Pith. sign in

REVIEW 3 major objections 5 minor 40 references

Alignment Drift in Single-Model Speculative Decoding for ASR: Mechanism, Correction, and Cost

T0 review · 3 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read In single-model speculative ASR, the draft's audio attention drifts between verifications; recentering it on the audio anchor recovers continuation acceptance.

desk verdict A genuinely useful empirical paper: the restart/continuation split and matched window intervention make a real case for alignment drift, but the anchor rests on MMS-FA and the artifact is not yet checkable. read the letter →

arxiv 2608.12703 v1 pith:QRGKLWZH submitted 2026-08-13 cs.SD cs.MM

Xinyu Wang , Huapeng Zhou , Ziyu Zhao , Silin Meng , Ke Bai , Dongming Shen , Xiao-Wen Chang , Alex Smola This is my paper · ORCID
classification cs.SDcs.MM
keywords single-modelspeculativedecodingautomaticspeechrecognitionalignmentdriftaudioanchorcontinuationacceptanceverificationattentionDraftdraft-verifyloop
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to establish that in single-model speculative decoding for speech recognition, the draft's proposals degrade between target verifications because its audio attention drifts away from the audio position of the next token; the authors call this alignment drift. They claim this drift is a recoverable cause of lost continuation acceptance, distinguishable from weak next-token prediction or limited draft capacity. The key evidence is a window intervention that keeps the visible audio width fixed and moves only the center: a window centered on the forced-alignment frame recovers continuation acceptance, while an equally narrow window at a wrong position reduces it. The paper also shows that the position can be reacquired cheaply at runtime from verification attention, or taught during training with a guided-attention loss called AnchorDraft, and that either correction speeds up end-to-end decoding only when the recovered accepted tokens cover its cost. If true, the result makes audio-position tracking a first-class requirement for self-speculative ASR, alongside token prediction and draft cost.

What carries the argument

The central object is the audio anchor: the encoded-audio frame that should feed the next draft token, operationally measured as the peak of the draft's audio cross-attention and validated against MMS-FA forced alignment. The argument's load-bearing tool is the window intervention, which fixes the attention mask width (half-width ±400 ms) and changes only its center, isolating audio position from token information and width effects. Two corrections attach to this machinery: a runtime correction that reads the target's verification attention (layer 21 selected on development data) as a fresh anchor, and AnchorDraft, a training-only loss that matches the draft's first-layer attention to a Gaussian centered on the forced-alignment frame for the next token while leaving the inference graph unchanged. A cost-screening inequality, $\frac{\Delta L}{1+L} \gtrsim \frac{\Delta C_{\mathrm{corr}}}{K\rho+1}$, with $\rho$ the draft-to-verification cost ratio, predicts which corrections will improve latency.

What would settle it

Run the fixed-width window intervention on utterances whose token-to-audio positions come from a gold-standard source (e.g., manually transcribed time boundaries or a forced aligner with independently verified accuracy on the target models). If windows centered on those frames no longer recover continuation acceptance relative to equally narrow wrong-position windows, or if the correct-versus-wrong contrast reverses, the alignment-drift explanation is refuted. A complementary test: use a synthetic audio condition where token durations are constant, so no drift should accumulate, and show that continuation acceptance under draft-only steps does not decay.

