Pith. sign in

REVIEW 5 major objections 8 minor 63 references

Should Audio Front-ends be Adaptive? Comparing Learnable and Adaptive Front-ends

T0 review · 5 major / 8 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read Adaptive audio front-ends beat fixed learnable ones on eight audio tasks.

desk verdict Solid clean-data comparison and a useful simplification, but the robustness claim rests on a misinterpreted learning-curve plot. read the letter →

arxiv 2502.03260 v1 pith:W7X4RUVC submitted 2025-02-05 eess.AS cs.SD

Qiquan Zhang , Buddhi Wickramasinghe , Eliathamby Ambikairajah , Vidhyasaharan Sethu , Haizhou Li This is my paper · ORCID
classification eess.AScs.SD
keywords RepresentationlearningAudiofront-endAdaptiveinferenceGaborfiltersLearnablefront-endsclassificationNeuralfeedbackcontrollerQ-factor
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 asks whether the front-end of an audio classifier—the stage that converts raw waveforms into features—should adapt to each input at inference time instead of staying fixed after training. To answer it, the authors compare Ada-FE, a front-end whose Gabor filters are reshaped frame by frame by a neural feedback controller, with LEAF and TD-fbanks on eight speech, sound-event, and music benchmarks using two back-end classifiers. They report that the adaptive front-ends are more accurate than the fixed learnable front-ends on every task and with both back-ends, in some cases by large margins, and that their test accuracy is more stable across training epochs. A simplified version that drops the hand-crafted level-dependent adaptation matches or exceeds the full model, suggesting the neural controller can learn the needed level-dependent behaviour on its own. The paper reads this as evidence that input-adaptive front-ends are a better default than fixed learnable front-ends across audio classification tasks.

What carries the argument

The central object is an adaptive Gabor filter layer whose only tunable shape parameter is the Q-factor, defined as the ratio of center frequency to bandwidth, $Q = f_c / BW$. The filter's frequency response at the center frequency satisfies $|W(\Omega_c)| \approx \sqrt{2\pi}\, Q / \Omega_c$, so raising $Q$ narrows the filter and raises its gain, while lowering $Q$ widens it and compresses the input. Each frame's Q value is the sum of two parallel terms: $Q_E$ from a hand-crafted level-dependent adaptation that reads subband energy (the amplitude-modulation pathway) and $Q_{FM}$ from a two-layer fully-connected neural adaptive feedback controller fed with frequency-modulation features. The entire system is trained end-to-end with a back-end classifier, with the controller's gradient path running counter to the main network path. A simplified version, Ada-FE-S, removes the hand-crafted $Q_E$ term so the neural controller alone determines the adaptive Q value.

What would settle it

Corrupt the test portions of ESC-50, Speech Commands V2, and VoxCeleb1 with additive noise, reverberation, or band-limiting at several signal-to-noise ratios and compare Ada-FE against LEAF; if LEAF matches or beats Ada-FE under those acoustic degradations, the claim that adaptive inference helps in varying acoustic environments would fail even if the clean-data accuracy advantage stands.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that adapting filter shape during inference, not just learning it during training, improves audio classification. Ada-FE operates on raw waveforms with a fixed Gabor filter layer followed by an adaptive Gabor filter layer; the adaptive layer's Q-factor, defined as the ratio of center frequency to bandwidth, is updated every frame as the sum of a level-dependent term computed from subband energy by a hand-crafted piecewise function and a term from a neural adaptive feedback controller fed with frame-averaged frequency-modulation features. Because the Q-factor directly sets gain and bandwidth through $|W(\Omega_c)| \approx \sqrt{2\pi}\, Q/\Omega_c$, the controller can make a filter more selective or more compressive as the input changes. Across ESC-50, GTZAN, FMA-S, CREMA-D, IEMOCAP, Speech Commands V1/V2, and VoxCeleb1, Ada-FE and its simplified variant Ada-FE-S-FM outperform LEAF on Top-1 accuracy with both EfficientNet-B0 and MobileNetV2-100 back-ends, with relative gains up to about 34% on VoxCeleb1 with 1-second segments. The paper also reports that removing the hand-crafted level-dependent adaptation does not hurt, and that removing the fixed first filter layer does hurt, so the fixed stage remains structurally important. The conclusion is that a feedback-controlled, input-adaptive front-end offers advantages over a fixed learnable front-end.

Load-bearing premise

The load-bearing premise is that the flatness of test-accuracy learning curves over training epochs is evidence of robustness to varying acoustic environments, even though all evaluations use clean recordings with no added noise, reverberation, or channel distortion at test time.

Editorial extensions

If this is right

  • Ada-FE and Ada-FE-S-FM improve Top-1 accuracy over LEAF on all eight datasets with both EfficientNet-B0 and MobileNetV2-100 back-ends.
  • The simplified adaptive front-end Ada-FE-S-FM, which removes the hand-crafted level-dependent adaptation, matches or exceeds the full Ada-FE on most tasks.
  • The fixed first Gabor filter layer is still needed: removing it degrades accuracy substantially on every benchmark.
  • Adaptive front-ends reach high accuracy with far fewer training epochs than LEAF, and their test accuracy remains stable across epochs, so early stopping is less risky.

