Exploiting Parallel Audio Recordings to Enforce Device Invariance in CNN-based Acoustic Scene Classification

Andreas Arzt; David Eitelsebner; Gerhard Widmer; Hamid Eghbal-zadeh; Khaled Koutini; Paul Primus

REVIEW 3 major objections 6 minor 28 references

Exploiting Parallel Audio Recordings to Enforce Device Invariance in CNN-based Acoustic Scene Classification

T0 review · 3 major / 6 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read Time-aligned audio pairs let a CNN enforce device invariance by minimizing mean squared error between hidden activations, lifting target-device accuracy by nearly 24 points.

desk verdict A simple, honest DA method that works in controlled comparisons; the abstract oversells the MMD comparison, but the paper is worth a serious look. read the letter →

arxiv 1909.02869 v1 pith:KJHNBJGW submitted 2019-09-04 eess.AS cs.LGcs.SDstat.ML

Paul Primus , Hamid Eghbal-zadeh , David Eitelsebner , Khaled Koutini , Andreas Arzt , Gerhard Widmer This is my paper · ORCID

classification eess.AScs.LGcs.SDstat.ML

keywords domainadaptationacousticsceneclassificationrecordingdevicemismatchparallelaudiorecordingsinvariancemeansquarederrormaximumdiscrepancyDCASE2019Task1b

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

Distribution mismatch between the microphone used in training and the microphones used in the field degrades acoustic scene classifiers. This paper proposes to remove that mismatch by exploiting parallel recordings: time-aligned clips of the same scenes captured by different devices. The idea is to add a term to the training loss that penalizes the mean squared difference between the hidden-layer activations of each paired recording, forcing the network to build device-invariant representations. Because the parallel-target clips are used without class labels, the adaptation signal is cheap to collect. On the DCASE 2019 Task 1b acoustic scene data the method lifts accuracy on unseen target devices from 35.3% to 59.2%, outperforming MMD-based domain adaptation and a no-adaptation baseline.

What carries the argument

The load-bearing object is the pairwise MSE invariance loss of Eq. (1), $L_{l,\mathrm{MSE}} = \frac{1}{n d}\sum_i \|\varphi_l(x^s_i) - \varphi_l(x^t_i)\|_2^2$, which is added to the classification loss with weight $\lambda$ in Eq. (2). It treats a time-aligned source/target spectrogram pair as two views of the same scene and asks the chosen hidden layer $l$ to produce identical activations. The DA mini-batch size $n$ is reported as critical (larger $n$ helps), and best results came from applying the loss at the output layer. No class labels are used to compute the DA term, so the parallel recordings function as a self-supervised invariance signal.

What would settle it

Take a dataset of time-aligned pairs where the target device's response removes a class-discriminative cue -- for instance, a low-pass filter that deletes the only spectral difference between two otherwise identical scenes -- and train the paper's MSE-DA model under its reported grid. If accuracy on target devices falls to or below the 35.3% no-DA baseline or below the MMD-DA baseline, the method's reliance on equal class information in parallel pairs is refuted.

Watch

Extended reading notes

Core claim

The paper's central claim is that when time-aligned parallel recordings exist, pointwise matching of hidden activations is a better and more robust domain-adaptation objective than matching whole distributions with MMD. Concretely, with the classification loss fixed, minimizing Eq. (1) -- the mean squared error between activations $\varphi_l(x^s)$ and $\varphi_l(x^t)$ for paired source and target spectrograms -- at the network's output layer raises per-class balanced accuracy on target devices B and C from 35.3% (no DA) to 59.2%, versus 49.2% for MMD-DA. On the challenge evaluation set, the same model with MSE-DA reaches 73.4% accuracy versus 71.3% without DA. On the controlled two-moons dataset, MSE-DA reaches 99.9% target accuracy for every tested hyperparameter combination, while MMD-DA's best is 99.7% and its results vary with the settings.

Load-bearing premise

Parallel time-aligned recordings contain exactly the same class-relevant information and differ only because of the device, so forcing their hidden activations to be identical cannot throw away anything the classifier needs.

