REVIEW 5 major objections 5 minor 51 references
Weight Factorization and Centralization for Continual Learning in Speech Recognition
T0 review · 5 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash
Pith's one-line read A continual-learning recipe that averages per-dataset low-rank adapters and merges them into a foundation model prevents catastrophic forgetting while still learning new code-switching speech tasks.
desk verdict A simple adapter-averaging recipe shows a genuine forgetting-mitigation effect when merging into Whisper, but the paper's centralization algorithm double-counts old adapters after the first merge, so the second-round results are not clearly supported. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The mechanism is the periodic averaging of LoRA update matrices. Each dataset learns a low-rank update $\Delta W = (\alpha/r) A B$ on query and key projections of the Whisper transformer. Periodically (every $K$ datasets), all adapters seen so far are averaged into $\Delta_{\text{avg}}$ and merged into the base weights. The paper's stated justification is that independent zero-mean Gaussian weights average to a smaller variance $\sigma^2/N$, so the merged update stays sparse and close to zero, limiting the drift of the knowledge base. The merge rule is the central object that is supposed to turn many individually destructive adapters into a nearly harmless cumulative update.
What would settle it
Measure the actual distribution of entries in the averaged adapter update $\Delta_{\text{avg}}$ after centralization on a different set of task streams (for example, languages outside the paper's six code-switching corpora) and with different values of $K$; if the average is not near zero or the backward error on held-out languages rises substantially after the merge, the centralization claim would be refuted. Concretely, run the same recipe on a second sequence of datasets and check whether per-language WER after centralization stays within a few points of the base model, as it does for German in Table 1.
Extended reading notes
Core claim
The central claim is that centrally merging multiple scattered low-rank adapters into a frozen foundation model prevents catastrophic forgetting without rehearsal. Concretely, after fine-tuning one LoRA adapter per code-switching dataset, the authors average all adapters and merge the average into the Whisper base model; repeating this periodically yields a model whose backward word-error rate on held-out languages stays close to the original (for example, German moves from 23.5% to 25.4% WER, and after the second merge Arabic improves from 23.5% to 19.9%), while forward average WER on six code-switching corpora improves from 39.4% to 28.7%. The same recipe outperforms a stochastic-weight-averaging with distillation baseline on both forward and backward metrics. The paper frames the effect as a Gaussian prior: averaging independent zero-mean weight updates reduces variance by a factor of the number of adapters, making the merged update sparse and near zero.
Load-bearing premise
The safety of merging averaged adapters into the base model rests on the empirical claim that the averaged update is sparse and close to zero, but the paper concedes its variance-reduction justification does not hold in practice because the cross-entropy loss dominates the weight-decay regularization.
Editorial extensions
If this is right
- A speech foundation model can be adapted to a sequence of code-switching datasets without access to the original training data or replay.
- The recipe needs only one hyperparameter, the number of adapters per centralization ($K$), making it easy to apply in practice.
- The centralized model delivers positive backward transfer, improving on some held-out languages relative to the base model rather than merely recovering them.
- The method can be implemented with constant extra memory by storing the running average of adapters rather than retaining all of them.
- The approach reaches only about 58% of the improvement of an oracle fine-tune on all datasets, leaving room for better merging strategies.
Reading between the lines
- If the averaging rationale holds generally, the same factorization-and-centralization recipe should transfer to other foundation-model adaptation problems, such as adapting vision or language models to a sequence of domains with LoRA-style adapters.
- The paper's Gaussian justification is violated by the data-dependent cross-entropy loss; a more direct test would be to measure the distribution of per-entry adapter weights after training and check how sparsity of the merged update correlates with backward stability across different task orders and $K$ values.
- The observed positive backward transfer suggests the merging acts as a form of implicit regularization that may favor flatter minima; whether this is specific to speech or general to merged low-rank updates could be tested by comparing loss landscapes before and after merging.
