Pith. sign in

REVIEW 4 major objections 7 minor 22 references

Improving Acoustic Scene Classification in Low-Resource Conditions

T0 review · 4 major / 7 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read A compact acoustic scene classifier outperforms larger low-complexity models on unseen devices.

desk verdict A plausible incremental win in low-complexity ASC, but the ADIR threshold may be tuned on test-set statistics and the accuracy claims need error bars before I'd trust the exact numbers. read the letter →

arxiv 2412.20722 v2 pith:ZOCN42G7 submitted 2024-12-30 eess.AS cs.SD

Zhi Chen , Yun-Fei Shao , Yong Ma , Mingsheng Wei , Le Zhang , Wei-Qiang Zhang This is my paper · ORCID
classification eess.AScs.SD
keywords acousticsceneclassificationlow-resourceconditionsdepthwiseseparableconvolutionquantization-awaretrainingcross-devicegeneralizationknowledgedistillationresidualnormalizationdataaugmentation
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

This paper claims that a small, quantized neural network can classify acoustic scenes more accurately than larger low-complexity baselines, even when tested on recording devices never seen during training. The proposed DS-FlexiNet combines depthwise separable convolutions with residual connections, and adds three robustness techniques: energy-gated device impulse response augmentation (ADIR), frequency-domain style mixing (FMS), and knowledge distillation from twelve fused teacher models. On the TAU Urban Acoustic Scenes 2022 Mobile test set, the sm2 variant reports 58.25% average accuracy after int8 quantization-aware training, using only 30.69K parameters and 8.27M MACs, compared with 57.0% for CP-ResNet and 55.6% for BSConv-CNN. If the results hold, they suggest that device heterogeneity in acoustic scene classification does not force a choice between accuracy and deployment efficiency.

What carries the argument

The central object is the DS-FlexiNet block: a depthwise 3x3 convolution followed by a pointwise 1x1 convolution, wrapped in a ResNet-style residual connection and augmented with a Residual Normalization branch, ResNorm(x) = λ·x + IN(x), where IN is instance normalization along the channel dimension and λ is a learnable scalar. This block carries the efficiency-versus-accuracy balance. The cross-device robustness is carried by ADIR's energy threshold, FMS's frequency-statistics mixing, and the fused-teacher knowledge distillation loss, while quantization-aware training converts the model to int8 without, the authors report, losing its advantage over larger baselines.

What would settle it

Retrain sm2 with the same code but recompute the ADIR energy threshold from the energy distribution of only the training devices (A, B, C, S1-S3), or sweep E over a held-out set of unseen devices; if the S4-S6 advantage over CP-ResNet disappears or the optimal threshold shifts, the reported cross-device gain depends on a threshold tuned to the test distribution.

Watch

Extended reading notes

Core claim

The paper's central discovery is that a highly compact model can beat larger published low-complexity systems on cross-device acoustic scene classification. The authors attribute the gain to Residual Normalization applied along the channel dimension, which preserves device-specific characteristics that batch normalization would blur, and to ADIR, which convolves only high-energy audio with impulse responses while skipping low-energy samples to avoid distortion. They also show that fusing the logits of twelve teacher models with learned weights and biases gives a better distillation target than simply averaging teacher outputs. Together, these components lift accuracy on the unseen simulated devices S4-S6 and on real devices, while quantization-aware training keeps the model small enough for embedded deployment.

Load-bearing premise

The ADIR augmentation gates impulse-response convolution on an energy threshold E=323 chosen from the TAU22 dataset's energy distribution, and the load-bearing assumption is that this threshold transfers to unseen devices without being a artifact of the training or test distribution.

Editorial extensions

If this is right

  • A model small enough for embedded deployment can outperform larger low-complexity acoustic scene classifiers on unseen devices, so device mismatch is not an inevitable accuracy penalty.
  • Quantization-aware training preserves the cross-device advantage, since sm2 drops only from 61.42% to 58.25% after int8 conversion while still beating 57.0% CP-ResNet, making on-device deployment practical.
  • Residual Normalization along the channel dimension separates confusable scenes such as parks and street traffic better than frequency-wise normalization, improving robustness to background noise and recording conditions.
  • Fusing teacher logits with learned weights and biases yields a stronger distillation target than averaging teacher outputs, as shown by the sm4 model's 65.26% versus 63.48% on the full test set.
  • Energy-gated impulse response augmentation (ADIR) improves over unconditional impulse response augmentation by avoiding distortion of low-energy audio samples, contributing to gains on unseen devices.
  • The full recipe of depthwise separable convolutions, residual normalization, ADIR, FMS, and knowledge distillation is what the paper claims drives the performance of the proposed models.
  • The proposed model is intended for resource-constrained and embedded systems, where both accuracy and low computational cost are required.
  • The results suggest that the approach is suitable for real-world deployment in mobile and embedded acoustic scene classification.

