence P iece: A simple and language independent subword tokenizer and detokenizer for neural text processing

    Taku Kudo and John Richardson. S entence P iece: A simple and language independent subword tokenizer and detokenizer for neural text processing. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing: System Demonstrations, pp.\ 66--71, Brussels, Belgium, November 2018. doi:10.18653/v1/D18-2012. URL https://aclanthology....

    doi:10.18653/v1/d18-2012 2018

  44. [45]

    MADLAD -400: A multilingual and document-level large audited dataset

    Sneha Kudugunta, Isaac Rayburn Caswell, Biao Zhang, Xavier Garcia, Derrick Xin, Aditya Kusupati, Romi Stella, Ankur Bapna, and Orhan Firat. MADLAD -400: A multilingual and document-level large audited dataset. In Thirty-seventh Conference on Neural Information Processing Systems Datasets and Benchmarks Track, 2023. URL https://openreview.net/forum?id=Y45ZCxslFx

    2023

  45. [46]

    Solomon Kullback and Richard A. Leibler. On information and sufficiency. The Annals of Mathematical Statistics, 22 0 (1): 0 79--86, 1951

    1951

  46. [47]

    CAST : Cross-attention in space and time for video action recognition

    Dongho Lee, Jongseo Lee, and Jinwoo Choi. CAST : Cross-attention in space and time for video action recognition. In Thirty-seventh Conference on Neural Information Processing Systems, 2023. URL https://openreview.net/forum?id=iATY9W5Xw7

    2023

  47. [48]

    Multimodal speech emotion recognition using cross attention with aligned audio and text

    Yoonhyung Lee, Seunghyun Yoon, and Kyomin Jung. Multimodal speech emotion recognition using cross attention with aligned audio and text. In Interspeech 2020, pp.\ 2717--2721, 2020. doi:10.21437/Interspeech.2020-2312

    work page doi:10.21437/interspeech.2020-2312 2020

  48. [50]

    Non-autoregressive end-to-end approaches for joint automatic speech recognition and spoken language understanding

    Mohan Li and Rama Doddipatla. Non-autoregressive end-to-end approaches for joint automatic speech recognition and spoken language understanding. In 2022 IEEE Spoken Language Technology Workshop (SLT), pp.\ 390--397, 2023. doi:10.1109/SLT54892.2023.10023042

    arXiv 2022

  49. [51]

    Cat: Cross attention in vision transformer

    Hezheng Lin, Xing Cheng, Xiangyu Wu, and Dong Shen. Cat: Cross attention in vision transformer. In 2022 IEEE International Conference on Multimedia and Expo (ICME), pp.\ 1--6, 2022. doi:10.1109/ICME52920.2022.9859720

    arXiv 2022

  50. [52]

    whisper-timestamped

    J \'e r \^o me Louradour. whisper-timestamped. https://github.com/linto-ai/whisper-timestamped, 2023

    2023

  51. [53]

    Cross attention is all you need: relational remote sensing change detection with transformer

    Kaixuan Lu, Xiao Huang, Ruiheng Xia, Pan Zhang, and Junping Shen and. Cross attention is all you need: relational remote sensing change detection with transformer. GIScience & Remote Sensing, 61 0 (1): 0 2380126, 2024. doi:10.1080/15481603.2024.2380126

    arXiv 2024

  52. [54]

    From understanding to utilization: A survey on explainability for large language models

    Haoyan Luo and Lucia Specia. From understanding to utilization: A survey on explainability for large language models. arXiv preprint arXiv:2401.12874, 2024

    Pith/arXiv arXiv 2024

  53. [55]

    Post-hoc Interpretability for Neural NLP: A Survey

    Andreas Madsen, Siva Reddy, and Sarath Chandar. Post-hoc Interpretability for Neural NLP: A Survey . ACM Computing Surveys, 55 0 (8), dec 2022. ISSN 0360-0300. doi:10.1145/3546577. URL https://doi.org/10.1145/3546577

    doi:10.1145/3546577 2022

  54. [56]