Watch

Extended reading notes

Core claim

In single-model speculative decoding for ASR, per-step audio access changes the first proposal after verification only modestly, yet it roughly doubles the acceptance of later draft-only proposals. The paper's central claim is that this continuation gap is caused by alignment drift: the draft's audio cross-attention peak moves away from the audio anchor by a growing number of frames as the draft runs without the target, with late-draft median error reaching 21 frames in the hardest condition, while target attention during verification stays within a 2-frame median. Because the accepted transcript records only token position, not audio position, the draft loses track of where the next token should be read from. The causal test is a fixed-width attention window whose center is the only thing that changes: a correct-position window recovers continuation (depth-two conditional acceptance rises by +0.144 over unrestricted attention), a wrong-position window lowers it (−0.111), and the matched correct-versus-wrong contrast is +0.254 with 95% CI [+0.241,+0.268] through depth three. The paper further argues that this position is recoverable: verification attention reduces late-draft error to a 2-frame median and gives a mean +3.4% speed gain at 1.7B after readout cost, and the AnchorDraft training objective (a Gaussian guided-attention loss on the first draft layer) improves five-set mean speed by +3.9% to +5.0% at 1.7B and +6.4% to +7.1% at 0.6B, with WER changes at most 0.01 percentage points. The paper concludes that ASR self-speculation depends on token prediction, audio-position tracking, and draft cost, with the cost condition $\frac{\Delta L}{1+L} \gtrsim \frac{\Delta C_{\mathrm{corr}}}{K\rho+1}$ determining when a correction helps.

Load-bearing premise

The causal claim that centering the window on the audio position recovers acceptance assumes the MMS-FA forced-alignment frame is the true per-token audio position; if MMS-FA is systematically biased for the target model, both the 'correct' and 'wrong' centers shift together and the measured correct-versus-wrong contrast could be distorted.

Editorial extensions

If this is right

  • Audio access without audio localization is insufficient: the first proposal after verification barely changes, so any single-model ASR draft must maintain position across draft-only steps.
  • A correct-centered window improves cumulative survival at depth two by +0.105 and accepted length by +0.262, defining a quantitative recovery target for any position-correcting proposal.
  • Verification attention is a cost-effective runtime position source on the tested Qwen checkpoints, reducing late-draft median error from up to 21 frames to 2 frames, with a five-set mean speed gain of +3.4% at 1.7B.
  • AnchorDraft training delivers speed gains at both tested scales without altering the decoding graph or degrading WER (max absolute change 0.01 percentage points).
  • The two corrections are complementary: after AnchorDraft training, adding the runtime readout gives a further +1.8% (95% CI [+0.9%,+2.7%]) speed gain at 1.7B.

Reading between the lines

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

  • Beyond the paper's tested ASR models, the same restart-continuation asymmetry should appear in any single-model speculative decoder that consumes a continuous input (video frames, raw audio for music, etc.) because the draft must track a moving anchor between verifications; measuring that anchor error would transfer the result directly.
  • The cost condition implies specific deployment thresholds: when batch size grows (the draft-to-verification cost ratio $\rho$ rises from 0.0275 at batch one to 0.1274 at batch 64), the runtime correction should become less beneficial; this is a quantitative prediction that the paper's own measurement setup could test.
  • Since correct positioning recovers only part of the continuation loss at depth four, a residual source of failure remains; combining anchor correction with a draft that exposes next-token features earlier (the token probe shows they peak late in the decoder) is a plausible next step to close the remaining gap.
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

3 major / 5 minor

Summary. The paper studies single-model speculative decoding for ASR, where a lightweight draft attached to the target model proposes tokens between target verifications. It identifies a phenomenon it calls 'alignment drift': the draft's audio cross-attention moves away from a reference audio frame (the 'anchor') during continuation, and this drift is associated with lower acceptance of later draft proposals. The authors support this with a matched intervention that fixes window width and changes only the window center: a window centered on forced-alignment frames recovers continuation acceptance, while an equally narrow wrong-position window reduces it. They propose two corrections: a runtime correction that reads audio position from verification attention, and a training correction (AnchorDraft) that adds a guided-attention loss using forced-alignment supervision. Both improve end-to-end speed when the recovered accepted tokens cover the correction cost, with the cost condition expressed by a screening rule. The paper includes extensive supplementary material with official-split replications, paired bootstrap intervals, cross-architecture comparisons, and deployment-scope measurements.