Reading between the lines

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

  • The ADIR energy threshold E=323 is a dataset-derived constant; a natural test is whether it transfers to new recording conditions, since it could encode TAU22-specific energy statistics rather than a general property of environments.
  • The 12-teacher fusion makes the training pipeline computationally expensive even though the deployed model is small, so the approach is low-resource at inference time but not necessarily at training time.
  • If the results reproduce, the same recipe of depthwise blocks, residual normalization, energy-gated impulse response augmentation, frequency-style mixing, and distillation could transfer to other device-mismatch audio tasks such as speaker verification or bioacoustic monitoring.
  • A threshold sweep on held-out devices would clarify whether ADIR's benefit is robust or sensitive to the chosen energy cutoff, which is currently an open question the paper does not fully resolve.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 7 minor

Summary. This paper addresses low-complexity acoustic scene classification (ASC) on the TAU Urban Acoustic Scenes 2022 Mobile dataset. The authors propose DS-FlexiNet, a MobileNetV2/ResNet-style architecture with depthwise separable convolutions and residual connections, and combine it with channel-wise Residual Normalization (RN), Freq-MixStyle (FMS), Automatic Device Impulse Response (ADIR) augmentation, quantization-aware training (QAT), and knowledge distillation (KD) from an ensemble of twelve PaSST/CP-ResNet teachers. Four model sizes (sm1-sm4) are evaluated; the mid-size sm2 reports 61.42% accuracy before QAT and 58.25% after QAT at 30.69K parameters and 8.27M MACs, which the paper compares with CP-ResNet (57.0%) and BSConv-CNN (55.6%). Tables IV and V break down accuracy by device and attribute gains to teacher fusion, RN, and ADIR. The central claim is that a quantized, very low-complexity model can beat larger baselines on unseen devices S4-S6 while staying deployable on embedded hardware.

Significance. If the cross-device results are reproducible, this is a useful demonstration that a very small quantized model can approximately match or exceed larger full-precision baselines on TAU22, and the per-device tables give the community a concrete picture of where the gains sit, mainly on simulated devices including unseen S4-S6. The public benchmark, the explicit parameter/MAC reporting, and the combination of standard ingredients (QAT, KD, MixStyle-style augmentation, impulse-response augmentation) make the work easy to contextualize and compare against future submissions. The significance is currently bounded by the absence of statistical evidence, by a data-split ambiguity in the ADIR threshold, and by a mismatch between the proposed quantized model and the unquantized baselines in the headline comparison.

