Pith. sign in

REVIEW 3 major objections 5 minor 33 references

persoDA: Personalized Data Augmentation for Personalized ASR

T0 review · 3 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 User-specific acoustics cut ASR errors by 13.9 percent.

desk verdict Plausible method, but the headline WER gain is confounded by using the evaluation set to build the augmentation, and the disjoint-split control that would fix it is not quantified. read the letter →

arxiv 2501.09113 v2 pith:JMF53X4M submitted 2025-01-15 eess.AS cs.SD

Pablo Peso Parada , Spyros Fontalis , Md Asif Jalal , Karthikeyan Saravanan , Anastasios Drosou , Mete Ozay , Gil Ho Lee , Jungin Lee
show 1 more author
Seokyeong Jung
This is my paper · ORCID
classification eess.AScs.SD
keywords personalizeddataaugmentationASRpersonalizationspeechrecognitionmulti-conditiontrainingvoiceactivitydetectionreverberationtimeon-deviceadaptationworderrorrate
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

persoDA is a data-augmentation strategy for personalizing on-device speech recognition to a specific user. Instead of corrupting the user's clean training utterances with randomly chosen noise and room impulse responses, it first estimates the user's own acoustic environment—background noise captured with voice activity detection and reverberation time measured from their recordings—and then uses those estimates to build the augmentation set. The paper reports that this personalized augmentation lowers word error rate by 13.9% relative to standard multi-condition training and reaches the same accuracy 16–20% sooner. The appeal is that better personalization can be obtained from the same limited unlabeled user data, with no extra labels or heavy computation.

What carries the argument

The load-bearing mechanism is a user-driven augmentation-parameter selector: the system solves $\arg\max_{\theta} P(\theta\mid E)$, where $E$ is the acoustic environment observed in the user's recordings and $\theta$ is a concrete audio distortion. persoNoise realizes this for additive noise by using voice activity detection to isolate non-speech frames, normalizing them to a fixed RMS level, crossfading them into long noise recordings, and optionally adding similar noise samples from an external corpus via cosine similarity in the power-spectral-density domain. persoReverb realizes it for reverberation by estimating $T_{60}$, the time for sound pressure to decay by 60 dB after the source stops, from the user's utterances and selecting the room impulse response with the nearest $T_{60}$ from a database. These selected distortions are applied only to the clean subset of the user's data during fine-tuning.

What would settle it

Run the disjoint-split experiment of Section IV-D with numeric reporting: extract persoDA parameters from one user subset, evaluate on the held-out acoustic environments of the other subset, and compare final WERs against MCT. If persoDA no longer beats random augmentation on that unseen set by a margin beyond run-to-run noise, the personalization-advantage claim collapses.

Watch

Extended reading notes

Core claim

On the paper's terms, the discovery is that the optimal data-augmentation parameters for personalizing an ASR model are not random but are the maximum-likelihood parameters of the user's own acoustic environment, $\theta$ chosen to maximize $P(\theta\mid E)$ from the environment characteristics observed in the user's data. For noise, the persoNoise module concatenates non-speech frames detected by a VAD into noise recordings; for reverberation, persoReverb estimates $T_{60}$ from the user's utterances and picks the room impulse response with the closest $T_{60}$. When a Conformer-based ASR model pretrained on LibriSpeech is fine-tuned on clean versions of the user's utterances augmented this way, and evaluated on the noisy reverberant versions of those utterances in the VOICES set, persoDA reaches 3.04% WER versus 3.53% for random noise-and-reverb augmentation, a 13.88% relative improvement, with no increase in global WER and with 16–20% faster convergence.

Load-bearing premise

The main result is measured on a validation set that also supplied the noise and reverberation characteristics used for augmentation, so the test conditions overlap with the training augmentation conditions.

