Towards Language-Agnostic Speech Inversion

Carol Espy-Wilson; Liran Oren; Mark Tiede; Saba Tabatabaee; Suzanne Boyce

REVIEW 3 major objections 5 minor 27 references

Towards Language-Agnostic Speech Inversion

T0 review · 3 major / 5 minor · reviewed 2026-07-11 · grok-4.5

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

Pith's one-line read An English-trained speech inversion system recovers oral tract variables, source features, and velopharyngeal opening on French and Russian speech.

desk verdict Useful first French/Russian SI numbers on a new multi-lingual EMA set, but the language-agnostic claim is ahead of the sample size and the English-rooted TV geometry. read the letter →

arxiv 2607.05060 v1 pith:MA36R7QH submitted 2026-07-06 eess.AS

Saba Tabatabaee , Mark Tiede , Suzanne Boyce , Liran Oren , Carol Espy-Wilson This is my paper

classification eess.AS

keywords SpeechInversionArticulatoryModelingTractVariablesCross-LingualAnalysisVelopharyngealPortSourceFeaturesWavLM

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 shows that a speech inversion system trained only on American English audio paired with articulatory measurements can recover the same articulatory timing patterns from French and Russian speech it never saw in training. The system jointly estimates six oral tract variables (lip and tongue constrictions), three source features (periodicity, aperiodicity, and fundamental frequency), and a velopharyngeal opening measure that tracks nasalance. On held-out French and Russian speakers the estimated trajectories match ground-truth articulatory and nasalance data with average Pearson correlations of 0.83 and 0.74 for the oral and source parameters, and 0.89 and 0.82 for velopharyngeal opening. Because collecting articulatory data is expensive and language-specific, a system that works across languages from English training alone would let researchers and clinicians study articulatory timing in many languages from ordinary audio recordings.

What carries the argument

Multi-task speech inversion that maps WavLM-Large embeddings through Conformer layers to six oral tract variables (LA, LP, TBCL, TBCD, TTCL, TTCD), three source features (Per, Aper, F0), and a velopharyngeal tract variable, trained with a combined Pearson-correlation and RMSE loss on English XRMB and YU data.

What would settle it

Collect EMA, nasalance, and audio from a larger set of French and Russian speakers (or additional languages) and recompute PPMC; a sharp drop below the reported correlations would falsify the cross-lingual claim.

Watch

Extended reading notes

Core claim

An SI system trained exclusively on co-recorded American English speech and articulatory kinematics simultaneously estimates six oral tract variables plus three source features on previously unseen French and Russian with average PPMC scores of 0.83 and 0.74 against ground-truth measurements, and the same English-trained system also estimates a velopharyngeal tract variable that correlates with nasalance at 0.89 (French) and 0.82 (Russian).

Load-bearing premise

The geometric mapping from flesh-point sensors to tract variables, and the speech features used, transfer well enough from English to French and Russian that results on a small number of held-out speakers measure true language-agnostic performance.

Editorial extensions

If this is right

Articulatory timing patterns for oral constrictions and source control can be recovered from audio alone in French and Russian without language-specific articulatory training data.
Velopharyngeal opening and closing can be estimated across languages that differ in nasalization patterns (English vs. French nasal vowels) from an English-trained model.
Multi-lingual speech processing and clinical articulatory assessment become feasible from ordinary audio recordings rather than specialized articulatory hardware.
Future collection of more speakers and under-resourced languages can test and extend the same English-trained inversion pipeline.

Reading between the lines

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

If the transfer holds for more language families, speech inversion could serve as a low-cost proxy for field articulatory phonetics where EMA or X-ray microbeam is impractical.
The same joint oral-plus-source-plus-VP architecture may improve robustness for disordered speech or child speech once more diverse training data are added.
Performance gaps between French and Russian (and the noted noisy Russian speaker) suggest recording conditions and phonetic inventory distance remain practical limits on claimed language-agnosticism.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Punchline: they trained an English SI system (WavLM-Large + Conformers, multi-task oral TVs + Per/Aper/F0) and report average PPMC 0.83 on French and 0.74 on Russian against held-out EMA, plus VP-nasalance correlations of 0.89/0.82. That cross-lingual table is new; almost everything before this was English-only or accent work.