major comments (4)
  1. [Sec. II-C, Eq. (3), with Sec. III-A and Fig. 2] The ADIR energy threshold E_threshold=323 is load-bearing for the cross-device claim, but the paper does not state whether the energy histogram in Fig. 2 was built from the 139,620 training samples only or from the full development set. Since Table I shows that S4, S5, and S6 occur only in the test split, a histogram over the full dataset would make the augmentation policy conditional on test-device energy statistics, and the 58.25% test accuracy would no longer be a clean prospective evaluation on unseen devices. Please recompute the threshold and Fig. 2 on the training split alone, report per-device accuracies with that threshold, and include a sensitivity sweep around E_threshold, for example for the RN+FMS rows of Table V, where S4 and S6 show gains of about 1.3 and 1.7 points.
  2. [Sec. IV-A, Tables II and III] The proposed-model row in Table III lists ACC=58.25, which is the QAT accuracy of sm2 in Table II; the pre-QAT accuracy is 61.42. The comparison rows (Baseline, CP-ResNet, BSConv-CNN) are taken from prior technical reports and are not indicated to be quantized. With the current caption, the headline result that a quantized model beats full-precision baselines is not established; the precision of every row must be stated, and the comparison should include either the full-precision sm2 or QAT versions of the baselines, together with identical teacher/KD conditions where applicable.
  3. [Sec. IV-B and IV-E, Tables II, IV and V] All accuracy numbers appear to come from a single training run per configuration, and no random seeds, error bars, or significance tests are reported for the 250-epoch student training. Several of the claimed margins are small (for example, RN+FMS+ADIR vs RN+FMS+DIR improves ACC from 64.58 to 65.26, and fused vs averaged teachers improves 63.48 to 65.26), so runs with at least three to five seeds with mean and standard deviation are needed to show that the improvements are not seed noise.
  4. [Sec. II-D, Eqs. (4)-(5), and Sec. II-C] The selection of the teacher-fusion weights alpha_k, the bias terms beta_i, the loss weight lambda, the FMS probability/mixing coefficient, and the ADIR probability p_A is not described. If any of these hyperparameters were chosen by inspecting the test-device accuracies in Tables IV-V, the reported numbers would be optimistic. Please specify the validation split used for tuning these quantities and report sensitivity of the final accuracy to at least the most influential ones, such as lambda and p_A; if lambda is learned during training, say so explicitly and explain the update procedure.
minor comments (7)
  1. [Sec. II-B, Eq. (1)] The text says normalization is applied 'on the channel dimension,' but the statistics mu_nc and sigma^2_nc are indexed by instance and channel, which implies they are computed over frequency and time; please rephrase to avoid this ambiguity.
  2. [Sec. II-A and II-B] Residual Normalization is introduced as a 'custom' layer, but Eq. (2) is essentially the residual normalization described in the already-cited references [15] and [16]; the text should explicitly state the relation and the incremental contribution.
  3. [Sec. II-C, Eq. (3)] The sentence 'we set E to an average value of 323' uses E inconsistently with E_threshold, and the paper does not define what 'energy' means (e.g., mean squared amplitude) or give its units; please fix the notation and definition.
  4. [Fig. 2] The histogram lacks axis labels and does not show whether it was computed on training, validation, or test audio; please add this information, as it is directly tied to Major Comment 1.
  5. [References] Reference [17] is assigned the same arXiv identifier as reference [14] (1603.03417), which is the Texture Networks paper, not the impulse-response-augmentation paper; reference [8] should read arXiv:1801.04381 rather than 1801.0438.
  6. [Table V caption] The caption says ACC is the average validation-set accuracy over the last ten rounds, while the text describes the rows as test-set or unseen-device results; please clarify which split is used and whether the S4-S6 numbers come from the official test split.
  7. [Table IV] The row label 'Teacher(w/o)' is not defined; state whether this row is the student without KD, the teacher ensemble without fusion, or something else.

Circularity Check

1 steps flagged · score 3.0 of 10

The headline accuracy is an independent measurement against external baselines, but the ADIR energy threshold is taken from the same TAU22 data that contains the 'unseen' test devices, so the cross-device generalization claim is partially data-conditioned.

  1. fitted input called prediction [Section II.C.2 (Auto Device Impulse Response), Eq. (3); Section III.A (Dataset, Table I)]
    ""We analyzed energy distribution in the TAU22 dataset (see Fig. 2)... Based on energy distribution analysis, we set E to an average value of 323." The dataset section states: "devices S4, S5, and S6 are only present in the test set, meaning their specific characteristics are unseen during training.""

    Eq. (3) gates DIR convolution on E(x) > E_threshold, and E_threshold = 323 is fixed from an energy histogram of 'the TAU22 dataset.' By the paper's own Table I, the TAU22 test partition (29,680 samples) contains all S4-S6 audio, so unless the histogram was computed on the 139,620-sample training split alone (which the paper never states), the augmentation rule applied to S4-S6 waveforms is a deterministic function of statistics of the very recordings that are later scored as 'unseen devices' in Tables III and V. The reported ADIR gains on S4 and S6 therefore cannot be separated from the threshold's provenance, and no train-only threshold, per-device energy comparison, or sensitivity analysis is given to show that E=323 transfers to genuinely new recorders.

