Pith. sign in

REVIEW 1 major objections 41 references

Reviewed by Pith at T0; open to challenge.

T0 means a machine referee read the full paper against a public rubric. The mark states how deep the mechanical check went, never who wrote it. the ladder, T0–T4 →

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

T0 review · grok-4.3

Cross-language mismatch causes larger drops in acoustic-to-articulatory inversion performance than cross-gender mismatch.

2026-06-26 15:22 UTC pith:Q3JDDRMT

load-bearing objection New bilingual EMA corpus with cross-gender and cross-language benchmarks on standard AAI setups; useful data point but thin on controls and stats. the 1 major comments →

arxiv 2606.20478 v1 pith:Q3JDDRMT submitted 2026-06-18 eess.AS

Beyond Speaker Independence: Evaluating Cross-Lingual Acoustic-to-Articulatory Inversion Across Finnish and Russian

Ruchi Pandey , Tomi Kinnunen This is my paper
classification eess.AS
keywords acoustic-to-articulatory inversioncross-lingual transferelectromagnetic articulographydomain shiftFinnishRussianPearson correlationspeaker independence
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper evaluates acoustic-to-articulatory inversion when models encounter shifts in speaker gender within a language or shifts between languages, using a new bilingual Finnish-Russian electromagnetic articulography corpus. It tests raw EMA coordinates against tract variables as targets, MFCC against self-supervised learning features as inputs, and BiLSTM against attention-based sequence models as predictors. The central result is that gender mismatch produces moderate Pearson correlation declines while language mismatch produces substantially larger ones. This work supplies the first systematic benchmarks on a resource designed to reduce English-centric bias in the field.

Core claim

On the FROST-EMA corpus the authors define protocols for within-language cross-gender transfer and within-gender cross-language transfer. They report that the former produces Pearson correlation drops of roughly 0.05 to 0.10 from the matched in-domain baseline, whereas the latter produces drops of roughly 0.10 to 0.20. These comparisons are obtained after benchmarking the three articulatory targets, two acoustic front-ends, and two inversion architectures.

What carries the argument

The cross-gender (within-language) and cross-language (within-gender) transfer evaluation protocols on the FROST-EMA bilingual EMA corpus that quantify Pearson correlation under controlled domain shifts.

Load-bearing premise

That the measured performance differences arise primarily from the intended gender and language mismatches rather than from imbalances in recording conditions, data volume per language, or speaker selection within the FROST-EMA corpus.

What would settle it

An experiment that equalizes training data volume, recording conditions, and speaker count across languages and genders and then re-computes the Pearson correlation gaps would show whether the reported language effect remains larger than the gender effect.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Language mismatch remains the dominant obstacle to generalization, so methods that succeed on gender variation will not automatically solve cross-lingual AAI.
  • Within-language gender-mixed training is more feasible than cross-lingual training on current architectures.
  • The FROST-EMA corpus supplies the first controlled test bed for isolating language effects from speaker effects in articulatory inversion.
  • Future model comparisons should report both matched and mismatched conditions using the defined protocols.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Multilingual training data may be required to close the larger language-induced gap, rather than simply adding more speakers of one language.
  • The moderate gender effect suggests that gender-balanced single-language corpora can still yield usable AAI systems.
  • Repeating the protocol on additional language pairs would test whether the 0.10–0.20 drop magnitude is specific to Finnish–Russian or holds more generally.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

1 major / 0 minor

Summary. The manuscript conducts a systematic evaluation of acoustic-to-articulatory inversion (AAI) under domain shifts using the FROST-EMA Finnish-Russian bilingual EMA corpus. It benchmarks articulatory targets (raw EMA coordinates vs. tract variables), acoustic front-ends (MFCC vs. SSL features), and inversion back-ends (BiLSTM vs. lightweight attention model), while defining explicit protocols for cross-gender transfer (within-language) and cross-language transfer (within-gender). The central empirical claim is that cross-gender mismatch produces moderate Pearson correlation declines (approximately 0.05–0.10) relative to in-domain baselines, whereas cross-language mismatch produces larger drops (approximately 0.10–0.20).