Reading between the lines

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

  • Because the paper only tests clean recordings, the reported accuracy advantage over LEAF is a lower bound on the benefit of adaptivity: the strongest gains would be expected under mismatched acoustic conditions such as noise or reverberation, which a straightforward test-set corruption experiment could measure.
  • The frame-wise Q-factor trajectories Ada-FE produces encode a per-frame estimate of the acoustic environment; exposing these trajectories to the back-end, or using them as an auxiliary training signal, is a testable extension the paper does not explore.
  • Plugging Ada-FE or Ada-FE-S into self-supervised pre-training pipelines is a natural next step, and the paper itself names pre-training as future work; a failure to see gains there would bound the value of adaptivity to low-level feature extraction.
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

5 major / 8 minor

Summary. This paper asks whether audio front-ends should be input-adaptive, and evaluates Ada-FE, a Gabor-filter front-end with a neural feedback controller that adjusts Q-factors frame-wise, against LEAF and TD-fbanks on eight audio benchmarks with two back-ends. It also introduces Ada-FE-S, a simplified version with the hand-crafted LDA module removed, and ablates the controller input choices. The main empirical claims are that Ada-FE and Ada-FE-S-FM outperform LEAF and TD-fbanks across all tasks, and that Ada-FE exhibits better stability or robustness on test samples over training epochs.

Significance. If the accuracy advantage is reproducible, the paper provides useful evidence that input-adaptive front-ends can outperform fixed learnable front-ends, and the 'w/o Adaptation' ablation is a valuable control. The evaluation across two backbones and eight tasks, the use of the official LEAF implementation, and the repeated-runs protocol are strengths. However, the central robustness claim is not supported by the clean-data experiments, and several comparison details (TD-fbanks coverage, error bars, baseline specification) need to be tightened before the headline claims can be accepted.

major comments (5)
  1. [Section V.C, Section VI, Abstract] The stability/robustness claim is not supported by the reported experiments. Figures 11-14 show test-accuracy learning curves on clean recordings, and no experiment applies noise, reverberation, channel distortion, or a domain shift at test time. The statement in Section V.C that 'The learning curves of LEAF suggest that LEAF is quite sensitive to varying acoustic conditions' is an inference from optimization dynamics on matched clean data, not from any controlled acoustic variation. Because the abstract's 'stability or robustness' claim and the conclusion's 'respond to varying conditions' rest on this inference, the adaptive-robustness claim is under-supported. Either add experiments with acoustic degradations or revise the claim to 'stable test accuracy across epochs'.
  2. [Section V.C, Table III] The claim that Ada-FE and Ada-FE-S-FM 'outperform LEAF and TD-fbanks, across all the tasks' is not supported by Table III. TD-fbanks results are reported only for SPC-V2 (87.70) and VoxCeleb1 (26.0) under EfficientNet-B0, and these are taken from the original LEAF paper [19]; no TD-fbanks numbers are given for ESC-50, GTZAN, FMA-S, CREMA-D, IEMOCAP, SPC-V1, or for the MobileNetV2 backbone. Please either re-run TD-fbanks under the same protocol on all benchmarks or restrict the superiority claim to LEAF.
  3. [Section IV.B, Section V.C] The paper reports means over three runs but no error bars, confidence intervals, or significance tests are provided. Many head-to-head differences are small (e.g., Ada-FE vs Ada-FE-S-FM on SPC-V1 with EfficientNet-B0: 91.29 vs 91.14 in Table II), so the reader cannot tell whether the reported ordering is reliable. Please report per-run variation or statistical significance for the main comparisons, at least for the Ada-FE vs LEAF differences.
  4. [Section VI, Section V.B] The conclusion that removing LDA demonstrates 'hand-crafted LDA is quite essential' contradicts the body of the paper. In Section V.B and Table II, Ada-FE-S-FM (without LDA) achieves comparable or better accuracy than Ada-FE on most benchmarks, and the text states that the Q-factor adaptive learning 'benefits from removing the hand-crafted module.' The conclusion should be corrected.
  5. [Table II, Section V.B] The 'w/o Adaptation (Not Learnable)' baseline is the key control for the adaptivity claim, but its configuration is not specified in the text. It is unclear whether the Gabor filters are fixed at their initial Q values, at some learned static Q values, or with only the fixed Gabor layer active. Please define the exact front-end used in this condition so that the adaptivity benefit can be properly interpreted.
minor comments (8)
  1. [Table III caption] The word 'netwroks' should be 'networks'.
  2. [Table IV header] The model name 'Ada-FE-S-F' should be 'Ada-FE-S-FM'.
  3. [Throughout] The dataset name 'V oxCeleb1' should be 'VoxCeleb1' without the stray space.
  4. [Figure 12 caption] The word 'MobilenNetV2' should be 'MobileNetV2'.
  5. [Section II] The reference to 'Jailty and Hinton' should be 'Jaitly and Hinton'.
  6. [Section II] The term 'Gammaton-filters' should be 'Gammatone-filters'.
  7. [Section IV.A] The phrase 'repost the accuracy' should be 'report the accuracy'.
  8. [Section IV.B] The typo 'TD-fbansks' should be 'TD-fbanks'.