What is actually new is the YU multi-lingual collection (EMA + nasalance + audio for English/French/Russian) and the first reported oral-TV + source-feature + VP numbers on French and Russian. The pipeline itself is a clean extension of their prior line, not a redesign. They do the basics right: speaker-independent splits, a direct XRMB comparison that edges their earlier model (0.86 vs 0.85), and trajectory plots that look plausible across languages. The French oral-TV result is the most credible piece of evidence for partial transfer. They also flag the noisy Russian recording themselves.

Soft spots are real and proportional, not fatal. Non-English N is tiny (4 French, 3 Russian; only one Russian for nasalance), with no speaker-level scores or intervals. Russian is already pulled down by that noisy speaker. The flesh-point-to-TV geometry and WavLM features are English-derived; treating pooled PPMC as proof of language-agnostic recovery overclaims what those cohorts can support. VP training on XRMB uses pseudo-labels from their own prior nasal SI system, so that path is less independent than the oral-TV results. No code or data release makes the numbers hard to re-run.

This is for people who need audio-only proxies for articulatory timing in multi-lingual or clinical settings. It is solid subfield progress, not a methods leap. I would send it to peer review: the French numbers and the new corpus deserve referee time, with the expectation that reviewers will ask for more speakers, variance, and a quieter title. Worth engaging if you work on SI or cross-lingual phonetics; otherwise the tables are enough.

Referee Report

3 major / 5 minor

Summary. The paper presents a multi-task speech inversion (SI) system that maps audio to six oral tract variables (LA, LP, TBCL, TBCD, TTCL, TTCD) plus three source features (Per, Aper, F0), and optionally a velopharyngeal (VP) TV, using WavLM-Large embeddings, Conformer layers, and a combined Pearson+RMSE loss. The system is trained only on American English (XRMB + YU English splits) and evaluated on held-out English as well as previously unseen French and Russian speakers from a newly collected multi-lingual EMA/nasalance corpus. Table 3 reports average PPMC of 0.86 (XRMB English), 0.85 (YU English), 0.83 (French), and 0.74 (Russian) across the nine oral+source parameters; Table 4 reports VP-TV vs. nasalance PPMC of 0.92 (English), 0.89 (French), and 0.82 (Russian). Qualitative trajectory plots and a comparison to a prior SI model on XRMB are also provided.

Significance. If the cross-lingual numbers hold under more rigorous scrutiny, the work supplies a practical English-trained SI pipeline that recovers both oral constriction timing and source/VP information for French and Russian without language-specific articulatory training data. That would be useful for multi-lingual articulatory research, clinical assessment of nasality, and low-resource settings where EMA is unavailable. Strengths include speaker-independent splits, simultaneous multi-parameter estimation, a modest improvement over the prior SI baseline on XRMB (0.86 vs 0.85), and the collection of a multi-lingual EMA+nasalance corpus. The language-agnostic framing is therefore of genuine interest to the speech-inversion and articulatory-phonetics communities, provided the small held-out cohorts and English-derived TV geometry are shown to be representative.

major comments (3)

Table 3 and Sec. 3.1: the central cross-lingual claim rests on pooled PPMC averages over only 4 French and 3 Russian speakers (and, for VP in Table 4 / Sec. 3.3, only 3 French and 1 Russian). No per-speaker scores, standard errors, or bootstrap intervals are reported. The text itself attributes the Russian drop (especially Per/F0) to one noisy recording. Without speaker-level or uncertainty statistics, it is not possible to judge whether 0.83 / 0.74 (and 0.89 / 0.82 for VP) are stable estimates of language-agnostic performance or are driven by individual speakers/recording conditions.
Sec. 2.1.1–2.1.2 and the geometric transform of [21]: oral TVs are obtained by an English-derived flesh-point-to-TV mapping that is assumed language-independent. The manuscript never tests whether that mapping remains faithful for French or Russian articulatory targets (e.g., by comparing alternative normalizations or reporting residual anatomical variance). If the transform itself is English-centric, the reported PPMC on French/Russian does not cleanly support a language-agnostic claim.
Sec. 3.3: VP training for XRMB uses nasalance pseudo-labels generated by the authors’ own prior nasal SI system [8]. While the final evaluation on YU French/Russian uses measured nasalance, the training target for a large fraction of the data is model-derived. The manuscript should quantify how sensitive the VP results are to these pseudo-labels (e.g., train without XRMB nasalance or report agreement between the prior nasal SI and true nasalance on YU English) so that the cross-lingual VP numbers are not partly circular.