Significance. If the central claim holds, the paper makes a valuable contribution by showing that in single-model speculative ASR, continuation quality depends not only on next-token prediction but also on audio-position tracking and on the cost of the draft. The study is unusually careful in several respects: the main causal estimate uses matched official-split arms with a fixed window width and paired bootstrap confidence intervals; the draft attention peak is validated against an external forced aligner (MMS-FA); the cost screening rule is tested across multiple settings; and the authors explicitly separate association from causation, including diagnostic boundaries and a limitations section. The artifact includes bootstrap code, run metadata, and mapping rules, which supports reproducibility. The main weakness is that MMS-FA, the sole external reference used to define 'correct' audio position, is itself a model-based aligner and is not independently validated for the Qwen checkpoints in this paper; this gap affects the interpretation of the causal contrast and should be addressed before publication.

major comments (3)
  1. [Section 4, Supp. A.2, Supp. B.2, Supp. table 7] The manuscript defines the 'correct' audio position using MMS-FA forced-alignment frames and uses that reference both to validate the draft attention peak and to center the window intervention. MMS-FA is itself a model-based aligner with no independent ground-truth check for Qwen in this paper. Supp. B.2 reports that Voxtral attention is systematically +205 ms late relative to MMS-FA, showing that attention-based and MMS-FA positions can differ by a large constant; for Qwen, Supp. table 7 reports only mean absolute errors (136–155 ms) and no signed offset. If Qwen has a similar systematic bias, both the 'correct' and 'shifted' window centers in the matched intervention are displaced from the actual acoustic anchor, and the depth-2 correct-versus-shifted contrast of +0.254 (fig. 2a, Supp. table 12) would demonstrate sensitivity to some reference position but would not specifically establish the mechanism of losing the true audio anchor. The 2-frame verification-attention error is also measured against MMS-FA, so it does not break the circularity. I recommend reporting the signed median offset for Qwen, performing a sensitivity analysis that shifts all window centers by that constant, and/or validating the anchor on a small set with an independent alignment source (for example, manually checked phoneme boundaries or a second aligner). This is a validation-gap concern rather than an observed contradiction, but it is load-bearing because the labels 'alignment drift' and 'the draft loses the true audio anchor' depend on the reference being accurate.
  2. [Section 3, Table 1, Supp. table 6] The matched 'with audio' vs. 'no audio' drafts differ not only in whether per-step audio access exists but also in architecture: the audio-conditioned draft adds cross-attention parameters, so the roughly doubled continuation acceptance could partly reflect additional capacity rather than audio access per se. The paper states that both drafts share the same layer budget, but it does not control for parameter count. This does not invalidate the window-intervention causal estimate, but it weakens the motivating claim that 'per-step audio access changes the first proposal modestly but roughly doubles later-proposal acceptance.' I suggest either adding a control with a non-audio cross-attention module of comparable parameter count or explicitly stating in the main text that this comparison confounds access with added capacity.
  3. [Section 5, Eq. (1), fig. 3c, Supp. D.1] The cost screening rule in Eq. (1) is used to claim that the measured result becomes positive only after the recovered accepted length clears the cost, with a reported 10/10 direction match across deployed evaluation-set point estimates. However, the rule's components—the draft-step cost ratio ρ, the correction cost ΔC_corr, and the accepted-length gain ΔL—are all measured on the same implementation and possibly on the same settings used to form the rule. The paper does not state whether the 19 settings in fig. 3c were used to choose the rule or whether the rule was specified before inspecting the timing results. As written, the sign-matching test has limited confirmatory value. I recommend clarifying the timeline and, if possible, reporting a leave-one-setting-out check on the direction match.
minor comments (5)
  1. [Abstract, Section 4] The abstract and main text report '21 frames' and '2-frame median' without stating in the same sentence that these are distances to MMS-FA reference frames; please state the reference explicitly to avoid implying a ground-truth acoustic position.
  2. [Supp. table 7] Please add the signed median offset for Qwen, alongside the Voxtral +205 ms offset, so readers can assess whether a constant-bias correction would change the window-intervention conclusions.
  3. [Figure 2c] The legend in the figure uses 'wrong' while other parts of the paper use 'shifted'; please standardize the terminology across the figure and text.
  4. [Section 4, 'Position Interventions'] The window-width robustness result in Supp. Section A is mentioned in the main text, but the actual tested half-width values are not given; please add the range (e.g., ±200 ms to ±800 ms) to the main text for completeness.
  5. [Section 4, 'Probe protocol'] The main text says the probe layer and ridge strength are fixed on development data but does not give the selected layer; Supp. table 9 reports layer 20, so please include this value in the main text.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the position intervention, cost rule, and AnchorDraft speed gains are measured independently of the quantities they are claimed to explain.