Circularity Check

1 steps flagged · score 2.0 of 10

Minor self-definitional Q-energy illustration; headline accuracy comparisons are independent and non-circular.

  1. self definitional [Section V-A (Q-Value Analysis, Figures 7-9)]
    "Given a frame with a low energy value, we expect Ada-FE to dynamically produce a corresponding high Q-factor value, leading to a high gain (sensitivity) and a low bandwidth (selectivity), and conversely, expect a low Q value, which is consistent with those shown in Figures 7-9. These observations about the relationship between energy and Q values over time illustrate how the proposed model adaptively responds to time-varying signal characteristics."

    The anti-correlation between energy and Q is built into the model by construction. Section III-B defines the level-dependent adaptation (LDA) module so that 'the Q value QE{E} is low for a high input level E' and 'QE is high for a low input level,' with the final Q given by Q = QE + QFM. Since LDA is a hand-crafted piecewise function of subband energy, observing that the overall Q curve opposes the energy curve is a restatement of the LDA definition rather than an emergent empirical discovery. The passage presents this definitional relationship as evidence that Ada-FE 'adaptively responds to time-varying signal characteristics,' which is a self-definitional illustration.

full rationale

The paper's primary empirical claim, that Ada-FE and Ada-FE-S-FM outperform LEAF and TD-fbanks across eight datasets and two back-ends, rests on experiments run with the official LEAF implementation and public benchmark datasets. No parameter is fitted to the target result, and no prediction is statistically forced by the experimental design. The numerous self-citations ([26], [44], [45], [49]) describe components of the authors' own system, but the comparative evaluation is new and externally grounded, so those citations are not load-bearing in a circular sense. The one genuine circular step is the Q-value analysis in Section V-A, where the observed inverse relationship between energy and Q is guaranteed by the hand-crafted LDA mapping defined earlier in Section III-B; presenting it as evidence of adaptive behavior is self-definitional, although it is ancillary to the headline results. Separately, the paper's robustness claim, which infers insensitivity to varying acoustic conditions from flat test-accuracy learning curves on clean recordings, is an unsupported inference and a validity concern, but it is not a circular reduction of the kind counted here. Overall, the central derivation chain is independent, with only a minor self-definitional illustration, warranting a score of 2.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The empirical claims rest on standard ML training assumptions and on design choices inherited from prior work. No mathematical derivation is attempted, so the main assumptions are domain-level (benchmark representativeness, training stability, comparability of baseline numbers).

free parameters (6)
  • Number of Gabor filters N = 40
    Hand-chosen architecture hyperparameter for the fixed Gabor filter layer; determines the number of adaptive filters (N-1=39).
  • Gabor filter length L = 150 samples
    Hand-chosen filter length in samples for both Gabor layers.
  • Frame length = 11 ms
    Hand-chosen segment length for frame-wise adaptation; affects temporal resolution of Q-factor updates.
  • Spatial differentiation order k = 1
    Hand-chosen differentiation order; determines number of output channels N-1.
  • Number of octaves for centroid magnitude = 5
    Hand-chosen number of octaves used to compute CM features per subband.
  • AFC hidden layer configuration = Not fully specified
    The AFC is described as a two-layer FC of size N-k, but exact hidden dimensions are not stated.
assumptions (4)
  • domain assumption Gabor filters parameterized by Q-factor can model auditory filter shape changes
    The paper assumes that adjusting Q-factor of Gabor filters emulates the cochlea's active mechanism; cited psychoacoustic evidence is indirect.
  • domain assumption Benchmark datasets are representative of varying acoustic environments
    The paper claims robustness to acoustic conditions based on results on eight clean datasets, which are not explicitly noise or channel stressed.
  • domain assumption Split-gradient backpropagation through the feedback controller yields a valid training signal
    The paper states that the two backprop paths allow AFC training without labeled Q-values, but provides no convergence or stability analysis.
  • domain assumption Reported TD-fbanks numbers from prior work are directly comparable
    The paper uses TD-fbanks results from [19] rather than rerunning under identical conditions; assumes comparability across papers.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Should Audio Front-ends be Adaptive? Comparing Learnable and Adaptive Front-ends." pith.science (2026). https://pith.science/paper/W7X4RUVC