    Directly comparing the listening strategies of humans and machines

    Michael I Mandel. Directly comparing the listening strategies of humans and machines. In INTERSPEECH, pp.\ 660--664, 2016

    2016

  55. [57]

    Visualizing automatic speech recognition--means for a better understanding? ISCA Symposium on Security and Privacy in Speech Communication, 2021

    Karla Markert, Romain Parracone, Mykhailo Kulakov, Philip Sperl, Ching-Yu Kao, and Konstantin B \"o ttinger. Visualizing automatic speech recognition--means for a better understanding? ISCA Symposium on Security and Privacy in Speech Communication, 2021

    2021

  56. [58]

    G lob E nc: Quantifying global token attribution by incorporating the whole encoder layer in transformers

    Ali Modarressi, Mohsen Fayyaz, Yadollah Yaghoobzadeh, and Mohammad Taher Pilehvar. G lob E nc: Quantifying global token attribution by incorporating the whole encoder layer in transformers. In Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pp.\ 258--271, Seatt...

    doi:10.18653/v1/2022.naacl-main.19 2022

  57. [59]

    D ecomp X : Explaining transformers decisions by propagating token decomposition

    Ali Modarressi, Mohsen Fayyaz, Ehsan Aghazadeh, Yadollah Yaghoobzadeh, and Mohammad Taher Pilehvar. D ecomp X : Explaining transformers decisions by propagating token decomposition. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pp.\ 2649--2664, Toronto, Canada, July 2023. doi:10.18653/v...

    doi:10.18653/v1/2023.acl-long.149 2023

  58. [60]

    Exploring the role of BERT token representations to explain sentence probing results

    Hosein Mohebbi, Ali Modarressi, and Mohammad Taher Pilehvar. Exploring the role of BERT token representations to explain sentence probing results. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pp.\ 792--806, Online and Punta Cana, Dominican Republic, November 2021. doi:10.18653/v1/2021.emnlp-main.61. URL https:...

    doi:10.18653/v1/2021.emnlp-main.61 2021

  59. [61]

    Homophone disambiguation reveals patterns of context mixing in speech transformers

    Hosein Mohebbi, Grzegorz Chrupa a, Willem Zuidema, and Afra Alishahi. Homophone disambiguation reveals patterns of context mixing in speech transformers. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pp.\ 8249--8260, Singapore, December 2023 a . doi:10.18653/v1/2023.emnlp-main.513. URL https://aclanthology.org/...

    doi:10.18653/v1/2023.emnlp-main.513 2023

  60. [62]

    Quantifying context mixing in transformers

    Hosein Mohebbi, Willem Zuidema, Grzegorz Chrupa a, and Afra Alishahi. Quantifying context mixing in transformers. In Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics, pp.\ 3378--3400, Dubrovnik, Croatia, May 2023 b . doi:10.18653/v1/2023.eacl-main.245. URL https://aclanthology.org/2023.eacl-main.245/

    doi:10.18653/v1/2023.eacl-main.245 2023

  61. [63]

    o rg Schl\

    Meike Nauta, Jan Trienes, Shreyasi Pathak, Elisa Nguyen, Michelle Peters, Yasmin Schmitt, J\" o rg Schl\" o tterer, Maurice van Keulen, and Christin Seifert. From anecdotal evidence to quantitative evaluation methods: A systematic review on evaluating explainable ai. ACM Comput. Surv., 55 0 (13s), July 2023. ISSN 0360-0300. doi:10.1145/3583558. URL https:...

    doi:10.1145/3583558 2023

  62. [64]

    A study of the plausibility of attention between rnn encoders in natural language inference

    Duc Hau Nguyen, Guillaume Gravier, and Pascale S \'e billot. A study of the plausibility of attention between rnn encoders in natural language inference. In 2021 20th IEEE International Conference on Machine Learning and Applications (ICMLA), pp.\ 1623--1629. IEEE, 2021

