Exploring the encoding of linguistic representations in the Fully-Connected Layer of generative CNNs for Speech

Bruno Ferenc \v{S}egedin; Gasper Begu\v{s}

REVIEW 2 major objections 5 minor 1 cited by

Exploring the encoding of linguistic representations in the Fully-Connected Layer of generative CNNs for Speech

T0 review · 2 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read A speech-generating CNN encodes similar vowels with similar internal weight patterns across different words, rather than storing each word as an unanalyzable whole.

desk verdict A genuinely novel set of FC-layer interpretability manipulations for a speech GAN, but the main claim about shared vowel encodings is undermined by output-dependent column selection and a lack of statistical controls. read the letter →

arxiv 2501.07726 v1 pith:2JAPKN25 submitted 2025-01-13 cs.CL

Bruno Ferenc Šegedin , Gasper Beguš This is my paper · ORCID

classification cs.CL

keywords AIinterpretabilitygenerativeadversarialnetworksspeechsynthesisfullyconnectedlayerlatentspacephoneticcategoriesciwGANsublexicalrepresentation

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 tries to show that the fully-connected layer of ciwGAN, a generative CNN for speech, does not merely memorize each lexical item as an indivisible code. Instead, the layer's learned weights contain shared, lexically invariant structure: the same vowel quality is encoded in similar ways across distinct word-specific latent codes. The authors develop two techniques to see this: feeding a latent variable's weight matrix directly into the convolutional layers, and extracting single feature-map columns that produce an isolated vowel. If the claim holds, it matters because it indicates that sublexical sound categories can emerge inside a generative model trained only to reproduce waveforms, strengthening the case that such networks can serve as models of phonological learning.

What carries the argument

The central object is the variable-specific weight matrix $W_i$ of the fully-connected layer: for each of the 100 latent inputs (9 one-hot lexical codes plus 91 noise variables), the trained weights form a $1024 \times 16$ feature-map matrix whose time axis mirrors the output waveform. Because the FC layer's pre-activation output is the sum of these 100 weight matrices each scaled by its latent variable value, each matrix can be treated as an independently testable blueprint of that variable's contribution. Passing a full weight matrix into the convolutional block reproduces the corresponding word, and passing a single extracted column in isolation yields a single vowel. Correlation matrices of these vowel columns, together with correlation matrices of the averaged output spectra, carry the argument that representational similarity in the FC layer tracks acoustic and phonological similarity.

What would settle it

Train a fresh ciwGAN on the same nine words with a different random seed, extract vowel columns by a fixed rule that does not listen to outputs (for example, the column at the temporal position matching each vowel's known duration in the training data), and compute the correlation matrix of those columns; if high-vowel columns no longer correlate more strongly with each other than with low-vowel columns, the reported structure is an artifact of output-guided column selection.

Watch

Extended reading notes

Core claim

The paper's central discovery is that in ciwGAN each latent code's fully-connected weight matrix is not a monolithic word code but contains temporally organized columns that, passed alone through the convolutional layers, produce a single vowel. Comparing 12 such vowel columns taken from the 9 lexical items shows that phonetically similar vowels have positively correlated weight patterns, while high and low vowels are negatively correlated; multidimensional scaling places the high vowels together and the low vowels apart. The authors state this as evidence that similar-sounding vowels are encoded in similar ways in the FC layer, both within and across the weight matrices of lexically specific latent codes. They further show that these vowel columns behave compositionally: placing a vowel column beside a copy of itself or beside a vowel column from another word consistently yields an /s/ plus vowel output, with the /s/ and vowel information located in distinct channel groups.

Load-bearing premise

The central result assumes that a single 1024-sample column of FC weights, chosen by inspecting which column produces only a vowel when run through the convolutional layers, is a valid and representative encoding of that vowel in the model; if that selection is confounded by how the convolutional layers respond to isolated columns, the clustering of similar vowels could be an artifact.