Editorial extensions

If this is right

The domain-adaptation loss needs no class labels for the parallel target recordings, so adapting to a new recording device only requires collecting a small set of synchronized clips.
Across every tested hyperparameter setting on the acoustic scenes task, MSE-DA improves over the no-DA baseline, while MMD-DA sometimes hurts; the paper also reports MSE-DA is more robust on the toy dataset.
Best results came from applying the DA loss at the output layer, indicating that high-level activations give the network the flexibility to absorb the device shift.
In challenge conditions with four-fold training and ensembling, the MSE-DA model still adds 2.1 percentage points on the evaluation set over the same architecture trained without DA.

Reading between the lines

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

The pairwise MSE constraint is a self-supervised alignment signal, so it could be relaxed to nearest-neighbor or softly aligned pairs, letting the method apply to casually recorded multi-device data without strict synchronization.
A direct diagnostic would measure how much class-relevant information is shared by each parallel pair before training; if pairs differ beyond device characteristics, the MSE constraint should be applied at lower layers or with a tolerance.
The same invariance idea transfers naturally to any synchronized sensor array, such as multi-microphone or multimodal recordings, where parallel observations arise without extra collection cost.
Because the paper reports that larger DA mini-batches help, one testable extension is to push n beyond 16 and check whether target accuracy continues to climb.

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.

Desk Editor's Note

read the letter

Quick take: this is a clean, small paper. The new thing is using MSE between hidden activations of time-aligned parallel recordings from different devices as a domain-adversarial-style loss, and showing it outperforms a same-architecture MMD baseline on both a toy problem and DCASE 2019 Task 1b. The result is credible and the paper is honest about where it doesn't win.

What it does well: it takes a standard objective (MSE between paired activations) and applies it in a setting where parallel recordings exist, which is exactly the DCASE 1b setup. No labels needed for the parallel data, which is a real practical advantage. The controlled comparison in Tables 1 and 3 is apples-to-apples: same ResNet, same training, only the DA loss differs. The toy dataset demonstrates the mechanism cleanly, and the real-data results show a large jump from 35.3% to 59.2% BC-accuracy with MSE-DA. Code is available. That is reproducible evidence.

Soft spots: all tables report single runs, no error bars or significance tests, which matters for small-N comparisons. Model selection uses the validation BC-accuracy to pick epochs and hyperparameters, so the reported numbers are optimistic. The MMD baseline is their own reimplementation, not the exact system from Eghbal-zadeh et al. [15], and the challenge evaluation actually shows [15]'s MMD-DA system at 74.5% vs. their 74.2% ensemble—slightly better. The stress-test note flags this as a contradiction. I don't think it's a contradiction; the paper explicitly explains the difference (architecture, mixup, ensemble, batch size). But the abstract and conclusion do overstate the comparison by not scoping it to the controlled setting. That's a presentation flaw, not a load-bearing one. The larger point is that the superiority of MSE over MMD is not universal; it holds for their architecture and setup. If a well-tuned MMD with all the tricks beats them, the 'more robust' claim needs a qualifier.

The assumption that parallel samples contain identical class-relevant information is stated clearly and is reasonable for time-aligned recordings of the same scene, but it's worth remembering. If devices have very different frequency responses that interact with class cues, forcing exact activation equality could hurt. Their toy experiment only tests one kind of covariate shift. Minor.

Bottom line: this is a legitimate empirical contribution. The central controlled comparison is solid, the method is simple and reproducible, and the paper is honest about the challenge results. It deserves a serious referee—mainly to make them add error bars and qualify the MMD comparison. I'd bring it to a reading group if we were discussing device-invariant audio, and I'd cite it if working on this specific problem, but it's not going to change how I think about domain adaptation generally.

Referee Report

3 major / 6 minor

Summary. The paper proposes a domain adaptation (DA) method for CNN-based acoustic scene classification that enforces device invariance by minimizing the mean squared error between hidden-layer activations of time-aligned parallel recordings from different devices. The DA loss (Eq. 1) is combined with the classification loss (Eq. 2) and requires no labels for the parallel samples. Experiments on a two-moons toy dataset and on the DCASE 2019 Task 1b dataset show that the proposed MSE-DA improves target-device accuracy over a no-DA baseline and over an MMD-DA baseline using the same architecture. The challenge submission achieves 73.4% single-model and 74.2% ensemble BC-accuracy on the evaluation set, and the paper claims 4th place in the team ranking. The central comparative claim is that MSE-DA is more accurate and more robust to DA hyperparameters than MMD-DA.