minor comments (5)

Table 2 / Sec. 2.1.2: the YU French and Russian test sets are small (59 min / 35 min). Explicitly state total hours and number of utterances per language in the abstract or introduction so readers can gauge statistical power immediately.
Eq. (1) and Sec. 2.2: the mixing weight α = 0.2 is described as “empirically chosen.” A short ablation (or at least the range explored) would strengthen reproducibility.
Figure 2 and Figure 4: axis labels and units for the TV / nasalance trajectories are hard to read in the manuscript rendering; ensure they remain legible in the final version.
Sec. 3.1: the claim that WavLM-Large “outperforms” other SSL models is supported only by citations; a one-sentence confirmation that the same ranking holds on the present multi-lingual test sets would be useful.
Typographical consistency: “product–moment” vs “product-moment,” and occasional spacing around PPMC values, should be standardized.

Circularity Check

1 steps flagged · score 2.0 of 10

Mild self-citation for XRMB VP pseudo-labels via authors' prior system [8]; main oral-TV/SF cross-lingual PPMC rests on independent EMA ground truth.

self citation load bearing [Section 3.3 (VP TV estimation paragraph)]
"To address this limitation, we employed the nasal SI system proposed in [8] to estimate nasalance ground-truth values for the XRMB dataset from the speech signals, effectively retrofitting XRMB with nasalance measurements. This approach is inspired by [8], which demonstrated that nasalance estimates generated by a nasal SI system for the XRMB dataset can be used to train an SI system effectively, as validated by comparison with a real ground-truth nasalance dataset."

Training targets for the VP TV component on the XRMB portion of the combined English training set are generated by the authors' own prior model rather than independent measurements. Although final PPMC scores (Table 4) are computed against real nasalance on held-out YU speakers, the multi-task model that produces those estimates was partially trained on self-generated labels, creating a mild self-citation dependency confined to the VP path.

full rationale

The core claims (Table 3 PPMC of 0.83 French / 0.74 Russian for six oral TVs + Per/Aper/F0) train on English XRMB+YU EMA-derived tract variables and APP-derived source features, then evaluate against independent co-recorded EMA ground truth on held-out French and Russian speakers. The geometric flesh-point-to-TV transform is applied uniformly to define targets for both train and test, so correlations are not forced by construction. WavLM features and the APP detector are external. The only self-citation dependency is auxiliary: Section 3.3 retrofits XRMB (which lacks nasalance) with pseudo-labels from the authors' own prior nasal SI system [8] before multi-task training that includes VP TV; evaluation of VP, however, uses real nasalance measurements on YU held-out speakers (Table 4). This is standard pseudo-labeling, not a definitional loop or fitted-as-prediction, and does not underwrite the primary language-agnostic oral-TV results. No uniqueness theorems, ansatz smuggling, or renaming of known results appear. The derivation is therefore empirical and largely self-contained; the single self-citation is minor and non-load-bearing for the strongest claims.