Editorial extensions

If this is right

If the claim is correct, ciwGAN's lexical codes are not holistic word templates; shared sublexical structure across words implies the model has learned emergent vowel categories without supervision.
The weight-matrix-as-input technique provides a label-free way to explore a latent space: each code's weight matrix alone can regenerate its lexical item's waveform, enabling bottom-up analysis of other trained models.
The column-extraction and recombination technique shows that FC columns act as compositional units, which could support segment-level manipulation in latent speech models, such as swapping vowel columns between words.
The finding that distinct channel groups specialize in frication versus vowel information suggests redundancy in the FC layer, implying the network might be compressible without losing its linguistically structured representations.
Because the methods only require a fully-connected projection from a latent space, the same techniques should transfer to other generative architectures, such as autoencoders, for probing their internal linguistic structure.

Reading between the lines

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

The paper selects vowel columns by listening to which isolated column yields a clean vowel; a fully blind selection rule based only on temporal position or spectral criteria would be a stronger test that the correlation structure is not an artifact of output-guided column choice.
The demonstrated compositionality of columns suggests a concrete next experiment the paper does not run: transplanting the /i/ column from 'greasy' into the feature map of 'suit' and checking whether the output becomes something like 'seat', which would test word-level recombination from sublexical units.
If the channel-localization result generalizes, ablating the specific channel groups that the paper found to carry pure vowel or pure frication information should remove those sounds from a full word output, giving a causal test of the claimed representational dissociation.
A natural extension is to apply column extraction to consonants rather than only vowels; the prediction would be that any sublexical segment with a consistent acoustic realization forms its own correlated cluster of FC columns across codes.

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

The useful part of this paper is the toolkit. Passing trained weight matrices into the convolutional layers as feature maps (Experiment 1) is a neat idea, and the result that each lexical code's weights produce the right word is a clean, concrete demonstration. The channel-slicing and column-recombination manipulations in Experiment 2 are also new for speech models — I haven't seen anyone isolate single FC columns and splice them into manually assembled feature maps before. The 64-channel zeroing experiment is a sensible way to localize information. As a methodological contribution, this is worth having.

The soft spot is exactly where the reader's stress-test lands. In Experiment 2, the authors select the vowel columns by generating outputs and picking the column that "contains only vowel information" (Section 4, Fig. 9). They then compute correlations among those same selected columns and report that similar vowels cluster. That is a selection on the outcome measure. If the selection procedure favors columns with clean, well-defined formants, those columns could cluster by acoustic cleanliness rather than by phonological height. There is no algorithmic selection rule, no inter-rater agreement, no baseline with randomly chosen columns or all 16 columns, and no significance testing. With N=12 and a visually interpreted MDS plot, the central claim that ciwGAN encodes lexical items in a "linguistically principled manner" is not yet supported. The concern is not circular parameter fitting — the correlation analysis itself does not use the vowel labels — but selection bias is enough to make the headline result fragile.

A second, related issue is the input regime. Passing isolated columns or weight matrices into convolutional layers trained on full feature maps is not a validated operation. The authors note this for the z-variables but not for the vowel columns. The bias term and all other latent variables are absent, so the outputs live in a part of input space the model never saw. The fact that those outputs are interpretable is interesting, but it does not license strong conclusions about how the model actually represents vowels during normal generation.

Who gets value: people working on GAN interpretability, speech representation learning, and maybe phonological acquisition. The methods are worth trying on other architectures. The main conclusion, however, should be treated as a suggestive observation, not an established result.

Recommendation: send it to peer review, but with the expectation of major revision. The authors need an algorithmic column-selection rule, a null baseline, significance tests, and a validation of the isolated-column regime (e.g., comparing against full-input generations). If those arrive, the paper could be solid. As is, it is a promising methods paper with an overstated conclusion.

Referee Report

2 major / 5 minor

Summary. The paper proposes two interpretability techniques for the fully-connected (FC) layer of ciwGAN, a WaveGAN variant with categorical latent codes trained on TIMIT words. Experiment 1 shows that latent-code weight matrices have larger average magnitudes than noise-variable matrices and, when passed through the convolutional layers, generate the corresponding lexical items. Experiment 2 extracts individual FC columns that, in isolation, produce single vowels; comparing 12 such columns across 9 lexical codes, the paper reports that high vowels cluster together and low vowels cluster together in a correlation matrix and MDS plot, and that recombining vowel columns yields predictable /s/+vowel outputs. The central claim is that ciwGAN encodes lexical information in a linguistically principled way, with shared lexically-invariant sublexical representations in the FC layer.