Editorial extensions

If this is right

  • Using matched user-environment augmentation instead of random noise and reverberation lowers the personalized WER from 3.53% to 3.04% on the VOICES validation set.
  • The gains come without harming general performance: global WER stays at 4.68%, so the method does not increase catastrophic forgetting.
  • The same target WER is reached 80–100 steps earlier, a 16–20% convergence speedup that matters for battery- and compute-limited on-device adaptation.
  • Both components contribute independently: persoReverb improves 7.45% and persoNoise improves 10.81% relative WER over their random counterparts.
  • Pseudo-label training also benefits, so the method can personalize without ground-truth transcripts from the user.

Reading between the lines

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

  • Because the headline 13.9% gain is measured on a validation set whose noise and reverberation parameters were extracted from that same set, the real-world gain on environments the user has not yet been recorded in is likely smaller; the disjoint-split experiment in Section IV-D reports only in figures, and its numerical WER gap should be read as an upper bound on transfer.
  • The VAD-based noise extraction inherits all VAD failure modes: if the user's recordings contain speech-like background such as television or other talkers, non-speech frames will be contaminated and the noise profile will drift toward speech, so the method's benefit depends on how clean the VAD decision is.
  • persoReverb can only select an impulse response that exists in the database; a user whose room acoustics fall outside the 325-RIR collection would silently receive a mismatched augmentation, which suggests testing database-coverage sensitivity as an extension.
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 proposes persoDA, a framework for personalized data augmentation in ASR personalization. It extracts per-user background noise via VAD-based segmentation (persoNoise) and per-user reverberation characteristics via blind T60 estimation (persoReverb), then uses these to augment a clean personalization training set T with matched noise and room impulse responses. The method is evaluated by fine-tuning a SpeechBrain Conformer ASR model pre-trained on LibriSpeech and adapting it to VOiCES user data, with comparison to standard Multi-Condition Training (MCT) augmentation using random noise and RIRs. The main reported result is a 13.9% relative WER reduction over MCT on the user validation set V, together with 16%-20% faster convergence. The paper also studies a disjoint train/validation split and a pseudo-label scenario.

Significance. If the reported gains hold under proper evaluation, persoDA is a practically attractive method for on-device ASR personalization: it is computationally light, requires only a small RIR database on the device, and uses unlabeled user data. The paper includes useful ablations (VAD vs. speech-separation noise extraction, two T60 estimators, with/without data grouping) and a pseudo-label experiment that addresses a realistic deployment setting. Its main weakness is that the headline 13.9% figure is measured on the same set V from which the augmentation parameters are estimated, and the disjoint-split control, which is the correct test, is only presented graphically without numeric WERs or a check of acoustic-condition overlap. The work is relevant to the speech community, but the central claim needs stronger experimental support.