Significance. If the reported correlation differences prove robust, the work supplies useful non-English benchmarks for AAI and quantifies the relative impact of gender versus language mismatch. The introduction of the FROST-EMA corpus directly addresses the English bias and limited speaker diversity noted in prior resources.

major comments (1)
  1. [Abstract] Abstract: the reported Pearson correlation declines (0.05–0.10 for cross-gender, 0.10–0.20 for cross-language) are presented without any accompanying information on training data volume per condition, speaker counts per split, statistical significance tests, error bars, or explicit controls for recording-condition or speaker-selection imbalances within FROST-EMA. These omissions are load-bearing for the central claim that the observed drops arise from the intended mismatches rather than from uncontrolled covariates.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address the major comment below and indicate planned revisions where appropriate.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the reported Pearson correlation declines (0.05–0.10 for cross-gender, 0.10–0.20 for cross-language) are presented without any accompanying information on training data volume per condition, speaker counts per split, statistical significance tests, error bars, or explicit controls for recording-condition or speaker-selection imbalances within FROST-EMA. These omissions are load-bearing for the central claim that the observed drops arise from the intended mismatches rather than from uncontrolled covariates.

    Authors: We agree that the abstract's brevity omits key supporting details that strengthen the central claim. The full manuscript (Section 3) specifies the FROST-EMA corpus composition (8 speakers total: 4 Finnish and 4 Russian, balanced by gender, with ~2 hours of data per speaker) and the evaluation protocols that enforce within-gender cross-language splits and within-language cross-gender splits. Section 4 reports per-condition training volumes (~70-80% of available utterances per split), Pearson correlations with standard deviations across 5-fold cross-validation, and paired statistical tests (Wilcoxon signed-rank, p<0.01 for cross-language effects). Recording-condition controls include per-session z-normalization and explicit checks for microphone and room consistency across languages. To make these elements visible at the abstract level without exceeding length limits, we will revise the abstract to reference the balanced corpus design and direct readers to Sections 3-4 for data volumes, significance testing, and imbalance controls. This addresses the concern directly. revision: yes

Circularity Check

0 steps flagged

No significant circularity: purely empirical measurement study

full rationale

The paper reports direct empirical benchmarks of acoustic-to-articulatory inversion models on the FROST-EMA corpus, measuring Pearson correlations under explicitly defined cross-gender and cross-language mismatch protocols. No derivation chain, equations, fitted parameters presented as predictions, or load-bearing self-citations appear in the abstract or described structure. All reported declines (0.05-0.10 for gender, 0.10-0.20 for language) are measured outcomes rather than constructed results. The study is self-contained against external data benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Empirical benchmarking study that relies on standard machine-learning assumptions (Pearson correlation as evaluation metric, validity of EMA sensor data) but introduces no explicit free parameters, domain axioms, or invented entities.

pith-pipeline@v0.9.1-grok · 5701 in / 1043 out tokens · 25366 ms · 2026-06-26T15:22:05.623240+00:00 · methodology

0 comments
read the original abstract

Acoustic-to-articulatory inversion (AAI) remains challenging under domain shifts where changes in speaker attributes and cross-language conditions often degrade performance. We conduct a systematic evaluation under such shifts and establish baseline benchmarks on FROST-EMA, a Finnish-Russian bilingual EMA corpus. FROST-EMA addresses the English bias and limited speaker diversity of existing resources. We benchmark (i) articulatory targets (raw EMA coordinates vs tract variables), (ii) acoustic front-ends (MFCC vs SSL features), and (iii) inversion back-ends (BiLSTM vs a lightweight attention-based sequence model). We further define evaluation protocols for cross-gender transfer (within language) and cross-language transfer (within gender). The results indicate that cross-gender mismatch introduces moderate Pearson correlation declines (approximately 0.05 to 0.10) relative to the in-domain baseline, whereas cross-language mismatch causes larger drops (approximately 0.10 to 0.20).

Figures

Figures reproduced from arXiv: 2606.20478 by Ruchi Pandey, Tomi Kinnunen.