Significance. If the central claim is substantiated, the paper makes a genuinely useful contribution to interpretability of generative speech models: it introduces weight-matrix-as-input probing and column/channel manipulation as tools that go beyond latent-space interpolation, and it provides evidence against purely holistic lexical encoding in a GAN generator. The work is exploratory and methodologically creative, and the channel-masking experiment (Fig. 17) is a nice illustration of how information can be localized. However, the main conclusion currently rests on a manual, potentially circular column-selection procedure and on descriptive statistics without significance testing, so the strength of the claim is not yet commensurate with the evidence.

major comments (2)

[Section 4.1, Figs. 10-13] The correlation and MDS analyses are purely descriptive, with no significance tests, confidence intervals, or effect sizes reported. With only 12 columns and strong dependencies among the 1024 values within each column, the observed high/low grouping could plausibly arise by chance. A permutation test (e.g., shuffling vowel-category labels or the set of columns) or a bootstrap over latent codes is needed before concluding that similar-sounding vowels are encoded similarly in the FC layer.
[Section 2.2 and Section 4] The isolated-column probe sets all other latent variables to zero and omits the trainable bias term, even though the convolutional layers were trained on feature maps that are sums of 100 scaled weight matrices plus a bias. The paper acknowledges this limitation for z-variables in Section 3.4 ('their outputs should be interpreted as mere approximations'), but it does not address the same issue in Experiment 2, where the isolated-column input is the basis for the main claim. The authors should show that the results are robust to including the bias term or to varying the other latent variables, or explicitly justify why an off-manifold single-column input is a valid probe of the FC layer's encoding.

minor comments (5)

[Section 1.2] There is a typo: 'have been frequently used for for image and audio generation' should read 'frequently used for image and audio generation.'
[Section 3.2] The phrase 'we demonstrates that' should be 'we demonstrate that.'
[Section 4.1 (Fig. 12 caption)] The caption states that high-front vowels have a bimodal peak 'representing a high F1 and low F2,' but high-front vowels have low F1 and high F2; mid-low vowels have higher F1 and lower F2. This should be corrected for accuracy.
[Section 4.2] The claim that feature columns 'can function as compositional units' is asserted on the basis of /s/+vowel outputs from repeated columns, but the paper itself proposes an alternative explanation in terms of convolutional kernel interactions and channel alignment. This wording should be moderated, since the evidence distinguishes predictable contextual effects from true compositionality only weakly.
[General] No code, model checkpoint, or data availability statement is provided; given that the analysis relies on manual selection and qualitative inspection, releasing code and the exact column indices would substantially improve reproducibility.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity; the central clustering result is not forced by construction, though manual column selection poses a validity risk.

full rationale

The paper's central claim, that lexically specific ciwGAN latent codes share sublexical FC-weight structure, is not derived from a fitted parameter or from the vowel labels themselves. Experiment 2 extracts one weight column per vowel by passing columns through the convolutional layers and selecting the column whose output contains only vowel information; the 12x12 FC correlation matrix and MDS are then computed on those same selected columns. The observed grouping of high versus low vowels is not guaranteed by construction: the selection criterion (output is a clear vowel) does not entail that columns for similar vowels will be mutually correlated, and the paper provides some independent contrasts, such as z-variable weight matrices yielding non-lexical outputs in Experiment 1b. The principal weakness is selection bias: columns are chosen manually by inspecting outputs, with no algorithmic rule, inter-rater reliability, or null baseline of randomly selected columns, so the clustering could partly reflect the selection criterion rather than a general lexically invariant encoding. That is a methodological validity concern, not a circular derivation. Self-citations to Beguš's prior ciwGAN work supply the model architecture and background, but the new weight-matrix analysis is conducted in this paper and does not reduce to those citations. The unsupported assertion that the current implementation showed 'similar success, according to metrics used by prior studies' is a reporting gap, not a circular step. No equation or claim is equivalent to its own input by construction; thus no significant circularity is found.