@misc{pith2026250203260,
  author       = {Pith},
  title        = {Pith review of: Should Audio Front-ends be Adaptive? Comparing Learnable and Adaptive Front-ends},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/W7X4RUVC}},
  note         = {Machine review of arXiv:2502.03260}
}
read the original abstract

Hand-crafted features, such as Mel-filterbanks, have traditionally been the choice for many audio processing applications. Recently, there has been a growing interest in learnable front-ends that extract representations directly from the raw audio waveform. \textcolor{black}{However, both hand-crafted filterbanks and current learnable front-ends lead to fixed computation graphs at inference time, failing to dynamically adapt to varying acoustic environments, a key feature of human auditory systems.} To this end, we explore the question of whether audio front-ends should be adaptive by comparing the Ada-FE front-end (a recently developed adaptive front-end that employs a neural adaptive feedback controller to dynamically adjust the Q-factors of its spectral decomposition filters) to established learnable front-ends. Specifically, we systematically investigate learnable front-ends and Ada-FE across two commonly used back-end backbones and a wide range of audio benchmarks including speech, sound event, and music. The comprehensive results show that our Ada-FE outperforms advanced learnable front-ends, and more importantly, it exhibits impressive stability or robustness on test samples over various training epochs.

Figures

Figures reproduced from arXiv: 2502.03260 by the authors.

Figure 1
Figure 1. Illustration of (a) fixed, (b) learnable but non-adaptive, and (c) learnable and adaptive audio front-ends. In real-world acoustic scenarios, speech and audio signals are inevitably shaped by varying acoustic conditions during transmission. The input speech level, for instance, varies with the distance between the speakers and the microphone and the voice levels of the speakers. (a) Fixed (hand-crafted) front-ends e… view at source ↗
Figure 2
Figure 2. Illustration of audio/speech representation methods, which mainly include Fixed, hand-crafted features (e.g., Mel- and [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Illustrations of (a) the overall diagram of the adaptive front-end (Ada-FE) [26] and (b) the simplified adaptive front￾end (Ada-FE-S), where the hand-crafted level-dependent adaptation function module is removed and the adaptive Q value is completely controlled by the neural adaptive feedback controller (orange box). respectively: Q = Q E + Q FM. The LDA module employs a hand-crafted piecewise function to calculate … view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: Illustrations of the time impulse and frequency re [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Illustration of the LDA module. Level-Dependent Adaptation. In our previous study [44], we propose a feed-forward adaptive front-end consisting of two filterbanks for spoofed speech detection, where an LDA module is employed to provide variable selective gain via Q val…
Figure 6
Figure 6. Figure 6: Illustration of the AFC module. QFM is added with the Q value QE that depends on input energy (relating to amplitude modulation (AM)) to obtain the final Q value to update the Gabor filters. This is motivated by evidence in psychoacoustic research, which shows that the…
Figure 7
Figure 7. Figure 7: Illustration of how (a) energy and (b) Q value change frame by frame for the Gabor filter with a center frequency of 1.12 kHz (low). Ada-FE infers one 1-second speech segment on the speaker ID task (VoxCeleb1). 0 20 40 60 80 Frame Index 0.5 1.0 1.5 Energy 1e 3 (a) The …
Figure 10
Figure 10. Figure 10: The illustrations of the AFC with (a) the energy of each subband (AFC-Energy) and (b) the concatenation of the energy and FM as the input (AFC-Energy+FM), respectively. L represents the concatenation operation. B. Analyses of Simplified Ada-FE In this section, we comp…
Figure 9
Figure 9. Figure 9: Illustration of how (a) energy and (b) Q value change frame by frame for the Gabor filter with a center frequency of 5.81 kHz (high). Ada-FE infers one 1-second speech segment on the speaker ID task (VoxCeleb1). In Figures 7-9, we illustrate three examples to visualize…
Figure 11
Figure 11. Figure 11: Top-1 accuracy (%) of the LEAF, Ada-FE, and [PITH_FULL_IMAGE:figures/full_fig_p010_11.png]
Figure 12
Figure 12. Figure 12: Top-1 accuracy (%) of the LEAF, Ada-FE, and [PITH_FULL_IMAGE:figures/full_fig_p011_12.png]
Figure 13
Figure 13. Figure 13: Top-1 accuracy (%) of the LEAF, Ada-FE, and [PITH_FULL_IMAGE:figures/full_fig_p011_13.png]
Figure 14
Figure 14. Figure 14: Top-1 accuracy (%) of the LEAF, Ada-FE, and [PITH_FULL_IMAGE:figures/full_fig_p011_14.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

63 extracted references · 51 canonical work pages

  1. [19]

    LEAF: A Learnable Frontend for Audio Classification,

    N. Zeghidour, O. Teboul, F. de Chaumont Quitry, and M. Tagliasacchi, “LEAF: A Learnable Frontend for Audio Classification,” in Proc. ICLR, 2022

    work page 2022

  2. [1]