- Algorithm 1's averaging step with weight $1/t$ is ambiguous about whether previously merged adapters are double-counted; a clean reformulation that tracks the running average explicitly would make the method more reproducible.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The manuscript proposes a rehearsal-free continual learning method for ASR foundation models. In the factorization stage, the incoming stream of datasets is scattered across per-dataset LoRA adapters. In the centralization stage, after every K=3 datasets all stored adapters are averaged and merged into the Whisper base model via a LoRA merge. Experiments on six code-switching datasets compare the centralized model with a fine-tuned ceiling and with an SWA-plus-distillation baseline (SW ADT). The paper reports forward WER improvements (average 39.4 to 28.7) and backward error-rate recovery from severe forgetting caused by individual adapters; the second centralization row in Table 2 is the main evidence that the method outperforms SW ADT on average backward metrics. The proposed mechanism is motivated by a Gaussian variance-reduction argument, but the paper concedes in Section 3.2 that this premise is violated in practice. The contribution is primarily empirical, with the method advertised as simple and scalable.
Significance. If the experimental results hold after a corrected implementation and with proper statistical support, the method would be a lightweight, rehearsal-free alternative for continually adapting foundation ASR models to streams of code-switched data. The reported backward recovery is striking: for example, Arabic WER moves from 49.0 after a single adapter to 25.4 after centralization, and Turkish from 43.5 to 15.4 in Table 1. The comparison against SW ADT is appropriate, and the method's small hyperparameter footprint (one centralization size K, no rehearsal data) is a practical strength. However, the central claim is currently supported by a single set of numbers, with no error bars, no significance tests, no ablation of K, and an accumulation rule that is internally inconsistent at the second centralization. The significance is therefore conditional on corrections that the manuscript itself needs to provide.
major comments (5)
- [Algorithm 1, lines 7-9; Table 2] The accumulation rule in Algorithm 1 double-counts the first block of adapters at the second centralization, so the '2nd Cent.' row of Table 2 does not follow from the stated procedure. After the first centralization at t=K, adapters Δθ_j for j=K+1,...,2K are trained on θ_K and are therefore deltas relative to θ_K, not to θ0. Writing A1=(1/K)Σ_{j=1}^K Δθ_j and B1=(1/K)Σ_{j=K+1}^{2K} Δθ_j, line 8 with t=2K gives θ_{2K}=θ0+A1+(1/2)A1+(1/2)B1, i.e., an effective weight of 1.5 on A1, whereas a non-double-counting cumulative average expressed relative to θ0 would be θ0+A1+(1/2)B1. The efficient variant described after Algorithm 1, which stores Δavg after each centralization and 'rescaled and added to the new adapters', does not specify the rescale factor or show that later adapters are expressed relative to θ0. Please state the actually implemented rule, derive its per-block weighting, and re-report the second-centralization results under a non-double-counting rule.
- [Section 3.2] The Gaussian independence argument is not supported as a justification for forgetting prevention. The paper itself states that 'in practice, however, model weights have different standard deviation because the regularization term is dominated by the main cross entropy loss', which directly concedes that the zero-mean Gaussian premise is violated. In addition, adapters trained after the first centralization are conditioned on previously learned adapters, so their updates are not independent draws around θ0 at the second merge. The abstract's claim that centralization 'can effectively prevent catastrophic forgetting' therefore rests on an empirical regularity that is neither derived nor directly validated. Please provide direct evidence for the claimed mechanism, such as per-layer and per-adapter norm/sparsity statistics across both centralization blocks, an ablation of the number of adapters averaged, and a comparison with a properly anchored averaging baseline that does not re-weight earlier adapters.
- [Tables 1 and 2; Section 4.1] The experimental comparison against SW ADT is reported without error bars, confidence intervals, or significance tests. Many differences in Table 2 are small in absolute terms (e.g., English 4.8 vs 5.0, Mandarin CER 6.9 vs 8.4, Spanish 5.8 vs 6.7), and the paper does not state how many independent training runs were performed. Because ASR test sets are fixed, the relevant uncertainty is training-seed variance; please report the number of runs, mean and standard deviation over seeds, and a paired significance test (or per-seed tables) for both forward and backward metrics. Without this information, the claim that the proposed method outperforms SW ADT is not yet established.