Assumptions & free parameters 1 free parameters · 2 assumptions · 0 invented entities

The paper introduces no new physical or model entities. Its analytical claims rest on treating FC weight slices as interpretable representations, which is an assumption rather than an independently evidenced entity.

free parameters (1)

Manually selected vowel column indices (12 columns) = not reported (only figure references)
The central clustering result depends on which 12 columns of the 9 latent-code weight matrices are chosen as vowel columns; the choice is made by inspecting generated outputs, and the exact indices are not listed in the paper.

assumptions (2)

ad hoc to paper Isolated weight matrices or columns, passed alone into the convolutional layers, are valid probes of what a latent code encodes, even though the convolutional layers were trained on feature maps that sum all 100 latent variables plus a bias term.
Section 3.3 and Experiment 2 feed single weight matrices or single columns as inputs without comparing to the full-input activation map or accounting for the bias term.
ad hoc to paper A manually selected FC column that produces only a vowel in isolation is a faithful representation of that vowel in the model's lexical code.
Section 4 selects columns by output inspection; no independent or algorithmic criterion is given, so the subsequent similarity analysis inherits this selection.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Exploring the encoding of linguistic representations in the Fully-Connected Layer of generative CNNs for Speech." pith.science (2026). https://pith.science/paper/2JAPKN25

@misc{pith2026250107726,
  author       = {Pith},
  title        = {Pith review of: Exploring the encoding of linguistic representations in the Fully-Connected Layer of generative CNNs for Speech},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2JAPKN25}},
  note         = {Machine review of arXiv:2501.07726}
}

read the original abstract

Interpretability work on the convolutional layers of CNNs has primarily focused on computer vision, but some studies also explore correspondences between the latent space and the output in the audio domain. However, it has not been thoroughly examined how acoustic and linguistic information is represented in the fully connected (FC) layer that bridges the latent space and convolutional layers. The current study presents the first exploration of how the FC layer of CNNs for speech synthesis encodes linguistically relevant information. We propose two techniques for exploration of the fully connected layer. In Experiment 1, we use weight matrices as inputs into convolutional layers. In Experiment 2, we manipulate the FC layer to explore how symbolic-like representations are encoded in CNNs. We leverage the fact that the FC layer outputs a feature map and that variable-specific weight matrices are temporally structured to (1) demonstrate how the distribution of learned weights varies between latent variables in systematic ways and (2) demonstrate how manipulating the FC layer while holding constant subsequent model parameters affects the output. We ultimately present an FC manipulation that can output a single segment. Using this technique, we show that lexically specific latent codes in generative CNNs (ciwGAN) have shared lexically invariant sublexical representations in the FC-layer weights, showing that ciwGAN encodes lexical information in a linguistically principled manner.

Figures

Figures reproduced from arXiv: 2501.07726 by the authors.

**Figure 2.** A schematic representation showing that the output of the FC-layer (before ReLU activation) can be [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** Average absolute weights of weight matrices for every variable in the latent space. The first 9 points are the [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Figures from the paper (14 more)

**Figure 4.** Figure 4: Average of absolute weight values along channel length. The blue curves represent the weights of uniformly [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]

**Figure 5.** Figure 5: Output waveforms, each derived from passing variable-specific weight matrices as inputs into the convolutional [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

**Figure 6.** Figure 6: Output waveforms derived from passing variable-specific weight matrices. Weight matrices for uniformly [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]

**Figure 7.** Figure 7: Visualization of entire weight matrix for the latent code that encodes "suit". The values within the red dashed [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]

**Figure 8.** Figure 8: Weight columns for every vowel. These 12 columns are extracted from the latent code weight matrices. Each [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]

**Figure 9.** Figure 9: The waveform output of each isolated vowel column when passed as input into the convolutional layers. The [PITH_FULL_IMAGE:figures/full_fig_p011_9.png]

**Figure 10.** Figure 10: Correlation Matrix of feature map FC-codes (columns in the FC feature map) that produce each vowel [PITH_FULL_IMAGE:figures/full_fig_p011_10.png]

**Figure 11.** Figure 11: Multidimensional-scaling plot. The MDS algorithm infers a 2d space on which to situate the points [PITH_FULL_IMAGE:figures/full_fig_p012_11.png]

**Figure 12.** Figure 12: Averaged spectra for the output of each vowel code. High-front vowels have a bimodal peak, representing a [PITH_FULL_IMAGE:figures/full_fig_p012_12.png]

**Figure 13.** Figure 13: Correlation Matrix of spectra (spectrograms averaged across the time dimension) derived from the waveform [PITH_FULL_IMAGE:figures/full_fig_p013_13.png]