Significance. If the result holds, the contribution is practically valuable: it offers a simple, label-free, parameter-light way to exploit parallel recordings for device-invariant acoustic scene classification. The paper ships reproducible code, uses controlled same-architecture comparisons on two datasets, and reports results on a real challenge evaluation. The main comparative claim against MMD-DA is, however, only tested with a simplified MMD baseline within the same architecture; the fully engineered MMD-DA system from the same group scores slightly higher on the evaluation set (Table 4), so the scope of the superiority claim needs to be qualified. The method's core assumption of identical class-relevant information in parallel samples is also stated but not stress-tested.

major comments (3)

[§4.4, Table 4] The independent evaluation does not support the unqualified claim that MSE-DA outperforms MMD-DA: the authors' MSE-DA single model scores 73.4% and their ensemble 74.2%, while the MMD-DA system of [15] scores 74.5% on the same evaluation set. The attribution of this gap to architecture, across-device mixup, ensemble size, and batch size is reasonable, but it means the controlled MMD baseline in Tables 1 and 3 is a simplified variant, not the strongest MMD-DA configuration. The abstract and conclusion should either add a strong MMD-DA baseline with the same architecture and MixUp/larger batches, or explicitly restrict the comparative claim to the tested same-architecture MMD baseline, noting that a fully tuned MMD-DA system can match or exceed the proposed method.
[§4.3.4, Table 3; §4.2.2, Table 1] The central robustness claim rests on single-run grid searches. No error bars, repeated seeds, or significance tests are reported, and several neighboring configurations differ by only a few percentage points (e.g., Table 3, n=8, lambda=1: 0.592 vs n=16, lambda=1: 0.592; n=8, lambda=0.1: 0.537 vs lambda=1: 0.592). The paper should report mean and standard deviation over at least several seeds for the key configurations, or otherwise temper the statement that MSE-DA is 'more robust' to DA parameter choices.
[§3, Eq. (1)] The method assumes that parallel samples contain identical class-relevant information and differ only by covariate shift, so that forcing exact equality of hidden activations cannot discard discriminative information. This assumption is stated but not validated; a concrete test would be to evaluate sensitivity to small temporal misalignments or to compare Eq. (1) to a relaxed alignment loss. If the assumption fails, the MSE constraint could collapse useful representations, so the paper should at least discuss this limitation and ideally provide evidence on alignment robustness.

minor comments (6)

[§4.3.3] The phrase 'the 8.645 non-parallel device A samples' is inconsistent with the 9,185 training samples stated in §4.3.1; please reconcile the number and the decimal/thousands separator style.
[§4.4.1] The heading 'Datset' should be 'Dataset'.
[Eq. (1)] The notation with double vertical bars around the norm is redundant; use a single pair of vertical bars for the L2 norm.
[§4.2.1] The number '10.000 samples' should be written as '10,000 samples'.
[Table 4] The column headers 'Tr./Te.' and '4-CV' are not self-explanatory; add a caption sentence defining these validation splits.
[§4.3] The typo 'DACSE' should be 'DCASE'.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation; empirical evaluation on held-out devices, self-citations are non-load-bearing.

full rationale

The paper's claimed derivation is that minimizing the MSE between hidden activations of parallel source/target samples (Eq. 1), jointly with the classification loss (Eq. 2), improves target-device accuracy. Eq. (1) is defined directly from paired activations and contains no fitted parameter that is later relabeled as a prediction, and no term is defined in terms of the reported accuracies. All accuracy results (Tables 1, 3, and 4) are measured on held-out validation splits or on the external DCASE 2019 evaluation set, so the evaluation is not forced by construction. The self-citations to the architecture [19], the MMD baseline [15], and the technical report [21] are used as tools or comparators, not as evidence that the MSE loss works; indeed Table 4 reports that the MMD-DA system of [15] scores 74.5% versus 73.4% for the single MSE-DA model, which shows the comparison is not rigged in favor of the proposed method. The covariate-shift assumption about parallel recordings is an explicit modeling assumption, not an equation that reduces the result to its input. No uniqueness theorem, ansatz smuggled via citation, or renaming of a known result is present. The possible weakness that the simplified MMD baseline in Tables 1 and 3 may not match the full MMD system of [15] is a fairness/correctness concern, not circularity. Therefore no significant circularity is found.