- [Section 4.1] The paper calls K the 'only hyper-parameter to select' but provides no ablation or selection procedure for it. All experiments use K=3, and no results are shown for K=1, K=2, or larger K. Since K directly controls how many adapters are averaged and how often the base model is modified, the robustness of the forgetting-prevention effect to K is central to the method's claim. Please add an ablation over K (at least K in {1,2,3,6}) and report whether the backward recovery persists across these values.
- [Sections 3.1 and 4.1] Essential LoRA hyperparameters are not reported, and some setup details are missing. In particular, the rank r, scaling factor α, learning rate, number of update steps per adapter, weight decay coefficient, and the LoRA target layers (query and key are mentioned, but not whether projection layers were also adapted) are not given. The 'Fine-tuned' ceiling row in Table 2 also lacks details such as how many adapters or steps were used and whether it was trained on all six datasets jointly. Without these details the method is not reproducible, and the claimed simplicity of the approach cannot be assessed.
minor comments (5)
- [Various] There are several typos, including 'contitnual' and 'framwework' in the Table 2 caption, 'serendeipitously' in the Section 4.2 heading, and 'robost' in the Introduction. Please proofread the manuscript.
- [Equation (1)] The indexing in Equation (1) is ambiguous: Nt and Nj are used without a clear definition, and the sums over previously seen datasets are not written consistently. Please define T, N_t, and the cumulative risk more explicitly.
- [Algorithm 1 note] The sentence after Algorithm 1 describing the efficient implementation says the stored Δavg is 'rescaled and added to the new adapters' but gives no rescale factor. Please specify the formula so that the rule can be checked against the reported experiments.
- [Section 4.2] The paper mentions measuring sparsity of Δθ_t and states that 'the averaged model has much higher sparsity than any other adapter', but no quantitative sparsity values are reported. Please add a table or figure with these statistics.
- [References] Some references are incomplete, e.g., reference [17] lacks full bibliographic information and reference [26] does not give a complete venue. Please complete the bibliography.
Circularity Check
No load-bearing circularity; the Gaussian justification is self-undermined and Algorithm 1 has a non-circular correctness ambiguity.
full rationale
The paper's central claim is empirical: averaging per-dataset LoRA adapters and merging them into the Whisper base reduces backward WER degradation. No parameter is fitted to the target results; the only mentioned hyperparameter, K, is fixed at 3 by grouping the six code-switching datasets. The Gaussian variance argument in Section 3.2 is explicitly conceded to be violated in practice ('In practice, however, model weights have different standard deviation because the regularization term is dominated by the main cross entropy loss, which depends on the data'), so it is a broken motivation rather than a self-fulfilling derivation. The paper also admits in the conclusion that explaining the effectiveness of this intuition is future work. Self-citations in the paper concern adapters, code-switching metrics, and benchmark datasets; they are not uniqueness theorems and do not carry the forgetting-prevention claim. The main correctness concern is Algorithm 1's accumulation rule: adapters trained after the first centralization are conditioned on the merged base, while line 8 averages them with earlier adapters trained relative to the original base. If taken literally, the second merge re-adds the old adapter average, which is an algorithmic artifact that could affect the '2nd Cent.' row, but it is not a circular reduction of a prediction to its inputs, and the first centralization result is reported independently. Therefore, no significant circularity is found.
Assumptions & free parameters
free parameters (6)
- K (adapters per centralization batch) =
3
- LoRA rank r =
not reported
- LoRA scaling factor alpha =
not reported
- Weight decay coefficient =
not reported
- Learning rate and update steps per adapter =
not reported
- Number of centralization rounds =
2
assumptions (4)
- standard math Independent zero-mean Gaussian adapter weights make averaging reduce variance by 1/N (Section 3.2 formula).