**Figure 14.** Figure 14: Visualization of the manually-constructed input into the convolutional layers we use to test how the outputs [PITH_FULL_IMAGE:figures/full_fig_p013_14.png]

**Figure 15.** Figure 15: Visualization of one example of a manually-constructed input into the convolutional layers we use to test [PITH_FULL_IMAGE:figures/full_fig_p014_15.png]

**Figure 16.** Figure 16: Outputs for every 3-column feature map consisting of columns that in isolation encode the first vowel in [PITH_FULL_IMAGE:figures/full_fig_p015_16.png]

**Figure 17.** Figure 17: Waveforms generated by passing 64-channel samples of the full weight matrix consisting of 3 identical [PITH_FULL_IMAGE:figures/full_fig_p016_17.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

A Technique for Isolating Lexically-Independent Phonetic Dependencies in Generative CNNs
cs.CL 2025-06 conditional novelty 6.0 of 10

A new probe that bypasses the fully-connected layer shows convolutional layers of a small-bottleneck WaveGAN can reflect a phonotactic restriction learned from lexical training data.

Reference graph

Works this paper leans on

36 extracted references · 30 canonical work pages · cited by 1 Pith paper

[1]

Ciwgan and fiwgan: Encoding information in acoustic data to model lexical learning with generative adversarial networks

Gašper Beguš. Ciwgan and fiwgan: Encoding information in acoustic data to model lexical learning with generative adversarial networks. Neural Networks, 139:305–325, 2021

work page 2021
[2]

Generative adversarial networks

Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial networks. Communications of the ACM, 63(11):139–144, 2020

2020
[3]

Modeling unsupervised phonetic and phonological learning in generative adversarial phonology

Gašper Beguš. Modeling unsupervised phonetic and phonological learning in generative adversarial phonology. In Proceedings of the Society for Computation in Linguistics 2020, pages 38–48, 2020

work page 2020
[4]

Exploring how generative adversarial networks learn phonological representations

Jingyi Chen and Micha Elsner. Exploring how generative adversarial networks learn phonological representations. arXiv preprint arXiv:2305.12501, 2023

arXiv 2023
[5]

Linguistic generalization and compositionality in modern artificial neural networks

Marco Baroni. Linguistic generalization and compositionality in modern artificial neural networks. Philosophical Transactions of the Royal Society B, 375(1791):20190307, 2020

work page 2020
[6]

Analyzing hidden representations in end-to-end automatic speech recognition systems

Yonatan Belinkov and James Glass. Analyzing hidden representations in end-to-end automatic speech recognition systems. Advances in Neural Information Processing Systems, 30, 2017

work page 2017
[7]

The influence of categories on perception: explaining the perceptual magnet effect as optimal statistical inference

Naomi H Feldman, Thomas L Griffiths, and James L Morgan. The influence of categories on perception: explaining the perceptual magnet effect as optimal statistical inference. Psychological review, 116(4):752, 2009

work page 2009
[8]

V owel normalization as perceptual constancy

Santiago Barreda. V owel normalization as perceptual constancy. Language, 96(2):224–254, 2020

work page 2020

Show all 36 references

[9]

Systematicity, but not compositionality: Examining the emergence of linguistic structure in children and adults using iterated learning

Limor Raviv and Inbal Arnon. Systematicity, but not compositionality: Examining the emergence of linguistic structure in children and adults using iterated learning. Cognition, 181:160–173, 2018

2018
[10]

Statistical learning of tone sequences by human infants and adults

Jenny R Saffran, Elizabeth K Johnson, Richard N Aslin, and Elissa L Newport. Statistical learning of tone sequences by human infants and adults. Cognition, 70(1):27–52, 1999

1999
[11]

Categorization of speech by infants: Support for speech-sound prototypes

DiAnne Grieser and Patricia K Kuhl. Categorization of speech by infants: Support for speech-sound prototypes. Developmental Psychology, 25(4):577, 1989

1989
[12]

Acquiring language from speech by learning to remember and predict

Cory Shain and Micha Elsner. Acquiring language from speech by learning to remember and predict. In Raquel Fernández and Tal Linzen, editors, Proceedings of the 24th Conference on Computational Natural Language Learning, pages 195–214, Online, November 2020. Association for Co...