major comments (3)
  1. [Section IV-C and Table II] The 13.9% relative WER improvement is computed on validation set V, which is the same set used to estimate the persoNoise and persoReverb parameters theta. The paper explicitly acknowledges this in Section IV: 'the distribution of theta applied to T overlaps with that found in V.' Consequently, the improvement may partly reflect matching of test conditions rather than a genuine personalization benefit. To support the central claim, the authors should report the disjoint-split experiment of Section IV-D with numeric WERs for V'' and demonstrate that the improvement persists when theta is estimated from data that does not share acoustic conditions with the evaluation set.
  2. [Section IV-D] The disjoint-split control does not guarantee acoustic disjointness. VOiCES was created by replaying LibriSpeech utterances in noisy, reverberant rooms; splitting V into V' and V'' by utterances does not prevent the same room, noise source, or T60 condition from appearing in both subsets. If a condition appears in both, then theta estimated from V' includes the very conditions later scored in V'', so Fig. 3 does not establish performance on unseen acoustic environments. The paper does not report the acoustic-session assignment or any check that the noise/reverberation conditions in V' and V'' are non-overlapping. Please provide this information or run the disjoint split at the condition/session level.
  3. [Tables I-III and Figs. 2-3] All reported WERs appear to come from a single run without error bars or multiple seeds. The gap between persoDA (3.04) and the best MCT combination (3.35) in Table II is 0.31 WER absolute; without variance estimates it is unclear whether this difference is statistically significant. Similarly, the convergence-speed claim (16%-20%) is derived from visual inspection of Fig. 2 without numeric step counts or confidence intervals. Please provide multiple seeds or confidence intervals for the main comparisons, and tabulate the disjoint-split and convergence results.
minor comments (5)
  1. [Table I caption] The caption contains a typo: 'T RAINED' should be 'TRAINED'.
  2. [Table II caption] The caption contains 'M ETHODS' and 'reverberation' should be 'reverberation'.
  3. [Section III-B] The sentence 'The total size of the training data is reduced to 940h after excluding the speakers included in validation set V to avoid data overlap with the evaluation set of the VOiCES dataset' is ambiguous: it seems to say LibriSpeech speakers are excluded because they appear in VOiCES evaluation, but V is defined as VOiCES validation data. Please clarify the exclusion procedure.
  4. [Section III-A] The 'best probability for MCT of including reverb or including noise is found empirically for each method' is not reported. Since these probabilities are free parameters, please state the chosen values or explain how they were selected.
  5. [Section II-B] When selecting 'the RIR from the database with the closest T60', it is not specified whether the selection is deterministic or whether ties are broken randomly. Please clarify.

Circularity Check

2 steps flagged · score 6.0 of 10

Headline 13.9% WER gain is measured on V after persoDA extracts noise/reverb parameters θ from V itself; the Sec. IV-D disjoint-split control only splits utterances, not acoustic conditions, so the loop may not actually be broken.

  1. fitted input called prediction [Section IV preamble and Section IV-C (Table II); also Section III-C metrics]
    "Noise and reverberation characteristics for persoNoise and persoReverb, parametrized by θ, are obtained from V, and randomly applied to the personalization task involving utterances in T . In sections IV-A to IV-C, the distribution of θ applied to T overlaps with that found in V."

    The augmentation parameters θ (VAD-extracted noise segments and T60-matched RIRs) are estimated from the very evaluation set V whose WER is reported as the headline result. Table II's 'Valid' column is computed on V (Section III-C: 'Valid WER, computed on V or subset of V, shows the personalization performance'). Thus persoDA is given the test acoustic distribution before evaluation, while the MCT baseline draws noise/RIRs randomly from generic databases. The advertised 13.9% relative improvement therefore reduces, at least in part, to training with the test conditions and then measuring on those same conditions; it is not an assessment on unseen user acoustics. The paper acknowledges the overlap and defers to Sec. IV-D, but the main claim stands on the overlapping setup.

  2. other [Section IV-D and Fig. 3]
    "A similar split is applied to V where V ′ and V ′′ comprise the noisy reverberant versions of the utterances in T ′ and T ′′ respectively. Thus, in this setup, the training phase of model personalization uses only T ′ set, and persoDA uses only V ′ to estimate information to guide the DA. This allows to measure the performance of models on the unseen acoustic environments in V ′′."

    The control claims to evaluate on 'unseen acoustic environments', but 'unseen' is defined only by utterance disjointness (T' vs T''), not by acoustic-condition disjointness. VOiCES consists of LibriSpeech utterances replayed in noisy, reverberant rooms; if the same room/noise/T60 condition occurs in both V' and V'', then θ estimated from V' (noise frames, T60-selected RIRs) includes the very conditions scored in V''. The ablation therefore does not guarantee that persoDA is tested outside the distribution used to choose its augmentations. The paper neither reports the acoustic session/room assignment of V' and V'' nor tabulates the disjoint-split WERs, so the only evidence that the loop is broken is a figure whose premise of acoustic disjointness is unstated and untested.