full rationale

The paper's central causal claim is tested by intervening on the center of a fixed-width attention window, where the 'correct' center is supplied by an external forced aligner (MMS-FA) rather than by the draft's own output. This intervention does not fit any parameter to the acceptance outcome, so the correct-versus-wrong contrast in Section 4 and Supp. table 12 is not a renamed prediction or a fitted input. The attention-peak anchor is validated by correlation with MMS-FA and by shuffled controls (Supp. table 7), and the paper explicitly documents the Voxtral +205 ms offset (Supp. B.2), showing that the aligner-dependence is a measurement-validity limitation rather than a hidden assumption that forces the result. AnchorDraft does train the draft's attention to match an MMS-FA-derived Gaussian target by construction, but the paper's claimed benefits for AnchorDraft are continuation acceptance and end-to-end speed measured on the real cached loop; those are not implied by the alignment loss, and the shared-mechanism and stacked-correction experiments further separate position maintenance from general draft confidence. Hyperparameters such as lambda, sigma, layer 21, and window width are fixed on holdout or development data before test evaluation, so they do not encode the target speed gains. The cost rule in Eq. (1) is a derived algebraic screening condition, not a fit to the timing results. The only self-citations are to the Qwen3-ASR technical report for model provenance and to MARS as an alternative verification scheme; neither is load-bearing for the alignment-drift mechanism. Therefore no step in the derivation reduces to its own input by construction, and the circularity score is 0.

Assumptions & free parameters 5 free parameters · 4 assumptions · 2 invented entities

The central claims rest on measured quantities (anchor error, acceptance, latency) and on a small number of design choices: an external forced aligner as ground truth, an operational definition of audio position as an attention peak, a fixed window geometry, and two holdout-selected training hyperparameters. No new physical entity is postulated; 'audio anchor' and 'alignment drift' are internal constructs with operational handles. The cost screening rule uses one measured draft-step cost ratio, not a fit to the speed results.