2020
[13]

Cnn-generated images are surprisingly easy to spot

Sheng-Yu Wang, Oliver Wang, Richard Zhang, Andrew Owens, and Alexei A Efros. Cnn-generated images are surprisingly easy to spot... for now. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 8695–8704, 2020

2020
[14]

Unsupervised cross-domain image generation

Yaniv Taigman, Adam Polyak, and Lior Wolf. Unsupervised cross-domain image generation. arXiv preprint arXiv:1611.02200, 2016

2016 arXiv
[15]

Unsupervised modeling of vowel harmony using wavegan

Sneha Ray Barman, Shakuntala Mahanta, and Neeraj Kumar Sharma. Unsupervised modeling of vowel harmony using wavegan. In Proc. SpeechProsody 2024, pages 200–204, 2024

2024
[16]

Identity-based patterns in deep convolutional networks: Generative adversarial phonology and reduplication

Gašper Beguš. Identity-based patterns in deep convolutional networks: Generative adversarial phonology and reduplication. Transactions of the Association for Computational Linguistics, 9:1180–1196, 2021

2021
[17]

Articulation GAN: Unsupervised modeling of articulatory learning

Gašper Beguš, Alan Zhou, Peter Wu, and Gopala K Anumanchipalli. Articulation GAN: Unsupervised modeling of articulatory learning. In ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 1–5. IEEE, 2023

2023
[18]

Measuring the perceptual availability of phonological features during language acquisition using unsupervised binary stochastic autoencoders

Cory Shain and Micha Elsner. Measuring the perceptual availability of phonological features during language acquisition using unsupervised binary stochastic autoencoders. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational L...

2019
[19]

How familiar does that sound? cross-lingual representational similarity analysis of acoustic word embeddings

Badr M Abdullah, Iuliia Zaitova, Tania Avgustinova, Bernd Möbius, and Dietrich Klakow. How familiar does that sound? cross-lingual representational similarity analysis of acoustic word embeddings. arXiv preprint arXiv:2109.10179, 2021

2021 arXiv
[20]

Disentanglement in a gan for unconditional speech synthesis

Matthew Baas and Herman Kamper. Disentanglement in a gan for unconditional speech synthesis. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 2024

2024
[21]

Interpreting intermediate convolutional layers in unsupervised acoustic word classification

Gašper Beguš and Alan Zhou. Interpreting intermediate convolutional layers in unsupervised acoustic word classification. In ICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 8207–8211. IEEE, 2022

2022
[22]

Interpreting intermediate convolutional layers of generative cnns trained on waveforms

Gašper Beguš and Alan Zhou. Interpreting intermediate convolutional layers of generative cnns trained on waveforms. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 30:3214–3229, 2022

2022
[23]

Local and non-local dependency learning and emergence of rule-like representations in speech data by deep convolutional generative adversarial networks

Gašper Beguš. Local and non-local dependency learning and emergence of rule-like representations in speech data by deep convolutional generative adversarial networks. Computer Speech & Language, 71:101244, 2022

2022
[24]

Modeling speech recognition and synthesis simultaneously: Encoding and decoding lexical and sublexical semantic information into speech with no direct access to speech data

Gašper Beguš and Alan Zhou. Modeling speech recognition and synthesis simultaneously: Encoding and decoding lexical and sublexical semantic information into speech with no direct access to speech data. In Proc. Interspeech 2022, pages 5298–5302, 2022

2022
[25]

Adversarial audio synthesis

Chris Donahue, Julian McAuley, and Miller Puckette. Adversarial audio synthesis. arXiv preprint arXiv:1802.04208, 2018

2018 arXiv
[26]

InfoGAN: Interpretable representation learning by information maximizing generative adversarial nets

Xi Chen, Yan Duan, Rein Houthooft, John Schulman, Ilya Sutskever, and Pieter Abbeel. InfoGAN: Interpretable representation learning by information maximizing generative adversarial nets. In Advances in Neural Information Processing Systems 29, pages 2172–2180. Curran Associate...