- ad hoc to paper Adapters trained on different datasets behave as independent draws from N(0, sigma^2).
- domain assumption The base model can be safely merged with the averaged adapters via lora_merge without retraining or data from the foundation model's original corpus.
- ad hoc to paper K=3 dataset grouping is a valid batch size for the continual learning stream and the two-round stream is representative.
Cite this review
Pith. "Pith review of Weight Factorization and Centralization for Continual Learning in Speech Recognition." pith.science (2026). https://pith.science/paper/MO64WUS2
@misc{pith2026250616574,
author = {Pith},
title = {Pith review of: Weight Factorization and Centralization for Continual Learning in Speech Recognition},
year = {2026},
howpublished = {\url{https://pith.science/paper/MO64WUS2}},
note = {Machine review of arXiv:2506.16574}
}
read the original abstract
Modern neural network based speech recognition models are required to continually absorb new data without re-training the whole system, especially in downstream applications using foundation models, having no access to the original training data. Continually training the models in a rehearsal-free, multilingual, and language agnostic condition, likely leads to catastrophic forgetting, when a seemingly insignificant disruption to the weights can destructively harm the quality of the models. Inspired by the ability of human brains to learn and consolidate knowledge through the waking-sleeping cycle, we propose a continual learning approach with two distinct phases: factorization and centralization, learning and merging knowledge accordingly. Our experiments on a sequence of varied code-switching datasets showed that the centralization stage can effectively prevent catastrophic forgetting by accumulating the knowledge in multiple scattering low-rank adapters.
Figures
Reference graph
Works this paper leans on
-
[1]
Introduction Large-scale multilingual speech models [1] have become the backbone of speech applications, thanks to the high perfor- mance achieved in multiple languages, enabled by scaling the models, computation and data. Despite such achievement, the one-size-fit-all solution might still require further fine- tuning, in order to optimize the performance...
arXiv 2025
-
[2]
Related Works We consider a setting where an automatic speech recognizer (ASR) is trained by observing the datasets [ D1,D 2,...D T ] sequentially. Each dataset Dt consists of samples (xi t,y i t) be- ing the input utterances and labels respectively. At each iter- ation t, not only does the model parameters θt have to opti- mize for the current dataset Dt...
-
[3]
Main approach Motivated by the daytime–nighttime cycle and the role of sleep in memory consolidation in the human brain, we propose de- coupling the typical learning and regularization objectives in continual learning into two distinct phases. In the factoriza- tion phase, we aim to expand the knowledge base model using temporary, learnable adapters on th...
-
[4]
This equation suggests that, the more factorization we have before centralization, the more possible it is to be robust to for- getting (when the summation of knowledge-base and adapters is less likely to deviate too far). In practice, however, model weights have different standard deviation because the regular- ization term is dominated by the main cross...
-
[5]
Experiments 4.1. Experiment Setup Code-switching—rapid alternation of languages within the same utterance—remains harder than monolingual ASR be- cause switch points are brief, pronunciation patterns overlap, and the available speech often exhibits distributional shifts such as telephone speech and data sparsity. Early work ap- plied language identificati...
-
[6]
Conclusion In this paper, we proposed a continual learning strategy moti- vated from the wake-sleep cycle of human brain activity, re- alized by having two separated phases of learning via factor- ization and consolidation via centralization. The main idea is to separate the input data into segments and maintain different model extensions, followed by a c...