Figure 1
Figure 1. Figure 1: Overview of experimental design for AAI: (a) FROST￾EMA speaker groups, (b) AAI, (c) Ablation study, and (d) Eval￾uation protocols. these data constraints, self-supervised learning (SSL) represen￾tations provide a useful acoustic front-end for AAI on under￾represented languages. Trained on large unlabeled speech cor￾pora, SSL front-ends can extract rich acoustic representations without requiring language-sp… view at source ↗
Figure 2
Figure 2. Figure 2: Per-channel Pearson correlation (r) for raw EMA tar￾gets across four language–gender groups (FIN-M, FIN-F, RUS￾M, RUS-F) under in-domain LOSO evaluation using Wav2Vec 2.0 with BiLSTM LA LP TTCL TBCL TDCL P e a r s o n C o r r ela tio n 0 0.1 0.2 0.3 0.4 0.5 0.6 FIN-M RUS-M FIN-F RUS-F [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Per-channel Pearson correlation (r) for tract variable targets across four language–gender groups under in-domain LOSO evaluation using Wav2Vec 2.0 with BiLSTM. sors (TT, TB, TD) consistently outperform the lip sensors (UL, LL), with vertical (Z-axis) coordinates predicted better than hor￾izontal (X-axis). This reflects stronger acoustic coupling of ver￾tical tongue displacement to formant structure. ULz r… view at source ↗
Figure 5
Figure 5. Figure 5: Predicted vs. ground-truth tract variable trajecto￾ries for a representative FIN-M test utterance under in-domain (FIN-M LOSO) and cross-language (RUS-M → FIN-M) condi￾tions, using Wav2Vec 2.0 with BiLSTM. matched conditions. All three tongue TVs achieve substan￾tially higher in-domain correlations than lip TVs. Under cross￾language transfer, the tongue CL predictions visibly diverge from the ground truth … view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

41 extracted references · 4 canonical work pages · 1 internal anchor

  1. [1]

    AAI is an ill-posed problem, as similar acoustic realizations can arise from different articulatory configurations [3]

    Introduction Acoustic-to-articulatory inversion (AAI) refers to estimating time-varying vocal-tract movements from the acoustic speech signal [1, 2]. AAI is an ill-posed problem, as similar acoustic realizations can arise from different articulatory configurations [3]. This one-to-many mapping from acoustics to articulation makes AAI a highly non-linear i...

  2. [2]

    RUS-F (2) Experiments

  3. [3]

    Combined (L+G) Front-end

  4. [4]

    Beyond Speaker Independence: Evaluating Cross-Lingual Acoustic-to-Articulatory Inversion Across Finnish and Russian

    Tract variables (a) FROST-EMA (L1 only) (d) Evaluation protocols (c) Ablation study (b) AAI: speech to articulation Figure 1:Overview of experimental design for AAI: (a) FROST- EMA speaker groups, (b) AAI, (c) Ablation study, and (d) Eval- uation protocols. these data constraints, self-supervised learning (SSL) represen- tations provide a useful acoustic ...

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  5. [5]

    Dataset FROST-EMA is a recently collected electromagnetic articulog- raphy corpus designed to address the limited coverage and En- glish bias of existing parallel acoustic–EMA resources [25]. It contains recordings from 18 bilingual speakers (11 native Finnish, 7 native Russian; 8 female, 10 male) under three con- trolled speaking conditions: native langu...

  6. [6]

    Preprocessing We extract articulatory trajectories from head-movement- corrected AG501 positional recordings sampled at 1250 Hz

    Proposed AAI Model 3.1. Preprocessing We extract articulatory trajectories from head-movement- corrected AG501 positional recordings sampled at 1250 Hz. Channel-to-sensor mapping follows the default FROST-EMA protocol [25], with speaker-specific overrides where sensor re- assignment is documented. We preprocess each of the 10 ar- ticulatory channels indep...

  7. [7]

    Experiments and Results 4.1. In-domain speaker-independent baselines We first establish in-domain baselines using leave-one-speaker- out (LOSO) evaluation within each language–gender group (FIN-M, FIN-F, RUS-M, and RUS-F). LOSO ensures that the test speaker is never seen during training, providing a rigor- ous measure of cross-speaker generalization. Vali...