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

The central contribution is an empirical training loss, so there are no invented physical or mathematical entities. The main free choices are the DA loss weight, DA batch size, DA layer, and MixUp parameters, all selected by validation or fixed by prior practice. The key domain assumption is that paired parallel recordings share identical class information, which is stated but not independently verified.

free parameters (4)

DA loss weight lambda = Grid: 0.1, 1, 5, 10 on toy; 0.1, 1, 10 on audio; best reported result uses lambda=1
Controls balance between classification and DA loss; selected by grid search on validation, so reported gains are conditional on this selection.
DA mini-batch size n = Grid: 8-256 on toy; 1, 8, 16 on audio; best audio result uses n=16
Number of parallel pairs per DA batch; the paper claims larger n helps in toy data but the audio results are less conclusive; selected on validation.
DA layer l = Output layer
The DA loss is applied to the output layer because the authors found it gave the best results in initial experiments (Section 4.1); this is a hand-chosen architectural choice.
MixUp alpha and beta = 0.2
Fixed regularization hyperparameters from [20], used for both classification and DA samples; not grid-searched in this paper.

assumptions (3)

domain assumption Time-aligned parallel recordings from different devices contain identical class-relevant information and differ only by covariate shift.
Introduced in Section 3 before Eq. (1) as the justification for forcing equal hidden activations; if false, MSE matching could erase class information.
domain assumption The target-domain label space is the same as the source-domain label space and no unseen classes appear on devices B and C.
Standard for the DCASE task and necessary for target accuracy to be a meaningful transfer metric.
domain assumption Standard deep learning training assumptions: the CNN, ReLU, BatchNorm, ADAM, and MixUp will generalize from validation to test if tuned on validation.
Common practice in empirical ML; not proven, and the paper provides no error bars to quantify validation-to-test stability.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Exploiting Parallel Audio Recordings to Enforce Device Invariance in CNN-based Acoustic Scene Classification." pith.science (2026). https://pith.science/paper/KJHNBJGW

@misc{pith2026190902869,
  author       = {Pith},
  title        = {Pith review of: Exploiting Parallel Audio Recordings to Enforce Device Invariance in CNN-based Acoustic Scene Classification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KJHNBJGW}},
  note         = {Machine review of arXiv:1909.02869}
}

read the original abstract

Distribution mismatches between the data seen at training and at application time remain a major challenge in all application areas of machine learning. We study this problem in the context of machine listening (Task 1b of the DCASE 2019 Challenge). We propose a novel approach to learn domain-invariant classifiers in an end-to-end fashion by enforcing equal hidden layer representations for domain-parallel samples, i.e. time-aligned recordings from different recording devices. No classification labels are needed for our domain adaptation (DA) method, which makes the data collection process cheaper.

Figures

Figures reproduced from arXiv: 1909.02869 by the authors.

**Figure 2.** From left to right: Time-aligned recordings from devices A (Soundman OKM II Klassik/Studio A3 Microphone & Zoom F8 [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

28 extracted references · 18 canonical work pages

[15]

A survey of transfer learning,

K. R. Weiss, T. M. Khoshgoftaar, and D. Wang, “A survey of transfer learning,” J. Big Data , vol. 3, p. 9, 2016. [Online]. Available: https://doi.org/10.1186/s40537-016-0043-6

doi:10.1186/s40537-016-0043-6 2016
[1]

INTRODUCTION Convolutional Neural Networks (CNNs) have become state of the art tools for audio related machine learning tasks, such as acous- tic scene classiﬁcation, audio tagging and sound event localization. While CNNs are known to generalize well if the recording condi- tions for training and unseen data remain the same, the generaliza- tion of this c...

work page 2019
[2]

Exploiting Parallel Audio Recordings to Enforce Device Invariance in CNN-based Acoustic Scene Classification