-
[7]
Acknowledgements This work was supported by BMBF (grant 01EF1803B), the EU Horizon program (grant 101135798, Meetween), the Helmholtz Association, the HoreKa supercomputer funded by the Min- istry of Science, Research and the Arts Baden-W¨urttemberg and BMBF, and Zoom Video Communications grant
-
[8]
Robust speech recognition via large-scale weak su- pervision,
A. Radford, J. W. Kim, T. Xu, G. Brockman, C. McLeavey, and I. Sutskever, “Robust speech recognition via large-scale weak su- pervision,” in International conference on machine learning
Show all 51 references
-
[9]
Code-switching with- out switching: Language agnostic end-to-end speech translation,
C. Huber, E. Y . Ugan, and A. Waibel, “Code-switching with- out switching: Language agnostic end-to-end speech translation,” arXiv preprint arXiv:2210.01512, 2022
2022 arXiv
-
[10]
Automatic extraction of named entity translingual equivalence based on multi-feature cost minimization,
F. Huang, S. V ogel, and A. Waibel, “Automatic extraction of named entity translingual equivalence based on multi-feature cost minimization,” inProceedings of the ACL 2003 workshop on Mul- tilingual and mixed-language named entity recognition , 2003
2003
-
[11]
Towards better language models for spontaneous speech,
B. Suhm and A. Waibel, “Towards better language models for spontaneous speech,” in Proc. ICSLP’94, vol. 2, 1994
1994
-
[12]
Multilingual ar- ticulatory features,
S. Stuker, T. Schultz, F. Metze, and A. Waibel, “Multilingual ar- ticulatory features,” in ICASSP’2003
2003
-
[13]
Simultaneous translation of open do- main lectures and speeches,
A. Waibel and C. Fuegen, “Simultaneous translation of open do- main lectures and speeches,” Jan. 3 2012, uS Patent 8,090,570
2012
-
[14]
Chil: Computers in the human interaction loop,
A. Waibe11, H. Steusloff, R. Stiefelhagen et al., “Chil: Computers in the human interaction loop,” 2005
2005
-
[15]
Efficient Weight Factorization for Multilingual Speech Recognition,
N.-Q. Pham, T.-N. Nguyen, S. St ¨uker, and A. Waibel, “Efficient Weight Factorization for Multilingual Speech Recognition,” in Proc. Interspeech 2021, 2021, pp. 2421–2425
2021
-
[16]
Lora: Low-rank adaptation of large lan- guage models,
E. J. Hu, Y . Shen, P. Wallis, Z. Allen-Zhu, Y . Li, S. Wang, L. Wang, and W. Chen, “Lora: Low-rank adaptation of large lan- guage models,” arXiv preprint arXiv:2106.09685, 2021
2021 arXiv
-
[17]
Catastrophic forgetting, rehearsal and pseudore- hearsal,
A. Robins, “Catastrophic forgetting, rehearsal and pseudore- hearsal,” Connection Science, vol. 7, no. 2, pp. 123–146, 1995
1995
-
[18]
Continual learning with deep generative replay,
H. Shin, J. K. Lee, and J. Kim, “Continual learning with deep generative replay,” inAdvances in Neural Information Processing Systems
-
[19]
Overcoming catastrophic forgetting in neural net- works,
J. Kirkpatrick, R. Pascanu, N. Rabinowitz, J. Veness, G. Des- jardins et al., “Overcoming catastrophic forgetting in neural net- works,” Proceedings of the national academy of sciences , 2017
2017
-
[20]
Regularize, expand and compress: Non- expansive continual learning,
J. Zhang, J. Zhang et al., “Regularize, expand and compress: Non- expansive continual learning,” in Proceedings of the IEEE/CVF winter conference on applications of computer vision , 2020
2020
-
[21]
Progressive prompts: Continual learning for lan- guage models,
A. Razdaibiedina, Y . Mao, R. Hou, M. Khabsa, M. Lewis, and A. Almahairi, “Progressive prompts: Continual learning for lan- guage models,” arXiv preprint arXiv:2301.12314, 2023
2023 arXiv
-
[22]
Towards continually learning new languages,
N.-Q. Pham, J. Niehues, and A. Waibel, “Towards continually learning new languages,” in Interspeech 2023, 2023
2023
-
[23]
Averaging weights leads to wider optima and better gen- eralization,
P. Izmailov, D. Podoprikhin, T. Garipov, D. Vetrov, and A. G. Wilson, “Averaging weights leads to wider optima and better gen- eralization,” arXiv preprint arXiv:1803.05407, 2018
2018 arXiv
-
[24]
Streaming diloco with overlapping communication: Towards a distributed free lunch
A. Douillard, Y . Donchev, K. Rush et al., “Streaming diloco with overlapping communication: Towards a distributed free lunch.”