    Learning filterbanks from raw speech for phone recogni- tion,

    N. Zeghidour, N. Usunier, I. Kokkinos, T. Schaiz, G. Synnaeve, and E. Dupoux, “Learning filterbanks from raw speech for phone recogni- tion,” in Proc. IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , 2018, pp. 5509–5513

    work page 2018

  3. [2]

    Acoustic modelling with cd-ctc-smbr lstm rnns,

    A. Senior, H. Sak, F. de Chaumont Quitry, T. Sainath, and K. Rao, “Acoustic modelling with cd-ctc-smbr lstm rnns,” in Proc. IEEE Work- shop on Automatic Speech Recognition and Understanding (ASRU) , 2015, pp. 604–609

    work page 2015

  4. [3]

    Speaker recognition from raw waveform with sincnet,

    M. Ravanelli and Y . Bengio, “Speaker recognition from raw waveform with sincnet,” in Proc. IEEE Spoken Language Technology Workshop (SLT), 2018, pp. 1021–1028

    work page 2018

  5. [4]

    End-to-end spoofing detection with raw waveform CLDNNS,

    H. Dinkel, N. Chen, Y . Qian, and K. Yu, “End-to-end spoofing detection with raw waveform CLDNNS,” in Proc. IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , 2017, pp. 4860– 4864

    work page 2017

  6. [5]

    ESC: Dataset for environmental sound classification,

    K. J. Piczak, “ESC: Dataset for environmental sound classification,” in Proceedings of ACM international conference on Multimedia , 2015, pp. 1015–1018

    work page 2015

  7. [6]

    Towards learning universal audio representations,

    L. Wang, P. Luc, Y . Wu, A. Recasens, L. Smaira, A. Brock, A. Jaegle, J.- B. Alayrac, S. Dieleman, J. Carreira et al., “Towards learning universal audio representations,” in Proc. IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , 2022, pp. 4593– 4597

    work page 2022

  8. [7]

    Filterbank design for end-to-end speech separation,

    M. Pariente, S. Cornell, A. Deleforge, and E. Vincent, “Filterbank design for end-to-end speech separation,” in Proc. IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2020, pp. 6364–6368

    work page 2020

Show all 63 references
  1. [8]

    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 et al., “Conformer: Convolution-augmented transformer for speech recognition,” Proc. INTERSPEECH 2020 , pp. 5036–5040, 2020. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2019 12

    2020

  2. [9]

    Deep scattering spectrum,

    J. And ´en and S. Mallat, “Deep scattering spectrum,” IEEE Transactions on Signal Processing , vol. 62, no. 16, pp. 4114–4128, 2014

    2014

  3. [10]

    Pitch-Adaptive Front-End Features for Robust Children’s ASR

    S. Shahnawazuddin, A. Dey, and R. Sinha, “Pitch-Adaptive Front-End Features for Robust Children’s ASR.” in Proc. INTERSPEECH, 2016, pp. 3459–3463

    2016

  4. [11]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in Proc. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016, pp. 770–778

    2016

  5. [12]

    Conv-tasnet: Surpassing ideal time– frequency magnitude masking for speech separation,

    Y . Luo and N. Mesgarani, “Conv-tasnet: Surpassing ideal time– frequency magnitude masking for speech separation,” IEEE/ACM Trans- actions on Audio, Speech, and Language processing , vol. 27, no. 8, pp. 1256–1266, 2019

    2019

  6. [13]

    Real time speech enhancement in the waveform domain,

    A. Defossez, G. Synnaeve, and Y . Adi, “Real time speech enhancement in the waveform domain,” in Proc. INTERSPEECH , 2020, pp. 3291– 3295

    2020

  7. [14]

    Learning the speech front-end with raw waveform cldnns,

    T. N. Sainath, R. J. Weiss, A. W. Senior, K. W. Wilson, and O. Vinyals, “Learning the speech front-end with raw waveform cldnns,” in Proc. INTERSPEECH, 2015, pp. 1–5

    2015

  8. [16]

    wav2vec: Unsupervised pre-training for speech recognition,

    S. Schneider, A. Baevski, R. Collobert, and M. Auli, “wav2vec: Unsupervised pre-training for speech recognition,” arXiv preprint arXiv:1904.05862, 2019

    1904 arXiv

  9. [17]

    A deep neural network integrated with filterbank learning for speech recognition,

    H. Seki, K. Yamamoto, and S. Nakagawa, “A deep neural network integrated with filterbank learning for speech recognition,” inProc. IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2017, pp. 5480–5484

    2017

  10. [18]

    Learning the speech front-end with raw waveform CLDNNs,

    T. N. Sainath, R. J. Weiss, A. Senior, K. W. Wilson, and O. Vinyals, “Learning the speech front-end with raw waveform CLDNNs,” in Proc. INTERSPEECH, 2015, pp. 1–5

    2015

  11. [20]