full rationale

The circularity here is not a self-citation chain or an imported uniqueness theorem; it is an evaluation-loop problem. The central quantitative claim (13.9% relative WER improvement over MCT, and the 16-20% faster convergence) is produced in a setup where the persoDA parameters θ are extracted from the same evaluation set V on which WER is reported. The paper honestly discloses this overlap and attempts a control in Section IV-D, which is the right experimental idea. However, the control is only shown as figures, and its claim to test 'unseen acoustic environments' is based on an utterance-level split rather than a verified acoustic-condition split. If the same rooms/noise conditions appear on both sides of the split, the control inherits the same leakage. These considerations make the headline result partially circular as evidence for generalization: persoDA's advantage may substantially reflect test-condition matching rather than a generally better personalization strategy. The method itself has independent content (VAD noise extraction, T60-based RIR selection, comparison against MCT), so the paper is not wholly derivative; score 6 reflects partial circularity confined to the evaluation protocol.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

No new physical entities are introduced. The method introduces several hand-set and validation-fitted hyperparameters; the central comparison uses the validation set to both guide augmentation and measure WER.

free parameters (4)
  • MCT inclusion probabilities for noise/reverb = tuned empirically per method
    Section III-A: 'The best probability for MCT of including reverb or including noise is found empirically for each method.' Fit to the validation set.
  • Noise RMS normalization level = 5 dB
    Section II-A1: chosen by hand to concatenate noise segments smoothly; affects the level of augmented noise.
  • Crossfade length = 100 ms
    Section IV-A: used for all persoNoise experiments; chosen by hand.
  • Noise segment length threshold = unspecified
    Section II-A1: 'noise segments longer than a threshold are concatenated'; the threshold value is not stated, making part of the pipeline irreproducible.
assumptions (3)
  • domain assumption The user's recording set U can be split into V (high distortion) and T (low distortion) such that noisy V_i = f(T_i).
    Section II and III-B: explicitly constructed from VOiCES, so this does not reflect natural user data distributions.
  • domain assumption Voice activity detection (WebRTC VAD) reliably identifies non-speech frames suitable for noise extraction.
    Section II-A1: the noise profile is taken from VAD-classified non-speech frames; if VAD mislabels speech as noise, the augmented noise is corrupted.
  • domain assumption T60 estimators from [22] and [26] give accurate room reverberation estimates for RIR selection.
    Section II-B: the closest RIR is chosen based on estimated T60; estimation errors propagate to the augmentation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of persoDA: Personalized Data Augmentation for Personalized ASR." pith.science (2026). https://pith.science/paper/JMF53X4M

@misc{pith2026250109113,
  author       = {Pith},
  title        = {Pith review of: persoDA: Personalized Data Augmentation for Personalized ASR},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JMF53X4M}},
  note         = {Machine review of arXiv:2501.09113}
}
read the original abstract

Data augmentation (DA) is ubiquitously used in training of Automatic Speech Recognition (ASR) models. DA offers increased data variability, robustness and generalization against different acoustic distortions. Recently, personalization of ASR models on mobile devices has been shown to improve Word Error Rate (WER). This paper evaluates data augmentation in this context and proposes persoDA; a DA method driven by user's data utilized to personalize ASR. persoDA aims to augment training with data specifically tuned towards acoustic characteristics of the end-user, as opposed to standard augmentation based on Multi-Condition Training (MCT) that applies random reverberation and noises. Our evaluation with an ASR conformer-based baseline trained on Librispeech and personalized for VOICES shows that persoDA achieves a 13.9% relative WER reduction over using standard data augmentation (using random noise & reverberation). Furthermore, persoDA shows 16% to 20% faster convergence over MCT.

Figures

Figures reproduced from arXiv: 2501.09113 by the authors.