2016
[27]

Timit acoustic-phonetic continuous speech corpus

John S Garofolo, Lori F Lamel, William M Fisher, David S Pallett, Nancy L Dahlgren, Victor Zue, and Jonathan G Fiscus. Timit acoustic-phonetic continuous speech corpus. (No Title), 1993

1993
[28]

Weiss, Samy Bengio, and Aäron van den Oord

Jan Chorowski, Ron J. Weiss, Samy Bengio, and Aäron van den Oord. Unsupervised speech representation learning using wavenet autoencoders. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 27(12):2041–2053, 2019

2019
[29]

Visualizing and understanding convolutional networks, 2013

Matthew D Zeiler and Rob Fergus. Visualizing and understanding convolutional networks, 2013

2013
[30]

Approaching an unknown communication system by latent space exploration and causal inference

Gašper Beguš, Andrej Leban, and Shane Gero. Approaching an unknown communication system by latent space exploration and causal inference. arXiv, 2024

2024
[31]

Speech perception without speaker normalization: An exemplar model

Keith Johnson et al. Speech perception without speaker normalization: An exemplar model. Talker variability in speech processing, pages 145–165, 1997

1997
[32]

Exemplar dynamics: Word frequency, lenition, and contrast

Janet Pierrehumbert. Exemplar dynamics: Word frequency, lenition, and contrast. Frequency and the Emergence of Linguistic Structure/John Benjamins, 2001

2001
[33]

Advancement of phonetics in the 21st century: Exemplar models of speech production

Matthew Goldrick and Jennifer Cole. Advancement of phonetics in the 21st century: Exemplar models of speech production. Journal of Phonetics, 99:101254, 2023

2023
[34]

The representation of speech variability and variation in deep neural networks

Badr Mohammed Badr Abdullah. The representation of speech variability and variation in deep neural networks. PhD thesis, Saarland University, 2024. 18 A PREPRINT - JANUARY 15, 2025

2024
[35]

Walking the tightrope: An investigation of the convolutional autoencoder bottleneck

Ilja Manakov, Markus Rohm, and V olker Tresp. Walking the tightrope: An investigation of the convolutional autoencoder bottleneck. arXiv preprint arXiv:1911.07460, 2019

1911 arXiv
[36]

Christina Zhao

Gašper Beguš, Alan Zhou, and T. Christina Zhao. Encoding of speech in convolutional layers and the brain stem based on language experience. Scientific Reports, 13(1):6480, 2023. 19

2023

Pith tools

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

[1] [1]

Ciwgan and fiwgan: Encoding information in acoustic data to model lexical learning with generative adversarial networks

Gašper Beguš. Ciwgan and fiwgan: Encoding information in acoustic data to model lexical learning with generative adversarial networks. Neural Networks, 139:305–325, 2021

work page 2021

[2] [2]

Generative adversarial networks

Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial networks. Communications of the ACM, 63(11):139–144, 2020

2020

[3] [3]

Modeling unsupervised phonetic and phonological learning in generative adversarial phonology

Gašper Beguš. Modeling unsupervised phonetic and phonological learning in generative adversarial phonology. In Proceedings of the Society for Computation in Linguistics 2020, pages 38–48, 2020

work page 2020

[4] [4]

Exploring how generative adversarial networks learn phonological representations

Jingyi Chen and Micha Elsner. Exploring how generative adversarial networks learn phonological representations. arXiv preprint arXiv:2305.12501, 2023

arXiv 2023

[5] [5]

Linguistic generalization and compositionality in modern artificial neural networks

Marco Baroni. Linguistic generalization and compositionality in modern artificial neural networks. Philosophical Transactions of the Royal Society B, 375(1791):20190307, 2020

work page 2020

[6] [6]

Analyzing hidden representations in end-to-end automatic speech recognition systems

Yonatan Belinkov and James Glass. Analyzing hidden representations in end-to-end automatic speech recognition systems. Advances in Neural Information Processing Systems, 30, 2017

work page 2017

[7] [7]

The influence of categories on perception: explaining the perceptual magnet effect as optimal statistical inference

Naomi H Feldman, Thomas L Griffiths, and James L Morgan. The influence of categories on perception: explaining the perceptual magnet effect as optimal statistical inference. Psychological review, 116(4):752, 2009

work page 2009

[8] [8]

V owel normalization as perceptual constancy

Santiago Barreda. V owel normalization as perceptual constancy. Language, 96(2):224–254, 2020

work page 2020