    Train- able frontend for robust and far-field keyword spotting,

    Y . Wang, P. Getreuer, T. Hughes, R. F. Lyon, and R. A. Saurous, “Train- able frontend for robust and far-field keyword spotting,” in Proc. IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2017, pp. 5670–5674

    2017

  12. [21]

    Automatic gain control in cochlear mechanics,

    R. F. Lyon, “Automatic gain control in cochlear mechanics,” The mechanics and biophysics of hearing , vol. 87, pp. 395–402, 1990

    1990

  13. [22]

    The cochlea as a smart structure,

    S. J. Elliott and C. A. Shera, “The cochlea as a smart structure,” Smart Materials and Structures , vol. 21, no. 6, p. 064001, 2012

    2012

  14. [23]

    Integrating the active process of hair cells with cochlear function,

    A. Hudspeth, “Integrating the active process of hair cells with cochlear function,” Nature Reviews Neuroscience , vol. 15, no. 9, pp. 600–614, 2014

    2014

  15. [24]

    Auditory processing of speech signals for robust speech recognition in real-world noisy environments,

    D.-S. Kim, S.-Y . Lee, and R. Kil, “Auditory processing of speech signals for robust speech recognition in real-world noisy environments,” IEEE Transactions on Speech and Audio Processing , vol. 7, no. 1, pp. 55–69, 1999

    1999

  16. [25]

    Power-normalized cepstral coefficients (pncc) for robust speech recognition,

    C. Kim and R. M. Stern, “Power-normalized cepstral coefficients (pncc) for robust speech recognition,” IEEE/ACM Transactions on Audio, Speech, and Language Processing, vol. 24, no. 7, pp. 1315–1329, 2016

    2016

  17. [26]

    Dnn controlled adaptive front-end for replay attack detection systems,

    B. Wickramasinghe, E. Ambikairajah, V . Sethu, J. Epps, H. Li, and T. Dang, “Dnn controlled adaptive front-end for replay attack detection systems,” Speech Communication, p. 102973, 2023

    2023

  18. [27]

    Towards learning a universal non- semantic representation of speech,

    J. Shor, A. Jansen, O. Lang, O. Tuval, F. de Chaumont Quitry, M. Tagliasacchi, and D. Emanuel, “Towards learning a universal non- semantic representation of speech,” in Proc. INTERSPEECH, 2020

    2020

  19. [28]

    Learning filter banks within a deep neural network framework,

    T. N. Sainath, B. Kingsbury, A.-r. Mohamed, and B. Ramabhadran, “Learning filter banks within a deep neural network framework,” inProc. IEEE Workshop on Automatic Speech Recognition and Understanding (ASRU), 2013, pp. 297–302

    2013

  20. [29]

    Data-driven harmonic filters for audio representation learning,

    M. Won, S. Chun, O. Nieto, and X. Serrc, “Data-driven harmonic filters for audio representation learning,” in Proc. IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2020, pp. 536–540

    2020

  21. [30]

    Learning a better representation of speech soundwaves using restricted boltzmann machines,

    N. Jaitly and G. Hinton, “Learning a better representation of speech soundwaves using restricted boltzmann machines,” in Proc. IEEE In- ternational Conference on Acoustics, Speech and Signal Processing (ICASSP), 2011, pp. 5884–5887

    2011

  22. [31]

    Estimating phoneme class conditional probabilities from raw speech signal using convolutional neural networks,

    D. Palaz, R. Collobert, and M. M. Doss, “Estimating phoneme class conditional probabilities from raw speech signal using convolutional neural networks,” in Proceedings of INTERSPEECH , 2013, pp. 1766– 1770

    2013

  23. [32]

    Speech acoustic modeling from raw multichannel waveforms,

    Y . Hoshen, R. J. Weiss, and K. W. Wilson, “Speech acoustic modeling from raw multichannel waveforms,” in Proc. IEEE International Con- ference on Acoustics, Speech and Signal Processing (ICASSP) , 2015, pp. 4624–4628

    2015

  24. [33]

    Learnable frontends that do not learn: Quantifying sensitivity to filterbank initialisation,

    M. Anderson, T. Kinnunen, and N. Harte, “Learnable frontends that do not learn: Quantifying sensitivity to filterbank initialisation,” in Proc. IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2023, pp. 1–5

    2023

  25. [34]

    What is Learnt by the LEArn- able Front-end (LEAF)? Adapting Per-Channel Energy Normalisation (PCEN) to Noisy Conditions,

    H. Meng, V . Sethu, and E. Ambikairajah, “What is Learnt by the LEArn- able Front-end (LEAF)? Adapting Per-Channel Energy Normalisation (PCEN) to Noisy Conditions,” inProc. INTERSPEECH, 2023, pp. 2898– 2902

    2023

  26. [35]

    Masked autoencoders that listen,

    P.-Y . Huang, H. Xu, J. Li, A. Baevski, M. Auli, W. Galuba, F. Metze, and C. Feichtenhofer, “Masked autoencoders that listen,”Advances in Neural Information Processing Systems , vol. 35, pp. 28 708–28 720, 2022