Figure 1
Figure 1. The personalized data augmentation (persoDA) framework. persoDA guides the training DA process to select the most adequate augmentation given user’s data. speech synthesis to synthesize personalized speech for rapid speaker adaptation is investigated for hybrid systems [11] and RNN-T [12] models. An informed method of enabling automatic selection and parameterization of a DA pipeline for self-supervised learning is … view at source ↗
Figure 2
Figure 2. shows the performance on V ′ indicating that the proposed approach outperforms the MCT approach w/ labels and w/o labels (i.e. using pseudo-labels). The graph also suggests faster converge for persoDA. Indeed, the final WER achieved by MCT w/ labels and w/o is reached 80 and 100 steps earlier respectively with the persoDA counterparts, thus, achieving 16% to 20% faster convergence. TABLE III COMPARISON OF STANDARD D… view at source ↗
Figure 3
Figure 3. WER achieved on the set V ′′ which comprises unseen data. Experiments ‘w/ labels’ were trained with ground-truth transcripts and ‘w/o labels’ were trained with pseudo-labels [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

33 extracted references · 28 canonical work pages

  1. [1]

    Fast contextual adaptation with neural associative memory for on-device personalized speech recognition,

    T. Munkhdalai, K. C. Sim, A. Chandorkar, F. Gao, M. Chua, T. Strohman, and F. Beaufays, “Fast contextual adaptation with neural associative memory for on-device personalized speech recognition,” in ICASSP, 2022, pp. 6632–6636

    work page 2022

  2. [2]

    Robust Continuous On-Device Personalization for Automatic Speech Recognition,

    K. C. Sim, A. Chandorkar, F. Gao, M. Chua, T. Munkhdalai, and F. Bea- ufays, “Robust Continuous On-Device Personalization for Automatic Speech Recognition,” in Interspeech, 2021, pp. 1284–1288

    work page 2021

  3. [3]

    UserLibri: A Dataset for ASR Personalization Using Only Text ,

    T. Breiner, S. Ramaswamy, E. Variani, S. Garg, R. Mathews, K. C. Sim, K. G. v Mingqing Chen, and L. McConnaughey, “UserLibri: A Dataset for ASR Personalization Using Only Text ,” in Interspeech, 2022, pp. 694–698

    work page 2022

  4. [4]

    V oices Obscured in Complex Environmental Settings (VOiCES) Corpus,

    C. Richey, M. A. Barrios, Z. Armstrong, C. Bartels, H. Franco, M. Graciarena, A. Lawson, M. K. Nandwana, A. Stauffer, J. van Hout, P. Gamble, J. Hetherly, C. Stephenson, and K. Ni, “V oices Obscured in Complex Environmental Settings (VOiCES) Corpus,” in Interspeech, 2018, pp. 1566–1570

    work page 2018

  5. [5]

    Recent advances in end-to-end automatic speech recog- nition,

    J. Li et al. , “Recent advances in end-to-end automatic speech recog- nition,” APSIPA Transactions on Signal and Information Processing , vol. 11, no. 1, 2022

    2022

  6. [6]

    Automatic speech recognition and speech variability: A review,

    M. Benzeghiba, R. De Mori, O. Deroo, S. Dupont, T. Erbes, D. Jouvet, L. Fissore, P. Laface, A. Mertins, C. Ris et al. , “Automatic speech recognition and speech variability: A review,” Speech communication , vol. 49, no. 10-11, pp. 763–786, 2007

    work page 2007

  7. [7]

    Adaptation algorithms for neural network-based speech recognition: An overview,

    P. Bell, J. Fainberg, O. Klejch, J. Li, S. Renals, and P. Swietojanski, “Adaptation algorithms for neural network-based speech recognition: An overview,” IEEE Open Journal of Signal Processing , vol. 2, pp. 33–66, 2020

    work page 2020

  8. [8]

    Improving the training recipe for a robust conformer-based hybrid model,

    M. Zeineldeen, J. Xu, C. L ¨uscher, R. Schl ¨uter, and H. Ney, “Improving the training recipe for a robust conformer-based hybrid model,” in Interspeech, 2022, pp. 1036–1040

    work page 2022

Show all 33 references
  1. [9]