full rationale

The central claim of the paper is an empirical measurement: the quantized sm2 model reaches 58.25% average accuracy on the TAU Urban Acoustic Scenes 2022 Mobile test set at 30.69K parameters and 8.27M MACs, compared against external baselines CP-ResNet (57.0%) and BSConv-CNN (55.6%) in Table III. That benchmark result is self-contained and is not forced by any equation in the paper; no output of the model equals an input by construction. Residual Normalization (Eq. (2), ResNorm(x) = lambda*x + IN(x)) is a defined architectural module whose benefit is supported by the paper's own ablations (Table V, Fig. 3), not derived from its definition. The self-citation [15] (the authors' prior EURASIP paper) is used only to motivate channel-wise Instance Normalization; it is not load-bearing for the headline numbers, and the RN formula itself comes from third-party work [16]. The one load-bearing concern is the ADIR threshold: the paper fixes E_threshold=323 from an energy histogram described as covering 'the TAU22 dataset,' which, per the paper's own dataset description, includes the S4-S6 test devices. If the histogram pooled the full development set, the augmentation decision for test-device audio is conditioned on test-device energy statistics, making the 'unseen device' accuracy in Tables III and V not a clean prospective out-of-device result. This is more a data-provenance and reproducibility problem than a by-construction reduction, and because the headline accuracy is still an honest measurement against external baselines, a circularity score near the low end is appropriate. Secondary underspecification (not scored as a circular step) includes the teacher-fusion weights alpha_k and biases beta_i of Eq. (4), whose fitting split is never described; Table IV's fused-teacher advantage would be inflated if those weights were tuned on the same validation partition being reported.

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

No new physical entities, particles, forces, or conserved quantities are introduced. ADIR is a data-augmentation rule and Residual Normalization is a layer variant based on existing normalization methods.

free parameters (4)
  • ADIR energy threshold Ethreshold = 323
    Chosen from the TAU22 energy distribution in Fig. 2 to decide when to apply device impulse response convolution; no sensitivity analysis is reported.
  • FMS application probability and mixing coefficient = not reported
    Hyperparameters controlling how often and how strongly frequency statistics are mixed; values are not given, so reimplementation is ambiguous.
  • Teacher fusion weights alpha_k and biases beta_i = learned, values not reported
    Eq. (4) fuses 12 teacher logits with per-model weights and per-class biases; how these are fitted and whether they are validated on held-out data is not stated.
  • Loss-balancing weight lambda in Eq. (5) = updated during training, not reported
    Controls the trade-off between hard-label cross-entropy and distillation loss; no schedule or final value given.
assumptions (4)
  • domain assumption The energy threshold computed from the TAU22 distribution is representative of unseen devices S4-S6.
    ADIR applies or withholds DIR convolution based on E(x) <= 323; if the threshold is overfit to training energy statistics, cross-device gains may not transfer.
  • domain assumption Convolving audio with MicIRP device impulse responses simulates the effect of unseen recording devices.
    ADIR inherits this from prior DIR work [10],[17]; no validation that MicIRP statistics match TAU22 test devices.
  • domain assumption Channel-wise instance normalization preserves scene-discriminative information while removing device-specific statistics.
    The Residual Normalization layer is built on IN along the channel dimension; the paper shows confusion matrices but no causal evidence for the mechanism.
  • ad hoc to paper Twelve teachers trained independently on the same training set provide a soft target that improves student generalization.
    The number of teachers, their diversity, and fusion fitting are chosen for this paper; no analysis of teacher correlation or overfitting is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Improving Acoustic Scene Classification in Low-Resource Conditions." pith.science (2026). https://pith.science/paper/ZOCN42G7

@misc{pith2026241220722,
  author       = {Pith},
  title        = {Pith review of: Improving Acoustic Scene Classification in Low-Resource Conditions},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZOCN42G7}},
  note         = {Machine review of arXiv:2412.20722}
}
read the original abstract

Acoustic Scene Classification (ASC) identifies an environment based on an audio signal. This paper explores ASC in low-resource conditions and proposes a novel model, DS-FlexiNet, which combines depthwise separable convolutions from MobileNetV2 with ResNet-inspired residual connections for a balance of efficiency and accuracy. To address hardware limitations and device heterogeneity, DS-FlexiNet employs Quantization Aware Training (QAT) for model compression and data augmentation methods like Auto Device Impulse Response (ADIR) and Freq-MixStyle (FMS) to improve cross-device generalization. Knowledge Distillation (KD) from twelve teacher models further enhances performance on unseen devices. The architecture includes a custom Residual Normalization layer to handle domain differences across devices, and depthwise separable convolutions reduce computational overhead without sacrificing feature representation. Experimental results show that DS-FlexiNet excels in both adaptability and performance under resource-constrained conditions.

Figures

Figures reproduced from arXiv: 2412.20722 by the authors.