RELA TED WORK Domain Adaptation (DA) is a popular ﬁeld of research in transfer learning with multiple areas of application, e.g. bird audio detec- tion [2]. Kouw et al. [3] distinguish between three types of data shifts which lead to a domain mismatches: prior, covariate and con- cept shift. In this work we focus on domain mismatches which are caused by c...

work page Pith review arXiv 1909
[3]

This loss exploits the fact that parallel samples contain the same information relevant for classiﬁcation and differ only due to a covariate shift, e.g

DOMAIN-INV ARIANT LEARNING We propose a symmetric feature-based loss function to encourage the network to learn device-invariant representations for parallel samples from the source X s, and the target domain X t. This loss exploits the fact that parallel samples contain the same information relevant for classiﬁcation and differ only due to a covariate sh...

work page
[4]

We compare our proposed DA objective to the multi-kernel MMD- based approach used by Eghbal-zadeh et al

EXPERIMENTS In the following we evaluate the performance of our approach on the two moons dataset as well as on real-world acoustic data: the DCASE 2019 Task 1b dataset on acoustic scene classiﬁcation [14]. We compare our proposed DA objective to the multi-kernel MMD- based approach used by Eghbal-zadeh et al. [15] for DCASE 2019 Subtask 1b. In all experi...

work page 2019
[5]

Our DA loss is designed to enforce equal hidden layer representations for different devices by exploiting time-aligned recordings

CONCLUSION & FUTURE WORK In this report, we have shown how an already well-performing ResNet-like model [19] can be further improved for DCASE 2019 task 1b by using a simple DA technique. Our DA loss is designed to enforce equal hidden layer representations for different devices by exploiting time-aligned recordings. In our experiment we ﬁnd that pointwis...

work page 2019
[6]

Detection and Classiﬁcation of Acoustic Scenes and Events 2019 25–26 October 2019, New York, NY , USA

ACKNOWLEDGMENT This work has been supported by the COMET-K2 Center of the Linz Center of Mechatronics (LCM), funded by the Austrian federal government and the Federal State of Upper Austria. Detection and Classiﬁcation of Acoustic Scenes and Events 2019 25–26 October 2019, New York, NY , USA

work page 2019
[7]

Deep within- class covariance analysis for robust deep audio representation learning,

H. Eghbal-zadeh, M. Dorfer, and G. Widmer, “Deep within- class covariance analysis for robust deep audio representation learning,” in Neural Information Processing Systems, Inter- pretability and Robustness in Audio, Speech, and Language Workshop, 2018

work page 2018

Show all 28 references

[8]

Bird audio detection - dcase 2018,

F. Berger, W. Freillinger, P. Primus, and W. Reisinger, “Bird audio detection - dcase 2018,” DCASE2018 Challenge, Tech. Rep., September 2018

2018
[9]

An introduction to domain adaptation and transfer learning,

W. M. Kouw, “An introduction to domain adaptation and transfer learning,” CoRR, vol. abs/1812.11806, 2018. [Online]. Available: http://arxiv.org/abs/1812.11806

2018 arXiv
[10]

Wasserstein distance guided representation learning for domain adaptation,

J. Shen, Y . Qu, W. Zhang, and Y . Yu, “Wasserstein distance guided representation learning for domain adaptation,” inPro- ceedings of the Thirty-Second AAAI Conference on Artiﬁcial Intelligence, (AAAI-18), the 30th innovative Applications of Artiﬁcial Intelligence (IAAI-18), ...

2018
[11]

Correcting sample selection bias by unlabeled data,

J. Huang, A. J. Smola, A. Gretton, K. M. Borgwardt, and B. Sch¨olkopf, “Correcting sample selection bias by unlabeled data,” in Advances in Neural Information Processing Systems 19, Proceedings of the Twentieth Annual Conference on Neural Information Processing Systems, Vancou...