    2021

  63. [65]

    Token-wise decomposition of autoregressive language model hidden states for analyzing model predictions

    Byung-Doh Oh and William Schuler. Token-wise decomposition of autoregressive language model hidden states for analyzing model predictions. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pp.\ 10105--10117, Toronto, Canada, July 2023. doi:10.18653/v1/2023.acl-long.562. URL https://aclantho...

    work page doi:10.18653/v1/2023.acl-long.562 2023

  64. [66]

    Librispeech: An asr corpus based on public domain audio books

    Vassil Panayotov, Guoguo Chen, Daniel Povey, and Sanjeev Khudanpur. Librispeech: An asr corpus based on public domain audio books. In 2015 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp.\ 5206--5210, 2015. doi:10.1109/ICASSP.2015.7178964

    arXiv 2015

  65. [67]

    Attention as a guide for simultaneous speech translation

    Sara Papi, Matteo Negri, and Marco Turchi. Attention as a guide for simultaneous speech translation. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pp.\ 13340--13356, Toronto, Canada, July 2023 a . Association for Computational Linguistics. doi:10.18653/v1/2023.acl-long.745. URL https://...

    doi:10.18653/v1/2023.acl-long.745 2023

  66. [68]

    Alignatt: Using attention-based audio-translation alignments as a guide for simultaneous speech translation

    Sara Papi, Marco Turchi, and Matteo Negri. Alignatt: Using attention-based audio-translation alignments as a guide for simultaneous speech translation. In Interspeech 2023, pp.\ 3974--3978, 2023 b . doi:10.21437/Interspeech.2023-170

    doi:10.21437/interspeech.2023-170 2023

  67. [69]

    Park, William Chan, Yu Zhang, Chung-Cheng Chiu, Barret Zoph, Ekin D

    Daniel S. Park, William Chan, Yu Zhang, Chung-Cheng Chiu, Barret Zoph, Ekin D. Cubuk, and Quoc V. Le. Specaugment: A simple data augmentation method for automatic speech recognition. In Interspeech 2019, pp.\ 2613--2617, 2019. doi:10.21437/Interspeech.2019-2680

    doi:10.21437/interspeech.2019-2680 2019

  68. [70]

    Explaining speech classification models via word-level audio segments and paralinguistic features

    Eliana Pastor, Alkis Koudounas, Giuseppe Attanasio, Dirk Hovy, and Elena Baralis. Explaining speech classification models via word-level audio segments and paralinguistic features. In Yvette Graham and Matthew Purver (eds.), Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers),...

    2024

  69. [71]

    Reproducing whisper-style training using an open-source toolkit and publicly available data

    Yifan Peng, Jinchuan Tian, Brian Yan, Dan Berrebbi, Xuankai Chang, Xinjian Li, Jiatong Shi, Siddhant Arora, William Chen, Roshan Sharma, Wangyou Zhang, Yui Sudo, Muhammad Shakeel, Jee-Weon Jung, Soumi Maiti, and Shinji Watanabe. Reproducing whisper-style training using an open-source toolkit and publicly available data. In 2023 IEEE Automatic Speech Recog...

    arXiv 2023

  70. [72]

    Owsm v3.1: Better and faster open whisper-style speech models based on e-branchformer

    Yifan Peng, Jinchuan Tian, William Chen, Siddhant Arora, Brian Yan, Yui Sudo, Muhammad Shakeel, Kwanghee Choi, Jiatong Shi, Xuankai Chang, Jee weon Jung, and Shinji Watanabe. Owsm v3.1: Better and faster open whisper-style speech models based on e-branchformer. In Interspeech 2024, pp.\ 352--356, 2024. doi:10.21437/Interspeech.2024-1194

    work page doi:10.21437/interspeech.2024-1194 2024

  71. [73]