    2022

  27. [36]

    Beats: Audio pre-training with acoustic tokenizers,

    S. Chen, Y . Wu, C. Wang, S. Liu, D. Tompkins, Z. Chen, W. Che, X. Yu, and F. Wei, “Beats: Audio pre-training with acoustic tokenizers,” in International Conference on Machine Learning . PMLR, 2023, pp. 5178–5193

    2023

  28. [37]

    Byol for audio: Exploring pre-trained general-purpose audio representations,

    D. Niizumi, D. Takeuchi, Y . Ohishi, N. Harada, and K. Kashino, “Byol for audio: Exploring pre-trained general-purpose audio representations,” IEEE/ACM Transactions on Audio, Speech, and Language Processing , vol. 31, pp. 137–151, 2022

    2022

  29. [38]

    AST: Audio Spectrogram Trans- former,

    Y . Gong, Y .-A. Chung, and J. Glass, “AST: Audio Spectrogram Trans- former,” in Proc. INTERSPEECH 2021 , 2021, pp. 571–575

    2021

  30. [39]

    SSAST: Self-supervised audio spectrogram transformer,

    Y . Gong, C.-I. Lai, Y .-A. Chung, and J. Glass, “SSAST: Self-supervised audio spectrogram transformer,” in Proceedings of the AAAI Conference on Artificial Intelligence , vol. 36, no. 10, 2022, pp. 10 699–10 709

    2022

  31. [40]

    Hubert: Self-supervised speech representation learning by masked prediction of hidden units,

    W.-N. Hsu, B. Bolte, Y .-H. H. Tsai, K. Lakhotia, R. Salakhutdinov, and A. Mohamed, “Hubert: Self-supervised speech representation learning by masked prediction of hidden units,” IEEE/ACM Transactions on Audio, Speech, and Language Processing, vol. 29, pp. 3451–3460, 2021

    2021

  32. [41]

    Wavlm: Large-scale self-supervised pre- training for full stack speech processing,

    S. Chen, C. Wang, Z. Chen, Y . Wu, S. Liu, Z. Chen, J. Li, N. Kanda, T. Yoshioka, X. Xiao et al. , “Wavlm: Large-scale self-supervised pre- training for full stack speech processing,” IEEE Journal of Selected Topics in Signal Processing , vol. 16, no. 6, pp. 1505–1518, 2022

    2022

  33. [42]

    wav2vec 2.0: A framework for self-supervised learning of speech representations,

    A. Baevski, Y . Zhou, A. Mohamed, and M. Auli, “wav2vec 2.0: A framework for self-supervised learning of speech representations,” Advances in neural information processing systems, vol. 33, pp. 12 449– 12 460, 2020

    2020

  34. [43]

    Learning neural audio features without supervision,

    S. Yadav and N. Zeghidour, “Learning neural audio features without supervision,” in Proc. INTERSPEECH, 2022, pp. 396–400

    2022

  35. [44]

    Biologically in- spired adaptive-q filterbanks for replay spoofing attack detection,

    B. Wickramasinghe, E. Ambikairajah, and J. Epps, “Biologically in- spired adaptive-q filterbanks for replay spoofing attack detection,” in Proc. INTERSPEECH, 2019, pp. 2953–2957

    2019

  36. [45]

    Replay detection in voice biometrics: an investiga- tion of adaptive and non-adaptive front-ends,

    B. Wickramasinghe, “Replay detection in voice biometrics: an investiga- tion of adaptive and non-adaptive front-ends,” Ph.D. dissertation, UNSW Sydney, 2021

    2021

  37. [46]

    Speech nonlinearities, mod- ulations, and energy operators,

    P. Maragos, T. F. Quatieri, and J. F. Kaiser, “Speech nonlinearities, mod- ulations, and energy operators,” in Proc. IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , 1991, pp. 421– 424

    1991

  38. [47]

    Encoding frequency modulation to improve cochlear implant performance in noise,

    K. Nie, G. Stickney, and F.-G. Zeng, “Encoding frequency modulation to improve cochlear implant performance in noise,” IEEE Transactions on Biomedical Engineering , vol. 52, no. 1, pp. 64–73, 2004

    2004

  39. [48]

    Detection of replay-spoofing attacks using frequency modulation features,

    T. Gunendradasan, B. Wickramasinghe, P. N. Le, E. Ambikairajah, and J. Epps, “Detection of replay-spoofing attacks using frequency modulation features,” in Proc. INTERSPEECH, 2018, pp. 636–640

    2018

  40. [49]

    Auditory inspired spatial differentiation for replay spoofing attack detection,

    B. Wickramasinghe, E. Ambikairajah, J. Epps, V . Sethu, and H. Li, “Auditory inspired spatial differentiation for replay spoofing attack detection,” in Proc. IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , 2019, pp. 6011–6015

    2019

  41. [50]

    SUPERB: Speech Processing Universal PERformance Benchmark,

    S. wen Yang, P.-H. Chi, Y .-S. Chuang, C.-I. J. Lai, K. Lakhotia, Y . Y . Lin, A. T. Liu, J. Shi, X. Chang, G.-T. Lin, T.-H. Huang, W.-C. Tseng, K. tik Lee, D.-R. Liu, Z. Huang, S. Dong, S.-W. Li, S. Watanabe, A. Mohamed, and H. yi Lee, “SUPERB: Speech Processing Universal PER...