free parameters (5)
  • AnchorDraft alignment-loss weight lambda = 0.1
    Selected on a fixed holdout from sensitivity options; the post-hoc sweep shows speed is positive nearby but peaks at 0.2.
  • Gaussian anchor width sigma = 5 frames
    Selected on a fixed holdout together with lambda in the AnchorDraft objective (Eq. 6).
  • Attention window half-width = 400 ms
    Fixed before test evaluation; the shifted control is 500 ms. The correct-position gain is stable over an intermediate width range, but the chosen width is a hand-set constant.
  • Verification-attention layer = 21
    Selected on development data from {7, 14, 18, 21, 24, 27}; the same layer is chosen for four Qwen heads.
  • Draft-step cost ratio rho = 0.0275
    Measured on Qwen-1.7B batch-one and used in Eq. (1) to predict the sign of speed gains; it is an empirical constant, not fitted to the acceptance outcomes.
assumptions (4)
  • domain assumption MMS-FA forced alignment provides the reference audio position for tokens.
    Used as the correct window center and as the AnchorDraft supervision target (Supp. A.3, Eq. 6); no independent ground-truth check beyond MMS-FA is provided.
  • domain assumption The peak of draft audio cross-attention is an operational measure of audio position.
    Validated by correlation with MMS-FA (table 7), but correlation is not causal; the window intervention is the causal check (Section 4).
  • ad hoc to paper Masking draft attention to a +-400 ms window changes only the position information available to the draft.
    The intervention holds width fixed and moves only the center, but a hard mask can also remove acoustic context; the width sweep and equally narrow wrong-center controls mitigate but do not eliminate this.
  • standard math Verification under exact arithmetic returns the target greedy sequence.
    Speculative verification is lossless in exact arithmetic; the paper measures low-precision deviations and reports small sign-mixed WER changes (Supp. A.4).
invented entities (2)
  • Audio anchor independent evidence
    purpose: The audio frame the draft should attend to when emitting its next token; alignment drift is defined as the growing distance between draft attention and this anchor.
    Operationalized via attention peaks correlated with MMS-FA and via the correct-position window intervention; it is not directly observed.
  • Alignment drift independent evidence
    purpose: Proposed mechanism for continuation acceptance loss in single-model speculative ASR.
    Anchor error grows with draft depth and the correct-versus-wrong window contrast changes acceptance in the predicted direction; it remains an internal-state construct rather than a directly measured quantity.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Alignment Drift in Single-Model Speculative Decoding for ASR: Mechanism, Correction, and Cost." pith.science (2026). https://pith.science/paper/QRGKLWZH

@misc{pith2026260812703,
  author       = {Pith},
  title        = {Pith review of: Alignment Drift in Single-Model Speculative Decoding for ASR: Mechanism, Correction, and Cost},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QRGKLWZH}},
  note         = {Machine review of arXiv:2608.12703}
}
read the original abstract

Speculative decoding speeds up generation by letting a cheap draft propose several tokens that a target model checks in one pass. In the single-model form, the draft is a lightweight module attached to the target rather than a separate model. Applying this design to Automatic Speech Recognition (ASR) introduces an extra problem. The draft can read the whole audio at every step, yet its proposals get worse as it runs on its own. Access is not localization. The accepted text keeps the transcript position explicit, but the draft must also track the changing audio position. In the primary matched comparison, per-step audio access changes the first proposal modestly but roughly doubles later-proposal acceptance. Fixed-width windows show that the audio position explains part of this gap. A correctly placed window recovers continuation, while an equally narrow window at the wrong position reduces it. Late-draft median error reaches 21 frames in the hardest reported condition, while target attention during verification stays within a 2-frame median. We test two ways to reduce this drift. The first reads the audio position from verification attention and uses it to guide the next draft round. It saves time only when the extra accepted tokens offset the readout cost. The second is AnchorDraft, which teaches the draft to track the audio position during training without changing the inference graph. The trained draft improves end-to-end speed at both tested target scales. These results show that ASR self-speculation depends on token prediction, audio-position tracking, and draft cost.

Figures

Figures reproduced from arXiv: 2608.12703 by the authors.

Figure 1
Figure 1. Overview of alignment drift. (a) Per-step audio access changes restart modestly and separates continuation. (b) Draft position drifts between target verifications and resets after verification. (c) Late-draft and verification anchor errors. used for the main timing results run on one NVIDIA A100- 80GB GPU at batch size one in bfloat16. Autoregressive and speculative timing use the same utterances and start from the … view at source ↗
Figure 2
Figure 2. Three measurements of audio position. (a) Depth-two conditional-acceptance contrasts on matched official-split [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Runtime and training-correction measurements. (a) Runtime speed change for the [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Held-out next-token probes. (a) Center-of-mass decoder depth for low- and high-NLL tokens. (b) Next-token predictability [PITH_FULL_IMAGE:figures/full_fig_p013_4.png]
Figure 5
Figure 5. Figure 5: Held-out audio-anchor probe R2 by decoder layer after removing the token-rank trend. same rounds. We follow them through four draft depths and compute anchor error, cumulative survival, conditional accep￾tance, and rejection odds over two seeds and five evaluation sets…
Figure 6
Figure 6. Figure 6: Depth-resolved checkpoint comparisons. (a) Anchor error for unrestricted attention and the pointer range over five [PITH_FULL_IMAGE:figures/full_fig_p017_6.png]
Figure 7
Figure 7. Figure 7: Accepted length and per-round draft cost by maxi [PITH_FULL_IMAGE:figures/full_fig_p019_7.png]
Figure 8
Figure 8. Figure 8: Deployment measurements. (a) Wall-clock composition. (b) Draft-to-verification cost ratio and projected speed by [PITH_FULL_IMAGE:figures/full_fig_p020_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