  8. [8]

    The key findings are as follows: • Language mismatch (∆r≈0.10–0.20) degrades AAI more than gender mismatch (∆r≈0.05–0.10) and combined shifts (L+G) produce the largest drops

    Conclusion We presented the first systematic, speaker-independent AAI benchmarks on non-English bilingual FROST-EMA corpus of 18 Finnish-Russian speakers. The key findings are as follows: • Language mismatch (∆r≈0.10–0.20) degrades AAI more than gender mismatch (∆r≈0.05–0.10) and combined shifts (L+G) produce the largest drops. This suggests that anatomic...

  9. [9]

    Authors carefully reviewed and edited the content and take full responsibility for the publication

    Generative AI Use Disclosure Generative AI tools were used for minor editing and polishing of the manuscript. Authors carefully reviewed and edited the content and take full responsibility for the publication

  10. [10]

    Inver- sion of articulatory-to-acoustic transformation in the vocal tract by a computer-sorting technique,

    B. S. Atal, J. J. Chang, M. V . Mathews, and J. W. Tukey, “Inver- sion of articulatory-to-acoustic transformation in the vocal tract by a computer-sorting technique,”The Journal of the Acoustical Society of America, vol. 63, no. 5, pp. 1535–1555, 1978

    1978

  11. [11]

    Towards an articulatory phonology,

    C. P. Browman and L. M. Goldstein, “Towards an articulatory phonology,”Phonology, vol. 3, pp. 219–252, 1986

    1986

  12. [12]

    Formant frequencies of some fixed-mandible vowels and a model of speech motor pro- gramming by predictive simulation,

    B. Lindblom, J. Lubker, and T. Gay, “Formant frequencies of some fixed-mandible vowels and a model of speech motor pro- gramming by predictive simulation,”The Journal of the Acousti- cal Society of America, vol. 62, no. S1, pp. S15–S15, 1977

    1977

  13. [13]

    An empirical investigation of the nonuniqueness in the acoustic-to-articulatory mapping

    C. Qin and M. ´A. Carreira-Perpi˜n´an, “An empirical investigation of the nonuniqueness in the acoustic-to-articulatory mapping.” in Interspeech, vol. 1, 2007, pp. 74–77

    2007

  14. [14]

    Articulatory information for noise robust speech recogni- tion,

    V . Mitra, H. Nam, C. Y . Espy-Wilson, E. Saltzman, and L. Gold- stein, “Articulatory information for noise robust speech recogni- tion,”IEEE Transactions on Audio, Speech, and Language Pro- cessing, vol. 19, no. 7, pp. 1913–1924, 2010

    1913

  15. [15]

    Automatic speech recognition using articulatory features from subject-independent acoustic-to- articulatory inversion,

    P. K. Ghosh and S. Narayanan, “Automatic speech recognition using articulatory features from subject-independent acoustic-to- articulatory inversion,”The Journal of the Acoustical Society of America, vol. 130, no. 4, pp. EL251–EL257, 2011

    2011

  16. [16]

    Articulatory con- trol of hmm-based parametric speech synthesis using feature- space-switched multiple regression,

    Z. H. Ling, K. Richmond, and J. Yamagishi, “Articulatory con- trol of hmm-based parametric speech synthesis using feature- space-switched multiple regression,”IEEE Transactions on Au- dio, Speech, and Language Processing, vol. 21, no. 1, pp. 207– 219, 2013

    2013

  17. [17]

    A real-time articula- tory visual feedback approach with target presentation for second language pronunciation learning,

    A. Suemitsu, J. Dang, T. Ito, and M. Tiede, “A real-time articula- tory visual feedback approach with target presentation for second language pronunciation learning,”The Journal of the Acoustical Society of America, vol. 138, no. 4, pp. EL382–EL387, 2015

    2015

  18. [18]

    Acoustic-articulatory modeling with the trajectory hmm,

    L. Zhang and S. Renals, “Acoustic-articulatory modeling with the trajectory hmm,”IEEE Signal Processing Letters, vol. 15, pp. 245–248, 2008