2006
[12]

A kernel two-sample test,

A. Gretton, K. M. Borgwardt, M. J. Rasch, B. Sch ¨olkopf, and A. J. Smola, “A kernel two-sample test,” J. Mach. Learn. Res., vol. 13, pp. 723–773, 2012. [Online]. Available: http://dl.acm.org/citation.cfm?id=2188410

2012
[13]

Beyond sharing weights for deep domain adaptation,

A. Rozantsev, M. Salzmann, and P. Fua, “Beyond sharing weights for deep domain adaptation,” IEEE Trans. Pattern Anal. Mach. Intell. , vol. 41, no. 4, pp. 801–814, 2019. [Online]. Available: https://doi.org/10.1109/TPAMI.2018. 2814042

2019 doi
[14]

Exploiting web images for event recognition in consumer videos: A multiple source domain adaptation approach,

L. Duan, D. Xu, and S.-F. Chang, “Exploiting web images for event recognition in consumer videos: A multiple source domain adaptation approach,” in 2012 IEEE Conference on Computer Vision and Pattern Recognition. IEEE, 2012, pp. 1338–1345

2012
[16]

Asymmetric and category invariant feature transformations for domain adaptation,

J. Hoffman, E. Rodner, J. Donahue, B. Kulis, and K. Saenko, “Asymmetric and category invariant feature transformations for domain adaptation,” International journal of computer vi- sion, vol. 109, no. 1-2, pp. 28–41, 2014

2014
[17]

Deep domain confusion: Maximizing for domain invariance,

E. Tzeng, J. Hoffman, N. Zhang, K. Saenko, and T. Darrell, “Deep domain confusion: Maximizing for domain invariance,” CoRR, vol. abs/1412.3474, 2014. [Online]. Available: http://arxiv.org/abs/1412.3474

2014 arXiv
[18]

Learning transferable features with deep adaptation networks,

M. Long, Y . Cao, J. Wang, and M. I. Jordan, “Learning transferable features with deep adaptation networks,” in Proceedings of the 32nd International Conference on Machine Learning, ICML 2015, Lille, France, 6-11 July 2015 , 2015, pp. 97–105. [Online]. Available: http: //proce...

2015
[19]

The Receptive Field as a Regularizer in Deep Convolutional Neural Networks for Acoustic Scene Classiﬁcation,

K. Koutini, H. Eghbal-zadeh, M. Dorfer, and G. Widmer, “The Receptive Field as a Regularizer in Deep Convolutional Neural Networks for Acoustic Scene Classiﬁcation,” in Pro- ceedings of the European Signal Processing Conference (EU- SIPCO), A Coru˜na, Spain, 2019

2019
[20]

Adversar- ial discriminative domain adaptation,

E. Tzeng, J. Hoffman, K. Saenko, and T. Darrell, “Adversar- ial discriminative domain adaptation,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recogni- tion, 2017, pp. 7167–7176

2017
[21]

A multi- device dataset for urban acoustic scene classiﬁcation,

A. Mesaros, T. Heittola, and T. Virtanen, “A multi- device dataset for urban acoustic scene classiﬁcation,” CoRR, vol. abs/1807.09840, 2018. [Online]. Available: http://arxiv.org/abs/1807.09840

2018 arXiv
[22]

Acoustic scene classiﬁcation and audio tagging with receptive-ﬁeld- regularized CNNs,

H. Eghbal-zadeh, K. Koutini, and G. Widmer, “Acoustic scene classiﬁcation and audio tagging with receptive-ﬁeld- regularized CNNs,” DCASE2019 Challenge, Tech. Rep., June 2019

2019
[23]

Delving deep into rectiﬁers: Surpassing human-level performance on imagenet classiﬁcation,

K. He, X. Zhang, S. Ren, and J. Sun, “Delving deep into rectiﬁers: Surpassing human-level performance on imagenet classiﬁcation,” in 2015 IEEE International Conference on Computer Vision, ICCV 2015, Santiago, Chile, December 7-13, 2015 , 2015, pp. 1026–1034. [Online]. Availabl...