    Recent progress in the CUHK dysarthric speech recognition system,

    S. Liu, M. Geng, S. Hu, X. Xie, M. Cui, J. Yu, X. Liu, and H. Meng, “Recent progress in the CUHK dysarthric speech recognition system,” IEEE/ACM Transactions on Audio, Speech, and Language Processing , vol. 29, pp. 2267–2281, 2021

    2021

  2. [10]

    Personalization of deep learning,

    J. Schneider and M. Vlachos, “Personalization of deep learning,” in iDSC2020, 2021, pp. 89–96

    2021

  3. [11]

    Using personalized speech synthesis and neural language generator for rapid speaker adaptation,

    Y . Huang, L. He, W. Wei, W. Gale, J. Li, and Y . Gong, “Using personalized speech synthesis and neural language generator for rapid speaker adaptation,” in ICASSP, 2020, pp. 7399–7403

    2020

  4. [12]

    Rapid RNN-T Adaptation Using Personalized Speech Synthesis and Neural Language Generator,

    Y . Huang, J. Li, L. He, W. Wei, W. Gale, and Y . Gong, “Rapid RNN-T Adaptation Using Personalized Speech Synthesis and Neural Language Generator,” in Interspeech, 2020, pp. 1256–1260

    2020

  5. [13]

    Automatic data augmentation selection and parametrization in contrastive self-supervised speech rep- resentation learning,

    S. Zaiem, T. Parcollet, and S. Essid, “Automatic data augmentation selection and parametrization in contrastive self-supervised speech rep- resentation learning,” in Interspeech, 2022, pp. 669–673

    2022

  6. [14]

    Automatic optimization of data perturbation distributions for multi-style training in speech recognition,

    M. Doulaty, R. Rose, and O. Siohan, “Automatic optimization of data perturbation distributions for multi-style training in speech recognition,” in 2016 IEEE Spoken Language Technology Workshop (SLT) , 2016, pp. 21–27

    2016

  7. [15]

    Personalized adversarial data augmentation for dysarthric and elderly speech recognition,

    Z. Jin, M. Geng, J. Deng, T. Wang, S. Hu, G. Li, and X. Liu, “Personalized adversarial data augmentation for dysarthric and elderly speech recognition,” arXiv:2205.06445, 2022

    2022

  8. [16]

    Personalized speech enhancement through self-supervised data augmentation and purification,

    A. Sivaraman, S. Kim, and M. Kim, “Personalized speech enhancement through self-supervised data augmentation and purification,” in Inter- speech, 2021, pp. 2676–2680

    2021

  9. [17]

    Specaugment: A simple data augmentation method for automatic speech recognition,

    D. S. Park, W. Chan, Y . Zhang, C.-C. Chiu, B. Zoph, E. D. Cubuk, and Q. V . Le, “Specaugment: A simple data augmentation method for automatic speech recognition,” in Interspeech, 2019, pp. 2613–2617

    2019

  10. [18]

    A study on data augmentation of reverberant speech for robust speech recognition,

    T. Ko, V . Peddinti, D. Povey, M. L. Seltzer, and S. Khudanpur, “A study on data augmentation of reverberant speech for robust speech recognition,” in ICASSP, 2017, pp. 5220–5224

    2017

  11. [19]

    pMCT: Patched Multi-Condition Training for Robust Speech Recognition,

    P. Peso Parada, A. Dobrowolska, K. Saravanan, and M. Ozay, “pMCT: Patched Multi-Condition Training for Robust Speech Recognition,” in Interspeech, 2022, pp. 3779–3783

    2022

  12. [20]

    G-augment: Searching for the meta-structure of data augmentation policies for asr,

    G. Wang, E. D. Cubuk, A. Rosenberg, S. Cheng, R. J. Weiss, B. Ramab- hadran, P. J. Moreno, Q. V . Le, and D. S. Park, “G-augment: Searching for the meta-structure of data augmentation policies for asr,” in SLT, 2022, pp. 23–30