    2008

  19. [19]

    On smoothing articulatory trajec- tories obtained from gaussian mixture model based acoustic-to- articulatory inversion,

    P. K. Ghosh and S. Narayanan, “On smoothing articulatory trajec- tories obtained from gaussian mixture model based acoustic-to- articulatory inversion,”The Journal of the Acoustical Society of America, vol. 134, no. 2, pp. EL258–EL264, 2013

    2013

  20. [20]

    Articulatory movement prediction using deep bidirectional long short-term memory based recurrent neural networks and word/phone embeddings

    P. Zhu, L. Xie, and Y . Chen, “Articulatory movement prediction using deep bidirectional long short-term memory based recurrent neural networks and word/phone embeddings.” inInterspeech, 2015, pp. 2192–2196

    2015

  21. [21]

    Low resource acoustic-to-articulatory inversion using bi-directional long short term memory

    A. Illa and P. K. Ghosh, “Low resource acoustic-to-articulatory inversion using bi-directional long short term memory.” inInter- speech, 2018, pp. 3122–3126

    2018

  22. [22]

    Raw speech-to-articulatory inversion by temporal filtering and decima- tion

    A. S. Shahrebabaki, S. M. Siniscalchi, T. Svendsenet al., “Raw speech-to-articulatory inversion by temporal filtering and decima- tion.” inInterspeech, 2021, pp. 1184–1188

    2021

  23. [23]

    Estimat- ing articulatory movements in speech production with transformer networks,

    S. Udupa, A. Roy, A. Singh, A. Illa, and P. K. Ghosh, “Estimat- ing articulatory movements in speech production with transformer networks,” inProc. Interspeech, 2021, pp. 1154–1158

    2021

  24. [24]

    Speaker-independent acoustic-to- articulatory inversion through multi-channel attention discrimina- tor,

    W.-J. Chung and H.-G. Kang, “Speaker-independent acoustic-to- articulatory inversion through multi-channel attention discrimina- tor,”arXiv preprint arXiv:2406.17329, 2024

    work page arXiv 2024

  25. [25]

    The mocha-timit articulatory database,

    A. Wrench, “The mocha-timit articulatory database,” 1999. [Online]. Available: https://www.cstr.ed.ac.uk/research/projects/ artic/mocha.html

    1999

  26. [26]

    Announcing the electro- magnetic articulography (day 1) subset of the mngu0 articulatory corpus,

    K. Richmond, P. Hoole, and S. King, “Announcing the electro- magnetic articulography (day 1) subset of the mngu0 articulatory corpus,” inInterspeech 2011: 12th Annual Conference of the In- ternational Speech Communication Association. International Speech Communication Association, 2011, pp. 1505–1508

    2011

  27. [27]

    Analysis of acoustic-to-articulatory speech inversion across different accents and languages,

    M. Wieling, G. Sivaraman, and C. Espy-Wilson, “Analysis of acoustic-to-articulatory speech inversion across different accents and languages,” inProceedings of INTERSPEECH, 2017, pp. 974–978

    2017

  28. [28]

    The impact of cross lan- guage on acoustic-to-articulatory inversion and its influence on articulatory speech synthesis,

    A. Illa, A. Nair, and P. K. Ghosh, “The impact of cross lan- guage on acoustic-to-articulatory inversion and its influence on articulatory speech synthesis,” inIEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2022, pp. 8267–8271

    2022

  29. [29]

    Combining language corpora in a japanese electromagnetic articulography database for acoustic-to-articulatory inversion,

    T. Yan, K. Maekawa, Y . Nota, and M. Hirata, “Combining language corpora in a japanese electromagnetic articulography database for acoustic-to-articulatory inversion,” inProc. INTER- SPEECH, vol. 2023, 2023, pp. 1464–1467

    2023

  30. [30]

    Exploring self-supervised speech representa- tions for cross-lingual acoustic-to-articulatory inversion,

    Y . Hao, R. Amooie, W. de Vries, T. Tienkamp, R. van Noord, and M. Wieling, “Exploring self-supervised speech representa- tions for cross-lingual acoustic-to-articulatory inversion,” inIn- terspeech 2024. ISCA, 2024, pp. 4603–4607