    2021

  42. [51]

    The GTZAN dataset: Its contents, its faults, their effects on evaluation, and its future use,

    B. L. Sturm, “The GTZAN dataset: Its contents, its faults, their effects on evaluation, and its future use,” CoRR, vol. abs/1306.1461, 2013. [Online]. Available: http://arxiv.org/abs/1306.1461

    2013 arXiv

  43. [52]

    Deep learning and music adversaries,

    C. Kereliuk, B. L. Sturm, and J. Larsen, “Deep learning and music adversaries,” IEEE Transactions on Multimedia , vol. 17, no. 11, pp. 2059–2071, 2015

    2015

  44. [53]

    FMA: A dataset for music analysis,

    M. Defferrard, K. Benzi, P. Vandergheynst, and X. Bresson, “FMA: A dataset for music analysis,” in Proc. of International Society for Music Information Retrieval Conference, ISMIR , S. J. Cunningham, Z. Duan, X. Hu, and D. Turnbull, Eds., 2017, pp. 316–323. JOURNAL OF LATEX CL...

    2017

  45. [54]

    CREMA-D: Crowd-sourced emotional multimodal actors dataset,

    H. Cao, D. G. Cooper, M. K. Keutmann, R. C. Gur, A. Nenkova, and R. Verma, “CREMA-D: Crowd-sourced emotional multimodal actors dataset,” IEEE Transactions on Affective Computing , vol. 5, no. 4, pp. 377–390, 2014

    2014

  46. [55]

    IEMOCAP: Interactive emotional dyadic motion capture database,

    C. Busso, M. Bulut, C.-C. Lee, A. Kazemzadeh, E. Mower, S. Kim, J. N. Chang, S. Lee, and S. S. Narayanan, “IEMOCAP: Interactive emotional dyadic motion capture database,” Language resources and evaluation , vol. 42, pp. 335–359, 2008

    2008

  47. [56]

    Speech commands: A dataset for limited-vocabulary speech recognition,

    P. Warden, “Speech commands: A dataset for limited-vocabulary speech recognition,” CoRR, vol. abs/1804.03209, 2018

    2018 arXiv

  48. [57]

    V oxceleb: A large-scale speaker identification dataset,

    A. Nagrani, J. S. Chung, and A. Zisserman, “V oxceleb: A large-scale speaker identification dataset,” in Proc. INTERSPEECH, F. Lacerda, Ed., 2017, pp. 2616–2620

    2017

  49. [58]

    Efficientnet: Rethinking model scaling for convolutional neural networks,

    M. Tan and Q. V . Le, “Efficientnet: Rethinking model scaling for convolutional neural networks,” in Proc. Internation Conference on Machine Learning (ICML) , vol. 97, 2019, pp. 6105–6114

    2019

  50. [59]

    Squeeze-and-excitation networks,

    J. Hu, L. Shen, and G. Sun, “Squeeze-and-excitation networks,” in Proc. CVPR, 2018, pp. 7132–7141

    2018

  51. [60]

    MobilenNetv2: Inverted residuals and linear bottlenecks,

    M. Sandler, A. Howard, M. Zhu, A. Zhmoginov, and L.-C. Chen, “MobilenNetv2: Inverted residuals and linear bottlenecks,” in Proc. IEEE Conference on Computer Vision and Pattern Recognition , 2018, pp. 4510–4520

    2018

  52. [61]

    Adam: A method for stochastic optimization,

    D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” in Proc. ICLR, 2015

    2015

  53. [62]

    Audio set: An ontology and human- labeled dataset for audio events,

    J. F. Gemmeke, D. P. Ellis, D. Freedman, A. Jansen, W. Lawrence, R. C. Moore, M. Plakal, and M. Ritter, “Audio set: An ontology and human- labeled dataset for audio events,” inProc. IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , 2017, pp. 776– 780

    2017

  54. [63]

    PANNs: Large-scale pretrained audio neural networks for audio pattern recognition,

    Q. Kong, Y . Cao, T. Iqbal, Y . Wang, W. Wang, and M. D. Plumbley, “PANNs: Large-scale pretrained audio neural networks for audio pattern recognition,” IEEE/ACM Transactions on Audio, Speech, and Language Processing, vol. 28, pp. 2880–2894, 2020

    2020

  55. [64]

    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 Proc. IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2015, pp. 5206–5210

    2015

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.