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

The central cross-lingual claim rests on domain assumptions that tract-variable geometry and SSL speech features transfer across languages, plus a few hand-chosen training hyperparameters. No new physical entities are postulated. Free parameters are standard ML knobs; the load-bearing scientific assumptions are about shared articulatory acoustics and the validity of small held-out non-English cohorts.

free parameters (3)

loss mixing weight α = 0.2
α=0.2 in L=(1−PC)+αRMSE is stated as empirically chosen for optimal performance; it directly weights the multi-task objective used for all reported PPMC scores.
WavLM layer weighted-sum coefficients
A learned weighted sum over all 25 WavLM-Large layers produces the input representation; these weights are free parameters fitted during training and affect every estimate.
AdamW learning rate and schedule patience = 5e-4 / patience 5+8
Initial LR 5e-4, weight decay 1e-3, plateau patience 5, early-stop patience 8 are hand-set training controls that determine the final model used for cross-lingual evaluation.

assumptions (4)

domain assumption Oral tract variables obtained by geometric transform of midsagittal flesh points are language-independent relative measures of constriction location and degree.
Invoked in Sections 2.1.1–2.1.2 when the same TV definitions from English XRMB/YU are applied without re-derivation to French and Russian EMA.
domain assumption WavLM-Large representations trained primarily on large English-heavy corpora still encode articulatory timing usable for non-English SI.
Core of the feature pipeline (Section 2.2); cross-lingual PPMC is taken as evidence rather than independently validated feature transfer.
ad hoc to paper Nasalance estimated by a prior nasal SI system is a valid training target for XRMB utterances that lack measured nasalance.
Section 3.3 explicitly retrofits XRMB with pseudo-nasalance from [8] to enable multi-task VP training; this is a paper-specific modeling choice, not a standard external fact.
domain assumption APP detector outputs (Per, Aper, F0) are reliable ground-truth source features across the languages tested.
Used for both English training and French/Russian evaluation (Section 2.2); no language-specific validation of APP is reported.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Towards Language-Agnostic Speech Inversion." pith.science (2026). https://pith.science/paper/MA36R7QH

@misc{pith2026260705060,
  author       = {Pith},
  title        = {Pith review of: Towards Language-Agnostic Speech Inversion},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MA36R7QH}},
  note         = {Machine review of arXiv:2607.05060}
}

read the original abstract

Characteristic timing patterns are reflected in the acoustic speech signal, encompassing both vocal tract configuration and acoustic excitation. Previous studies have demonstrated that speech inversion (SI) systems can recover these timing patterns from speech, including oral tract variables (tongue and lip constrictions) and source information such as periodic and aperiodic energies and fundamental frequency. In this study, we develop an SI system that simultaneously estimates oral tract variables and three source information parameters trained on co-recorded American English speech audio and articulatory kinematics and investigate cross-linguistic generalizability by evaluating performance on previously unseen languages. Pearson product-moment correlation scores of 0.83 and 0.74 were achieved on untrained French and Russian respectively, across oral tract variables and source information when comparing estimated data with ground-truth measurements.

Figures

Figures reproduced from arXiv: 2607.05060 by the authors.

**Figure 2.** shows the estimated TBCD, TTCD, and LA TVs produced by the SI system, together with the corresponding ground-truth measurements, for randomly selected English, Russian and French utterances from the YU test set. A detailed comparison between the SI system estimates and the groundtruth values shows that, despite being trained exclusively on English speech, the SI system effectively captures the articulatory pattern… view at source ↗

**Figure 3.** shows a randomly selected example from the XRMB test set, illustrating the nasalance ground-truth and the SI system estimates for the utterances “It’s seam ore Sid” and “It’s see more Sid”. For the instances of the phoneme /m/ in “seam” and “more” the estimated nasalance exhibits a peak in each case, which is consistent with the nasalance ground-truth. In the utterance “It’s seam ore Sid”, the nasalance peak begin… view at source ↗

Figures from the paper (1 more)

**Figure 4.** Figure 4: Waveforms and spectrograms of French and Russian utterances, alongside comparisons between ground-truth nasalance and the VP TV estimated by the SI system. 4. Conclusions and future work In this paper, we investigated the performance of an SI system trained on Englis…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

27 extracted references · 2 linked inside Pith

[21]

Unsuper- vised acoustic-to-articulatory inversion with variable vocal tract anatomy