Figure 2
Figure 2. Energy Distribution Analysis of Audio in the TAU22 Dataset. [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. The confusion matrix classified by different dimensions: MS:metro [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

22 extracted references · 13 canonical work pages

  1. [15]

    Deep semantic learning for acoustic scene classification,

    Shao, Y .F., Ma, X.X., Ma, Y ., Zhang, W.Q, “Deep semantic learning for acoustic scene classification,” EURASIP J. Audio Speech Music. Pro- cess. vol. 2024, pp. 1 (2024). https://asmp-eurasipjournals.springeropen. com/articles/10.1186/s13636-023-00323-5

    work page doi:10.1186/s13636-023-00323-5 2024

  2. [16]

    QTI Submission to DCASE 2021: residual normalization for device-imbalanced acoustic scene classification with efficient design

    Kim, B., Yang, S., Kim, J., Chang, S, “QTI submission to DCASE 2021: Residual normalization for device-imbalanced acoustic scene clas- sification with efficient design,” ArXiv eprint: abs/2206.13909 (2022). doi:10.48550/arXiv.2206.13909

    work page Pith review arXiv doi:10.48550/arxiv.2206.13909 2022

  3. [1]

    Receptive field regu- larization techniques for audio classification and tagging with deep convolutional neural networks,

    Koutini, K., Eghbal-zadeh, H., Widmer, G, “Receptive field regu- larization techniques for audio classification and tagging with deep convolutional neural networks,” IEEE/ACM Transactions on Audio, Speech, and Language Processing, vol. 29, pp. 1987-2000. IEEE (2021). doi:10.1109/TASLP.2021.3082307

    work page arXiv 2021

  4. [2]

    CNNs-based Acoustic Scene Classification using Multi-Spectrogram Fusion and Label Expansions

    Zheng, W., Mo, Z., Xing, X., Zhao, G, “CNNs-based acoustic scene classification using multi-spectrogram fusion and label expansions,” ArXiv eprint: abs/1809.01543 (2018). doi:10.48550/arXiv.1809.01543

    work page Pith review arXiv doi:10.48550/arxiv.1809.01543 2018

  5. [3]

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

    Koutini, K., Eghbal-zadeh, H., Dorfer, M, “Widmer, G.: The recep- tive field as a regularizer in deep convolutional neural networks for acoustic scene classification,” ArXiv eprint: abs/1907.01803 (2019). doi:10.48550/arXiv.1907.01803

    work page Pith review arXiv doi:10.48550/arxiv.1907.01803 2019

  6. [4]

    Histogram of gradients of Time-Frequency Representations for Audio scene detection

    Rakotomamonjy, A., Gasso, G, “Histogram of gradients of time- frequency representations for audio scene detection,” ArXiv eprint: abs/1508.04909 (2015).doi:10.48550/arXiv.1508.04909

    work page Pith review arXiv doi:10.48550/arxiv.1508.04909 2015

  7. [5]

    Gammatone cepstral coefficients: Bi- ologically inspired features for non-speech audio classification,

    Valero, X., Al ´ıas-Pujol, F, “Gammatone cepstral coefficients: Bi- ologically inspired features for non-speech audio classification,” IEEE Transactions on Multimedia, vol. 14, pp. 1684-1689 (2012). doi:10.1109/TMM.2012.2199972

    arXiv 2012

  8. [6]

    Acoustic scene classification using block-based MFCC features,

    Ghodasara, V ., Waldekar, S., Paul, D., Saha, G, “Acoustic scene classification using block-based MFCC features,” ResearchGate eprint: abs/RG.2.2.18614.09287 (2016). doi:10.13140/RG.2.2.18614.09287

    work page arXiv 2016

Show all 22 references
  1. [7]

    Acoustic scene classification using deep residual networks with late fusion of separated high and low frequency paths,

    McDonnell, M.D., Gao, W, “Acoustic scene classification using deep residual networks with late fusion of separated high and low frequency paths,” ICASSP 2020 - 2020 IEEE International Conference on Acous- tics, Speech and Signal Processing (ICASSP), Barcelona, Spain (2020). do...