40 extracted references · 23 canonical work pages

  1. [1]

    Proceedings of the 40th International Conference on Machine Learning , articleno =

    Leviathan, Yaniv and Kalman, Matan and Matias, Yossi , title =. Proceedings of the 40th International Conference on Machine Learning , articleno =. 2023 , publisher =

    2023

  2. [2]

    2023 , eprint=

    Accelerating Large Language Model Decoding with Speculative Sampling , author=. 2023 , eprint=

    2023

  3. [3]

    Speculative Decoding: Exploiting Speculative Execution for Accelerating Seq2seq Generation

    Xia, Heming and Ge, Tao and Wang, Peiyi and Chen, Si-Qing and Wei, Furu and Sui, Zhifang. Speculative Decoding: Exploiting Speculative Execution for Accelerating Seq2seq Generation. Findings of the Association for Computational Linguistics: EMNLP 2023. 2023. doi:10.18653/v1/2023.findings-emnlp.257

    doi:10.18653/v1/2023.findings-emnlp.257 2023

  4. [4]

    Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies , pages=

    Decoding Speculative Decoding , author=. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies , pages=. 2025 , publisher=. doi:10.18653/v1/2025.naacl-long.328 , url=

    doi:10.18653/v1/2025.naacl-long.328 2025

  5. [5]

    2026 , eprint=

    MARS: Unleashing the Power of Speculative Decoding via Margin-Aware Verification , author=. 2026 , eprint=

    work page 2026

  6. [6]

    SpecASR: Accelerating LLM-based Automatic Speech Recognition via Speculative Decoding , year=

    Wei, Linye and Zhong, Shuzhang and Xu, Songqiang and Wang, Runsheng and Huang, Ru and Li, Meng , booktitle=. SpecASR: Accelerating LLM-based Automatic Speech Recognition via Speculative Decoding , year=

    work page

  7. [7]

    Model-Free Speculative Decoding for Transformer-Based ASR with Token Map Drafting , year=

    Ho, Tuan Vu and Kokubo, Hiroaki and Yamamoto, Masaaki and Kawaguchi, Yohei , booktitle=. Model-Free Speculative Decoding for Transformer-Based ASR with Token Map Drafting , year=

    work page

  8. [8]

    2023 , eprint=

    Distil-Whisper: Robust Knowledge Distillation via Large-Scale Pseudo Labelling , author=. 2023 , eprint=

    work page 2023

Show all 40 references
  1. [9]

    L ayer S kip: Enabling Early Exit Inference and Self-Speculative Decoding

    Elhoushi, Mostafa and Shrivastava, Akshat and Liskovich, Diana and Hosmer, Basil and Wasti, Bram and Lai, Liangzhen and Mahmoud, Anas and Acun, Bilge and Agarwal, Saurabh and Roman, Ahmed and Aly, Ahmed and Chen, Beidi and Wu, Carole-Jean. L ayer S kip: Enabling Early Exit Inf...