-
[25]
wav2vec 2.0: A framework for self-supervised learning of speech representa- tions,
A. Baevski, Y . Zhou, A. Mohamed, and M. Auli, “wav2vec 2.0: A framework for self-supervised learning of speech representa- tions,” Advances in Neural Information Processing Systems
-
[26]
Continual learning through synaptic intelligence,
F. Zenke, B. Poole, and S. Ganguli, “Continual learning through synaptic intelligence,” in International Conference on Machine Learning. PMLR, 2017, pp. 3987–3995
2017
-
[27]
Variational continual learning,
C. V . Nguyen, Y . Li, T. D. Bui, and R. E. Turner, “Variational continual learning,” arXiv preprint arXiv:1710.10628, 2017
2017 arXiv
-
[28]
A unifying bayesian view of continual learning,
S. Farquhar and Y . Gal, “A unifying bayesian view of continual learning,” arXiv preprint arXiv:1902.06494, 2019
1902 arXiv
-
[29]
Gradient episodic memory for continual learning,
D. Lopez-Paz and M. Ranzato, “Gradient episodic memory for continual learning,” Advances in neural information processing systems, vol. 30, 2017
2017
-
[30]
Coda-prompt: Con- tinual decomposed attention-based prompting for rehearsal-free continual learning,
J. S. Smith, L. Karlinsky, V . Gutta et al. , “Coda-prompt: Con- tinual decomposed attention-based prompting for rehearsal-free continual learning,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition
-
[31]
Learn and don’t forget: Adding a new language to asr foundation models,
M. Qian, S. Tang, R. Ma, K. M. Knill, and M. J. Gales, “Learn and don’t forget: Adding a new language to asr foundation models,” in Interspeech 2024, 2024, pp. 2544–2548
2024
-
[32]
Continuously learning new words in automatic speech recognition,
C. Huber and A. Waibel, “Continuously learning new words in automatic speech recognition,” 2024
2024
-
[33]
Rehearsal-free online continual learning for automatic speech recognition,
S. Vander Eeckt et al., “Rehearsal-free online continual learning for automatic speech recognition,”Proceedings Interspeech 2023
2023
-
[34]
Lvcsr-based language iden- tification,
T. Schultz, I. Rogina, and A. Waibel, “Lvcsr-based language iden- tification,” in 1996 IEEE International Conference on Acoustics, Speech, and Signal Processing Conference Proceedings
1996
-
[35]
Experiments on cross-language acous- tic modeling
T. Schultz and A. Waibel, “Experiments on cross-language acous- tic modeling.” in INTERSPEECH, 2001, pp. 2721–2724
2001
-
[36]
Language- agnostic code-switching in sequence-to-sequence speech recogni- tion,
E. Y . Ugan, C. Huber, J. Hussain, and A. Waibel, “Language- agnostic code-switching in sequence-to-sequence speech recogni- tion,” arXiv preprint arXiv:2210.08992, 2022
2022 arXiv
-
[37]
Arzen: A speech corpus for code-switched egyptian arabic-english,
I. Hamed, N. T. Vu, and S. Abdennadher, “Arzen: A speech corpus for code-switched egyptian arabic-english,” in Proceedings of the 12th Language Resources and Evaluation Conference , 2020
2020
-
[38]
End-to-end speech translation for code switched speech,
O. Weller, M. Sperber et al. , “End-to-end speech translation for code switched speech,” inFindings of the Association for Compu- tational Linguistics: ACL 2022 , 2022
2022
-
[39]
Seame: a mandarin-english code- switching speech corpus in south-east asia
D.-C. Lyu, T. P. Tan et al. , “Seame: a mandarin-english code- switching speech corpus in south-east asia.” Interspeech, 2010
2010
-
[40]
Leveraging data collec- tion and unsupervised learning for code-switched tunisian arabic automatic speech recognition,
A. A. B. Abdallah, A. Kabboudi et al., “Leveraging data collec- tion and unsupervised learning for code-switched tunisian arabic automatic speech recognition,” ICASSP, 2024
2024
-
[41]
ASCEND: A spontaneous Chinese-English dataset for code- switching in multi-turn conversation,
H. Lovenia, S. Cahyawijaya, G. Winata, P. Xu, Y . Xu, et al. , “ASCEND: A spontaneous Chinese-English dataset for code- switching in multi-turn conversation,” in Proceedings of the 13th Language Resources and Evaluation Conference
-
[42]
Talcs: An open-source mandarin-english code-switching corpus and a speech recognition baseline
C. Li, S. Deng, Y . Wang, G. Wang, Y . Gong, C. Chen, and J. Bai, “Talcs: An open-source mandarin-english code-switching corpus and a speech recognition baseline.”