    2020

  2. [8]

    Mo- bileNetV2: Inverted residuals and linear bottlenecks,

    Sandler, M., Howard, A., Zhu, M., Zhmoginov, A., Chen, L.C, “Mo- bileNetV2: Inverted residuals and linear bottlenecks,” ArXiv eprint: abs/1801.0438 (2018). doi:10.48550/arXiv.1801.04381

    2018 doi

  3. [9]

    Deep Residual Learn- ing for Image Recognition,

    He, K., Zhang, X., Ren, S., Sun, J., “Deep Residual Learn- ing for Image Recognition,” ArXiv eprint: abs/1512.03385 (2015). doi:10.48550/arXiv.1512.03385

    2015 doi arXiv

  4. [10]

    Device-robust acoustic scene classification based on two-stage cat- egorization and data augmentation,

    Hu, Hu, Yang, Chao-Han Huck, Xia, Xianjun, Bai, Xue, Tan, et al, “Device-robust acoustic scene classification based on two-stage cat- egorization and data augmentation,” arXiv eprint/2007.08389 (2020). doi:10.48550/arXiv.2007.08389

    2020 doi arXiv

  5. [11]

    Deep convolutional neural networks and data augmentation for acoustic event detection,

    Takahashi, N., Gygli, M., Pfister, B., Van Gool, L, “Deep convolutional neural networks and data augmentation for acoustic event detection,” ArXiv eprint: abs/1604.07160. doi:10.48550/arXiv.1604.07160

    doi arXiv

  6. [12]

    Domain gener- alization with relaxed instance frequency-wise normalization for multi- device acoustic scene classification,

    Kim, B., Yang, S., Kim, J., Park, H., Lee, J., Chang, S, “Domain gener- alization with relaxed instance frequency-wise normalization for multi- device acoustic scene classification,” ArXiv eprint: abs/2206.12513 (2022). doi:10.48550/arXiv.2206.12513

    2022 doi arXiv

  7. [13]

    Domain general- ization with MixStyle,

    Zhou, K., Yang, Y ., Qiao, Y ., Xiang, T, “Domain general- ization with MixStyle,” ArXiv eprint: abs/2104.02008 (2021). doi:10.48550/arXiv.2104.02008

    2021 doi arXiv

  8. [14]

    Texture networks: Feed-forward synthesis of textures and stylized images,

    Ulyanov, D., Lebedev, V ., Vedaldi, A., Lempitsky, V , “Texture networks: Feed-forward synthesis of textures and stylized images,” ArXiv eprint: abs/1603.03417 (2016). doi:10.48550/arXiv.1603.03417

    2016 doi arXiv

  9. [17]

    Device- robust acoustic scene classification via impulse response augmentation,

    Morocutti, T., Schmid, F., Koutini, K., Widmer, G, “Device- robust acoustic scene classification via impulse response augmentation,” ArXiv eprint: abs/1603.03417 (2023). doi:10.23919/EUSIPCO58844.2023.10289983

    2023 arXiv

  10. [18]

    CP-JKU submission to Dcase23: Efficient acoustic scene classification with CP-Mobile, technical report,

    Schmid, F., Morocutti, T., Masoudian, S., Koutini, K., Widmer, G, “CP-JKU submission to Dcase23: Efficient acoustic scene classification with CP-Mobile, technical report,” (2023). https://api.semanticscholar. org/CorpusID:265222203

    2023

  11. [19]

    Acoustic scene classification in DCASE 2020 challenge: Generalization across devices and low complexity solutions,

    Heittola, T., Mesaros, A., Virtanen, T, “Acoustic scene classification in DCASE 2020 challenge: Generalization across devices and low complexity solutions,” (2020). https://api.semanticscholar.org/CorpusID: 219124055

    2020

  12. [20]

    Low-complexity acoustic scene classification in DCASE 2022 challenge,

    Mart ´ın-Morat´o, I., Paissan, F., Ancilotto, A., Heittola, T., Mesaros, A., Farella, E., Brutti, A., Virtanen, T, “Low-complexity acoustic scene classification in DCASE 2022 challenge,” ArXiv eprint: abs/2206.03835 (2022). doi:10.48550/arXiv.2206.03835

    2022 doi arXiv

  13. [21]

    Low-complexity deep learning solution for acoustic scene classifi- cation,

    Cai, W., Zhang, M., Zhang, X.:Tencent Submission to DCASE 23 Task 1, “Low-complexity deep learning solution for acoustic scene classifi- cation,” Tech. rep., Detection and classification of acoustic scenes and events 2023 (2023). https://dcase.community/documents/challenge2023...

    2023

  14. [22]

    Low-complexity Acoustic Scene Classification Us- ing Blueprint Separable Convolution and Knowledge Distillation,

    Tan, J., Li, Y , “Low-complexity Acoustic Scene Classification Us- ing Blueprint Separable Convolution and Knowledge Distillation,” Tech. Rep., Detection and Classification of Acoustic Scenes and Events 2023 (2023). https://dcase.community/documents/challenge2023/ technical re...

    2023

Pith tools

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