2015 doi
[24]

Adam: A method for stochastic optimization,

D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” in 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings , 2015. [Online]. Available: http://arxiv.org/abs/1412.6980

2015 arXiv
[25]

Acoustic scene classiﬁcation with fully convolutional neural networks and I-vectors,

M. Dorfer, B. Lehner, H. Eghbal-zadeh, H. Christop, P. Fabian, and W. Gerhard, “Acoustic scene classiﬁcation with fully convolutional neural networks and I-vectors,” DCASE2018 Challenge, Tech. Rep., September 2018

2018
[27]

mixup: Beyond empirical risk minimization,

H. Zhang, M. Ciss ´e, Y . N. Dauphin, and D. Lopez-Paz, “mixup: Beyond empirical risk minimization,” CoRR, vol. abs/1710.09412, 2017. [Online]. Available: http: //arxiv.org/abs/1710.09412

2017 arXiv
[28]

Acoustic scene classiﬁcation with mismatched recording devices,

P. Primus and D. Eitelsebner, “Acoustic scene classiﬁcation with mismatched recording devices,” DCASE2019 Challenge, Tech. Rep., June 2019

2019
[608]

Available: http://papers.nips.cc/paper/3075- correcting-sample-selection-bias-by-unlabeled-data

[Online]. Available: http://papers.nips.cc/paper/3075- correcting-sample-selection-bias-by-unlabeled-data

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.

[1] [15]

A survey of transfer learning,

K. R. Weiss, T. M. Khoshgoftaar, and D. Wang, “A survey of transfer learning,” J. Big Data , vol. 3, p. 9, 2016. [Online]. Available: https://doi.org/10.1186/s40537-016-0043-6

doi:10.1186/s40537-016-0043-6 2016

[2] [1]

INTRODUCTION Convolutional Neural Networks (CNNs) have become state of the art tools for audio related machine learning tasks, such as acous- tic scene classiﬁcation, audio tagging and sound event localization. While CNNs are known to generalize well if the recording condi- tions for training and unseen data remain the same, the generaliza- tion of this c...

work page 2019

[3] [2]

Exploiting Parallel Audio Recordings to Enforce Device Invariance in CNN-based Acoustic Scene Classification

RELA TED WORK Domain Adaptation (DA) is a popular ﬁeld of research in transfer learning with multiple areas of application, e.g. bird audio detec- tion [2]. Kouw et al. [3] distinguish between three types of data shifts which lead to a domain mismatches: prior, covariate and con- cept shift. In this work we focus on domain mismatches which are caused by c...

work page Pith review arXiv 1909

[4] [3]

This loss exploits the fact that parallel samples contain the same information relevant for classiﬁcation and differ only due to a covariate shift, e.g

DOMAIN-INV ARIANT LEARNING We propose a symmetric feature-based loss function to encourage the network to learn device-invariant representations for parallel samples from the source X s, and the target domain X t. This loss exploits the fact that parallel samples contain the same information relevant for classiﬁcation and differ only due to a covariate sh...

work page

[5] [4]

We compare our proposed DA objective to the multi-kernel MMD- based approach used by Eghbal-zadeh et al

EXPERIMENTS In the following we evaluate the performance of our approach on the two moons dataset as well as on real-world acoustic data: the DCASE 2019 Task 1b dataset on acoustic scene classiﬁcation [14]. We compare our proposed DA objective to the multi-kernel MMD- based approach used by Eghbal-zadeh et al. [15] for DCASE 2019 Subtask 1b. In all experi...

work page 2019

[6] [5]

Our DA loss is designed to enforce equal hidden layer representations for different devices by exploiting time-aligned recordings

CONCLUSION & FUTURE WORK In this report, we have shown how an already well-performing ResNet-like model [19] can be further improved for DCASE 2019 task 1b by using a simple DA technique. Our DA loss is designed to enforce equal hidden layer representations for different devices by exploiting time-aligned recordings. In our experiment we ﬁnd that pointwis...

work page 2019

[7] [6]

Detection and Classiﬁcation of Acoustic Scenes and Events 2019 25–26 October 2019, New York, NY , USA

ACKNOWLEDGMENT This work has been supported by the COMET-K2 Center of the Linz Center of Mechatronics (LCM), funded by the Austrian federal government and the Federal State of Upper Austria. Detection and Classiﬁcation of Acoustic Scenes and Events 2019 25–26 October 2019, New York, NY , USA

work page 2019

[8] [7]

Deep within- class covariance analysis for robust deep audio representation learning,

H. Eghbal-zadeh, M. Dorfer, and G. Widmer, “Deep within- class covariance analysis for robust deep audio representation learning,” in Neural Information Processing Systems, Inter- pretability and Robustness in Audio, Speech, and Language Workshop, 2018

work page 2018