    2024

  31. [31]

    Self-supervised models of speech infer universal articulatory kinematics,

    C. J. Cho, A. Mohamed, A. W. Black, and G. K. Anumanchipalli, “Self-supervised models of speech infer universal articulatory kinematics,” inIEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2024, pp. 12 061– 12 065

    2024

  32. [32]

    Enhancing acoustic-to-articulatory inversion with multi-target pretraining for low-resource settings,

    J. Bandekar and P. K. Ghosh, “Enhancing acoustic-to-articulatory inversion with multi-target pretraining for low-resource settings,” inProc. Interspeech 2025, 2025, pp. 5588–5592

    2025

  33. [33]

    The presence of articulatory representations in self-supervised speech models: A cross-linguistic perspective,

    A. Pollio San Pedro, “The presence of articulatory representations in self-supervised speech models: A cross-linguistic perspective,” Master’s thesis, It¨a-Suomen yliopisto, 2026

    2026

  34. [34]

    Frost-ema: Finnish and russian oral speech dataset of electromagnetic articulography measure- ments with l1, l2 and imitated l2 accents,

    S. Hopponen, T. Kinnunen, A. Nikolaev, R. G. Hautam ¨aki, L. Tavi, and E. Meister, “Frost-ema: Finnish and russian oral speech dataset of electromagnetic articulography measure- ments with l1, l2 and imitated l2 accents,”arXiv preprint arXiv:2506.08981, 2025

    work page arXiv 2025

  35. [35]

    The geometric vocal tract variables controlled for vowel production: proposals for con- straining acoustic-to-articulatory inversion,

    L.-J. Bo ¨e, P. Perrier, and G. Bailly, “The geometric vocal tract variables controlled for vowel production: proposals for con- straining acoustic-to-articulatory inversion,”Journal of Phonetics, vol. 20, no. 1, pp. 27–38, 1992

    1992

  36. [36]

    Recovering articulatory movement from for- mant frequency trajectories using task dynamics and a genetic algorithm: Preliminary model tests,

    R. S. McGowan, “Recovering articulatory movement from for- mant frequency trajectories using task dynamics and a genetic algorithm: Preliminary model tests,”Speech Communication, vol. 14, no. 1, pp. 19–48, 1994

    1994

  37. [37]

    Unsupervised speaker adaptation for speaker indepen- dent acoustic to articulatory speech inversion,

    G. Sivaraman, V . Mitra, H. Nam, M. Tiede, and C. W. Espy- Wilson, “Unsupervised speaker adaptation for speaker indepen- dent acoustic to articulatory speech inversion,”The Journal of the Acoustical Society of America, vol. 146, no. 1, pp. 316–329, 2019

    2019

  38. [38]

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

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

    2020

  39. [39]

    Xls-r: Self-supervised cross-lingual speech representation learning at scale.arXiv preprint arXiv:2111.09296, 2021

    A. Babu, C. Wang, A. Tjandra, K. Lakhotia, Q. Xu, N. Goyal, K. Singh, P. V . Platen, Y . Saraf, J. Pinoet al., “Xls-r: Self- supervised cross-lingual speech representation learning at scale,” arXiv preprint arXiv:2111.09296, 2021

    work page arXiv 2021

  40. [40]

    Scaling speech technology to 1,000+ languages,

    V . Pratap, A. Tjandra, B. Shiet al., “Scaling speech technology to 1,000+ languages,”Journal of Machine Learning Research, vol. 25, no. 97, pp. 1–52, 2024

    2024

  41. [41]

    Towards end-to-end speech articulation and spoken language analysis using deep learning,

    T. Weise, K. C. Demir, P. A. P ´erez-Toro, T. Arias-Vergara, A. Maier, E. N ¨oth, M. Schuster, B. Heismann, and S. H. Yang, “Towards end-to-end speech articulation and spoken language analysis using deep learning,”Human-Centric Intelligent Systems, vol. 5, no. 1, pp. 103–122, 2025

    2025