    2024 doi

  2. [10]

    2025 , eprint=

    EAGLE-3: Scaling up Inference Acceleration of Large Language Models via Training-Time Test , author=. 2025 , eprint=

    2025

  3. [11]

    2025 , eprint=

    EAGLE: Speculative Sampling Requires Rethinking Feature Uncertainty , author=. 2025 , eprint=

    2025

  4. [12]

    and Chen, Deming and Dao, Tri , title =

    Cai, Tianle and Li, Yuhong and Geng, Zhengyang and Peng, Hongwu and Lee, Jason D. and Chen, Deming and Dao, Tri , title =. Proceedings of the 41st International Conference on Machine Learning , articleno =. 2024 , publisher =

    2024

  5. [13]

    Draft& Verify: Lossless Large Language Model Acceleration via Self-Speculative Decoding , url=

    Zhang, Jun and Wang, Jue and Li, Huan and Shou, Lidan and Chen, Ke and Chen, Gang and Mehrotra, Sharad , year=. Draft& Verify: Lossless Large Language Model Acceleration via Self-Speculative Decoding , url=. doi:10.18653/v1/2024.acl-long.607 , booktitle=

    doi

  6. [14]

    BERT Rediscovers the Classical NLP Pipeline

    Tenney, Ian and Das, Dipanjan and Pavlick, Ellie. BERT Rediscovers the Classical NLP Pipeline. Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics. 2019. doi:10.18653/v1/P19-1452

    2019 doi

  7. [15]

    Designing and Interpreting Probes with Control Tasks

    Hewitt, John and Liang, Percy. Designing and Interpreting Probes with Control Tasks. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019. doi:10.18...

    2019 doi

  8. [16]

    2020 , howpublished=

    nostalgebraist , title=. 2020 , howpublished=

    2020

  9. [17]

    2026 , eprint=

    Self-Speculative Decoding for LLM-based ASR with CTC Encoder Drafts , author=. 2026 , eprint=

    2026

  10. [18]

    Hybrid Decoding: Rapid Pass and Selective Detailed Correction for Sequence Models , url=

    Lim, Yunkyu and Park, Jihwan and Kim, Hyung Yong and Lee, Hanbin and Kim, Byeong-Yeol , year=. Hybrid Decoding: Rapid Pass and Selective Detailed Correction for Sequence Models , url=. doi:10.1109/asru65441.2025.11434678 , booktitle=

    2025

  11. [19]

    Simultaneous Masked and Unmasked Decoding with Speculative Decoding Masking for Fast

    Okabe, Koji and Yamamoto, Hitoshi , booktitle=. Simultaneous Masked and Unmasked Decoding with Speculative Decoding Masking for Fast. 2025 , doi=

    2025

  12. [20]

    2024 , eprint=

    Speculative Speech Recognition by Audio-Prefixed Low-Rank Adaptation of Language Models , author=. 2024 , eprint=

    2024

  13. [21]

    2025 , eprint=

    WhisperKit: On-device Real-time ASR with Billion-Scale Transformers , author=. 2025 , eprint=

    2025

  14. [22]

    2024 , eprint=

    Recurrent Drafter for Fast Speculative Decoding in Large Language Models , author=. 2024 , eprint=

    2024

  15. [23]

    Listen, attend and spell: A neural network for large vocabulary conversational speech recognition , year=

    Chan, William and Jaitly, Navdeep and Le, Quoc and Vinyals, Oriol , booktitle=. Listen, attend and spell: A neural network for large vocabulary conversational speech recognition , year=

  16. [24]

    Proceedings of the 29th International Conference on Neural Information Processing Systems - Volume 1 , pages =

    Chorowski, Jan and Bahdanau, Dzmitry and Serdyuk, Dmitriy and Cho, Kyunghyun and Bengio, Yoshua , title =. Proceedings of the 29th International Conference on Neural Information Processing Systems - Volume 1 , pages =. 2015 , publisher =

    2015

  17. [25]

    International Conference on Learning Representations , year=

    Monotonic Chunkwise Attention , author=. International Conference on Learning Representations , year=

  18. [26]

    Efficiently Trainable Text-to-Speech System Based on Deep Convolutional Networks with Guided Attention , year=