Y . Sun, Q. Huang, X. Wu, and M. Perception, “Unsuper- vised acoustic-to-articulatory inversion with variable vocal tract anatomy.” inINTERSPEECH, 2022, pp. 4656–4660

2022
[8]

A cross-language study of voicing in initial stops: Acoustical measurements,

L. Lisker and A. S. Abramson, “A cross-language study of voicing in initial stops: Acoustical measurements,”Word, vol. 20, no. 3, pp. 384–422, 1964

1964
[1]

Direct observation of articulation provides the most reliable evidence of language-specific temporal patterns

Introduction Speech articulation is a complex process that requires precise, temporally coordinated movements of multiple articulators, in- cluding the lips, tongue, jaw, velum, and glottis [1]. Direct observation of articulation provides the most reliable evidence of language-specific temporal patterns. However, collecting ar- ticulatory data requires sp...

arXiv 2026
[2]

Dataset description and pre-processing 2.1.1

Methodology 2.1. Dataset description and pre-processing 2.1.1. XRMB dataset The University of Wisconsin X-Ray Microbeam (XRMB) dataset [20] provides speech recordings accompanied by artic- ulatory trajectory data collected using a rasterized X-ray mi- crobeam system tracking pellets placed midsagittally on the oral articulators. The dataset includes recor...
[3]

It’s seam ore Sid

Results and discussion 3.1. Performance of the SI system Table 3 summarizes the performance of the proposed SI sys- tem on the XRMB and YU test sets, evaluated using Pear- son product–moment correlation (PPMC) scores, which mea- sure the alignment between estimated and ground-truth artic- ulatory parameters. On the XRMB test set, the proposed SI system ac...
[4]

We developed an SI system capable of simultaneously estimating oral tract variables and three source features, and evaluated its performance on both English and non-English speech

Conclusions and future work In this paper, we investigated the performance of an SI sys- tem trained on English data when applied to non-English lan- guages, including French and Russian. We developed an SI system capable of simultaneously estimating oral tract variables and three source features, and evaluated its performance on both English and non-Engl...
[5]

Generative AI Use Disclosure Generative AI tools were used solely for spelling and grammar correction
[6]

K. N. Stevens,Acoustic phonetics. MIT press, 2000, vol. 30

2000

Show all 27 references

[7]

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
[9]

Towards an articulatory phonology,

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

1986
[10]

A dynamical approach to gestural patterning in speech production,

E. L. Saltzman and K. G. Munhall, “A dynamical approach to gestural patterning in speech production,”Ecological psychology, vol. 1, no. 4, pp. 333–382, 1989

1989
[11]

Phonics: A large phoneme-grapheme frequency count revised,

E. Fry, “Phonics: A large phoneme-grapheme frequency count revised,”Journal of Literacy Research, vol. 36, no. 1, pp. 85–98, 2004

2004
[12]

Speaker-independent speech inversion for recovery of velopharyngeal port constriction de- gree,

Y . M. Siriwardena, S. E. Boyce, M. K. Tiede, L. Oren, B. Fletcher, M. Stern, and C. Y . Espy-Wilson, “Speaker-independent speech inversion for recovery of velopharyngeal port constriction de- gree,”The Journal of the Acoustical Society of America, vol. 156, no. 2, pp. 1380–1390, 2024

2024
[13]

Enhancing Acoustic-to-Articulatory Speech Inversion by Incor- porating Nasality,

S. Tabatabaee, S. Boyce, L. Oren, M. Tiede, and C. Espy-Wilson, “Enhancing Acoustic-to-Articulatory Speech Inversion by Incor- porating Nasality,” inInterspeech 2025, 2025, pp. 325–329

2025
[14]

The secret source: In- corporating source features to improve acoustic-to-articulatory speech inversion,

Y . M. Siriwardena and C. Espy-Wilson, “The secret source: In- corporating source features to improve acoustic-to-articulatory speech inversion,” inICASSP 2023-2023 IEEE International Con- ference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2023, pp. 1–5

2023
[15]