    2022

  13. [21]

    Robust signal-to-noise ratio estimation based on waveform amplitude distribution analysis,

    C. Kim and R. M. Stern, “Robust signal-to-noise ratio estimation based on waveform amplitude distribution analysis,” in Interspeech, 2008, pp. 2598–2561

    2008

  14. [22]

    An improved algorithm for blind reverberation time estimation,

    H. L ¨ollmann, E. Yilmaz, M. Jeub, and P. Vary, “An improved algorithm for blind reverberation time estimation,” in IWAENC, 2010, pp. 1– 4, https://www.mathworks.com/matlabcentral/fileexchange/35740-blind- reverberation-time-estimation

    2010

  15. [23]

    Https://github.com/wiseman/py-webrtcvad

  16. [24]

    Attention is all you need in speech separation,

    C. Subakan, M. Ravanelli, S. Cornell, M. Bronzi, and J. Zhong, “Attention is all you need in speech separation,” in ICASSP, 2021, pp. 21–25, https://huggingface.co/speechbrain/sepformer-libri3mix

    2021

  17. [25]

    Kuttruff, Room acoustics

    H. Kuttruff, Room acoustics . CRC Press, 2016

    2016

  18. [26]

    Noise-robust reverberation time estimation using spectral decay distributions with reduced compu- tational cost,

    J. Eaton, N. D. Gaubitch, and P. A. Naylor, “Noise-robust reverberation time estimation using spectral decay distributions with reduced compu- tational cost,” in ICASSP, 2013, pp. 161–165

    2013

  19. [27]

    SpeechBrain: A general- purpose speech toolkit,

    M. Ravanelli, T. Parcollet, P. Plantinga, A. Rouhe, S. Cornell, L. Lu- gosch, C. Subakan, N. Dawalatabad, A. Heba, J. Zhong, J.-C. Chou, S.-L. Yeh, S.-W. Fu, C.-F. Liao, E. Rastorgueva, F. Grondin, W. Aris, H. Na, Y . Gao, R. D. Mori, and Y . Bengio, “SpeechBrain: A general- p...

    2021 arXiv

  20. [28]

    Conformer: Convolution- augmented Transformer for Speech Recognition,

    A. Gulati, J. Qin, C.-C. Chiu, N. Parmar, Y . Zhang, J. Yu, W. Han, S. Wang, Z. Zhang, Y . Wu, and R. Pang, “Conformer: Convolution- augmented Transformer for Speech Recognition,” in Interspeech, 2020, pp. 5036–5040

    2020

  21. [29]

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

    V . Panayotov, G. Chen, D. Povey, and S. Khudanpur, “Librispeech: An ASR corpus based on public domain audio books,” in ICASSP, 2015, pp. 5206–5210

    2015

  22. [30]

    Catastrophic interference in connec- tionist networks: The sequential learning problem,

    M. McCloskey and N. J. Cohen, “Catastrophic interference in connec- tionist networks: The sequential learning problem,” in Psychology of learning and motivation . Elsevier, 1989, vol. 24, pp. 109–165

    1989

  23. [31]

    MUSAN: A music, speech, and noise corpus,

    D. Snyder, G. Chen, and D. Povey, “MUSAN: A music, speech, and noise corpus,” in arXiv, vol. abs/1510.08484 , 2015

    2015 arXiv

  24. [32]

    Self-training for end-to-end speech recognition,

    J. Kahn, A. Lee, and A. Hannun, “Self-training for end-to-end speech recognition,” in ICASSP, 2020, pp. 7084–7088

    2020

  25. [33]

    slim- IPL: Language-Model-Free Iterative Pseudo-Labeling,

    T. Likhomanenko, Q. Xu, J. Kahn, G. Synnaeve, and R. Collobert, “slim- IPL: Language-Model-Free Iterative Pseudo-Labeling,” in Interspeech 2021, 2021, pp. 741–745

    2021

Pith tools

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