-
[43]
Pier: A novel metric for evaluating what matters in code-switching,
E. Y . Ugan, N.-Q. Pham, L. B ¨armann, and A. Waibel, “Pier: A novel metric for evaluating what matters in code-switching,” arXiv preprint arXiv:2501.09512, 2025
2025 arXiv
-
[44]
Ted-lium 3: Twice as much data and corpus repartition for experiments on speaker adaptation,
F. Hernandez, V . Nguyen, S. Ghannay, N. Tomashenko, and Y . Es- teve, “Ted-lium 3: Twice as much data and corpus repartition for experiments on speaker adaptation,” in SPECOM 2018
2018
-
[45]
Mls: A large-scale multilingual dataset for speech research,
V . Pratap, Q. Xu, A. Sriram, G. Synnaeve, and R. Collobert, “Mls: A large-scale multilingual dataset for speech research,” 2020
2020
-
[46]
The mgb-2 challenge: Arabic multi-dialect broadcast media recognition,
A. Ali, P. Bell, J. Glass, Y . Messaoui, H. Mubarak, S. Renals, and Y . Zhang, “The mgb-2 challenge: Arabic multi-dialect broadcast media recognition,” in 2016 IEEE Spoken Language Technology Workshop (SLT). IEEE, 2016, pp. 279–284
2016
-
[47]
Multilingual speech recognition for turkic languages,
S. Mussakhojayeva, K. Dauletbek, R. Yeshpanov, and H. A. Varol, “Multilingual speech recognition for turkic languages,” Informa- tion, vol. 14, no. 2, p. 74, 2023
2023
-
[48]
Aishell-1: An open- source mandarin speech corpus and a speech recognition base- line,
H. Bu, J. Du, X. Na, B. Wu, and H. Zheng, “Aishell-1: An open- source mandarin speech corpus and a speech recognition base- line,” in Oriental COCOSDA 2017, 2017, p. Submitted
2017
-
[49]
Common voice: A massively- multilingual speech corpus,
R. Ardila, M. Branson et al. , “Common voice: A massively- multilingual speech corpus,” in Proceedings of the Twelfth Lan- guage Resources and Evaluation Conference, 2020
2020
-
[50]
Decm: Evaluating bilin- gual asr performance on a code-switching/mixing benchmark,
E. Y . Ugan, N.-Q. Pham, and A. Waibel, “Decm: Evaluating bilin- gual asr performance on a code-switching/mixing benchmark,” 2024, pp. 4468–4475
2024
-
[51]
Dark experience for general continual learning: a strong, sim- ple baseline,
P. Buzzega, M. Boschini, A. Porrello, D. Abati, and S. Calderara, “Dark experience for general continual learning: a strong, sim- ple baseline,” Advances in neural information processing systems, vol. 33, pp. 15 920–15 930, 2020
2020
Reviewed August 15, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.