    Tachibana, Hideyuki and Uenoyama, Katsuya and Aihara, Shunsuke , booktitle=. Efficiently Trainable Text-to-Speech System Based on Deep Convolutional Networks with Guided Attention , year=

  19. [27]

    and Weiss, Ron J

    Raffel, Colin and Luong, Minh-Thang and Liu, Peter J. and Weiss, Ron J. and Eck, Douglas , title =. Proceedings of the 34th International Conference on Machine Learning - Volume 70 , pages =. 2017 , publisher =

    2017

  20. [28]

    2017 , eprint=

    Get To The Point: Summarization with Pointer-Generator Networks , author=. 2017 , eprint=

    2017

  21. [29]

    2026 , eprint=

    Qwen3-ASR Technical Report , author=. 2026 , eprint=

    2026

  22. [30]

    and Ehrenberg, Andy and Lo, Andy and Denoix, Cl

    Liu, Alexander H. and Ehrenberg, Andy and Lo, Andy and Denoix, Cl. arXiv preprint arXiv:2507.13264 , year=

    arXiv

  23. [31]

    Proceedings of the 40th International Conference on Machine Learning , articleno =

    Radford, Alec and Kim, Jong Wook and Xu, Tao and Brockman, Greg and McLeavey, Christine and Sutskever, Ilya , title =. Proceedings of the 40th International Conference on Machine Learning , articleno =. 2023 , publisher =

    2023

  24. [32]

    Pratap, Vineel and Tjandra, Andros and Shi, Bowen and Tomasello, Paden and Babu, Arun and Kundu, Sayani and Elkahky, Ali and Ni, Zhaoheng and Vyas, Apoorv and Fazel-Zarandi, Maryam and Baevski, Alexei and Adi, Yossi and Zhang, Xiaohui and Hsu, Wei-Ning and Conneau, Alexis and ...

    2024

  25. [33]

    2019 , url=

    Language Models are Unsupervised Multitask Learners , author=. 2019 , url=

    2019

  26. [34]

    2024 , eprint=

    Whisper in Medusa's Ear: Multi-head Efficient Decoding for Transformer-based ASR , author=. 2024 , eprint=

    2024

  27. [35]

    Librispeech: An ASR corpus based on public domain audio books , year=

    Panayotov, Vassil and Chen, Guoguo and Povey, Daniel and Khudanpur, Sanjeev , booktitle=. Librispeech: An ASR corpus based on public domain audio books , year=

  28. [36]

    TED-LIUM 3: Twice as Much Data and Corpus Repartition for Experiments on Speaker Adaptation

    Hernandez, Fran c ois and Nguyen, Vincent and Ghannay, Sahar and Tomashenko, Natalia and Est \`e ve, Yannick. TED-LIUM 3: Twice as Much Data and Corpus Repartition for Experiments on Speaker Adaptation. Speech and Computer. 2018

    2018

  29. [37]

    GigaSpeech: An Evolving, Multi-domain ASR Corpus with 10,000 Hours of Transcribed Audio , booktitle=

  30. [38]

    2022 , eprint=

    FLEURS: Few-shot Learning Evaluation of Universal Representations of Speech , author=. 2022 , eprint=

    2022

  31. [39]

    arXiv e-prints , keywords =

    SPGISpeech: 5,000 hours of transcribed financial audio for fully formatted end-to-end speech recognition. arXiv e-prints , keywords =

  32. [40]

    V ox P opuli: A Large-Scale Multilingual Speech Corpus for Representation Learning, Semi-Supervised Learning and Interpretation

    Wang, Changhan and Riviere, Morgane and Lee, Ann and Wu, Anne and Talnikar, Chaitanya and Haziza, Daniel and Williamson, Mary and Pino, Juan and Dupoux, Emmanuel. V ox P opuli: A Large-Scale Multilingual Speech Corpus for Representation Learning, Semi-Supervised Learning and I...

    2021 doi

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.