    MLS: A Large-Scale Multilingual Dataset for Speech Research

    Vineel Pratap, Qiantong Xu, Anuroop Sriram, Gabriel Synnaeve, and Ronan Collobert. MLS: A Large-Scale Multilingual Dataset for Speech Research . In Proc. Interspeech 2020, pp.\ 2757--2761, 2020. doi:10.21437/Interspeech.2020-2826

    doi:10.21437/interspeech.2020-2826 2020

  72. [74]

    Krishna C. Puvvada, Piotr Żelasko, He Huang, Oleksii Hrinchuk, Nithin Rao Koluguri, Kunal Dhawan, Somshubra Majumdar, Elena Rastorgueva, Zhehuai Chen, Vitaly Lavrukhin, Jagadeesh Balam, and Boris Ginsburg. Less is more: Accurate speech recognition & translation without web-scale data. In Interspeech 2024, pp.\ 3964--3968, 2024. doi:10.21437/Interspeech.2024-2294

    doi:10.21437/interspeech.2024-2294 2024

  73. [75]

    Robust speech recognition via large-scale weak supervision

    Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine Mcleavey, and Ilya Sutskever. Robust speech recognition via large-scale weak supervision. In Proceedings of the 40th International Conference on Machine Learning, volume 202 of Proceedings of Machine Learning Research, pp.\ 28492--28518, 23--29 Jul 2023. URL https://proceedings.mlr.press/v202/r...

    2023

  74. [76]

    COMET : A neural framework for MT evaluation

    Ricardo Rei, Craig Stewart, Ana C Farinha, and Alon Lavie. COMET : A neural framework for MT evaluation. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pp.\ 2685--2702, Online, November 2020. Association for Computational Linguistics. doi:10.18653/v1/2020.emnlp-main.213. URL https://aclanthology.org/2020...

    doi:10.18653/v1/2020.emnlp-main.213 2020

  75. [77]

    Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead

    Cynthia Rudin. Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nature machine intelligence, 1 0 (5): 0 206--215, 2019

    2019

  76. [78]

    NLP evaluation in trouble: On the need to measure LLM data contamination for each benchmark

    Oscar Sainz, Jon Campos, Iker Garc \'i a-Ferrero, Julen Etxaniz, Oier Lopez de Lacalle, and Eneko Agirre. NLP evaluation in trouble: On the need to measure LLM data contamination for each benchmark. In Findings of the Association for Computational Linguistics: EMNLP 2023, pp.\ 10776--10787, Singapore, December 2023. Association for Computational Linguisti...

    doi:10.18653/v1/2023.findings-emnlp.722 2023

  77. [79]

    Anders, and Klaus-Robert Müller

    Wojciech Samek, Grégoire Montavon, Sebastian Lapuschkin, Christopher J. Anders, and Klaus-Robert Müller. Explaining Deep Neural Networks and Beyond: A Review of Methods and Applications . Proceedings of the IEEE, 109 0 (3): 0 247--278, 2021. doi:10.1109/JPROC.2021.3060483

    arXiv 2021

  78. [80]

    Sofia Serrano and Noah A. Smith. Is attention interpretable? In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pp.\ 2931--2951, Florence, Italy, July 2019. doi:10.18653/v1/P19-1282. URL https://aclanthology.org/P19-1282/

    doi:10.18653/v1/p19-1282 2019

  79. [81]

    Exploring explainable ai: a bibliometric analysis

    Chhavi Sharma, Swati Sharma, Kavita Sharma, et al. Exploring explainable ai: a bibliometric analysis. Discover Applied Sciences, 6 0 (1): 0 615, 2024. doi:10.1007/s42452-024-06324-z. URL https://doi.org/10.1007/s42452-024-06324-z

    work page doi:10.1007/s42452-024-06324-z 2024

  80. [82]

    Understanding the role of self attention for efficient speech recognition

    Kyuhong Shim, Jungwook Choi, and Wonyong Sung. Understanding the role of self attention for efficient speech recognition. In International Conference on Learning Representations, 2022. URL https://openreview.net/forum?id=AvcfxqRy4Y

    2022

Showing first 80 references.