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

W.-N. Hsu, B. Bolte, Y .-H. H. Tsai, K. Lakhotia, R. Salakhutdi- nov, and A. Mohamed, “Hubert: Self-supervised speech represen- tation learning by masked prediction of hidden units,”IEEE/ACM transactions on audio, speech, and language processing, vol. 29, pp. 3451–3460, 2021

2021
[16]

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
[17]

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. Xiaoet 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
[18]

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
[19]

Ev- idence of vocal tract articulation in self-supervised learning of speech,

C. J. Cho, P. Wu, A. Mohamed, and G. K. Anumanchipalli, “Ev- idence of vocal tract articulation in self-supervised learning of speech,” inICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2023, pp. 1–5

2023
[20]

Acoustic to articulatory speech inversion for children with velopharyngeal insufficiency,

S. Tabatabaee, S. Boyce, L. Oren, M. Tiede, and C. Espy-Wilson, “Acoustic to articulatory speech inversion for children with velopharyngeal insufficiency,”arXiv preprint arXiv:2509.09489, 2025

2025 arXiv
[22]

Speaker-independent acoustic-to-articulatory speech inversion,

P. Wu, L.-W. Chen, C. J. Cho, S. Watanabe, L. Goldstein, A. W. Black, and G. K. Anumanchipalli, “Speaker-independent acoustic-to-articulatory speech inversion,” inICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2023, pp. 1–5

2023
[23]

Towards noise-robust speech inversion through multi-task learning with speech enhancement,

S. Tabatabaee and C. Espy-Wilson, “Towards noise-robust speech inversion through multi-task learning with speech enhancement,” arXiv preprint arXiv:2601.14516, 2026

2026
[24]

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
[25]

Speech production database user’s handbook,

J. R. Westbury, “Speech production database user’s handbook,” IEEE Personal Communications-IEEE Pers. Commun., vol. 0, no, 1994

1994
[26]

Improv- ing speech inversion through self-supervised embeddings and en- hanced tract variables,

A. A. Attia, Y . M. Siriwardena, and C. Espy-Wilson, “Improv- ing speech inversion through self-supervised embeddings and en- hanced tract variables,” in2024 32nd European Signal Processing Conference (EUSIPCO). IEEE, 2024, pp. 306–310

2024
[27]

Use of temporal information: Detection of periodicity, aperiodicity, and pitch in speech,

O. Deshmukh, C. Y . Espy-Wilson, A. Salomon, and J. Singh, “Use of temporal information: Detection of periodicity, aperiodicity, and pitch in speech,”IEEE Transactions on Speech and Audio Processing, vol. 13, no. 5, pp. 776–786, 2005

2005

Pith tools

Reviewed July 11, 2026 · model on record in the stance chip above.

[1] [21]

Unsuper- vised acoustic-to-articulatory inversion with variable vocal tract anatomy

Y . Sun, Q. Huang, X. Wu, and M. Perception, “Unsuper- vised acoustic-to-articulatory inversion with variable vocal tract anatomy.” inINTERSPEECH, 2022, pp. 4656–4660

2022

[2] [8]

A cross-language study of voicing in initial stops: Acoustical measurements,

L. Lisker and A. S. Abramson, “A cross-language study of voicing in initial stops: Acoustical measurements,”Word, vol. 20, no. 3, pp. 384–422, 1964

1964

[3] [1]

Direct observation of articulation provides the most reliable evidence of language-specific temporal patterns

Introduction Speech articulation is a complex process that requires precise, temporally coordinated movements of multiple articulators, in- cluding the lips, tongue, jaw, velum, and glottis [1]. Direct observation of articulation provides the most reliable evidence of language-specific temporal patterns. However, collecting ar- ticulatory data requires sp...

arXiv 2026

[4] [2]

Dataset description and pre-processing 2.1.1

Methodology 2.1. Dataset description and pre-processing 2.1.1. XRMB dataset The University of Wisconsin X-Ray Microbeam (XRMB) dataset [20] provides speech recordings accompanied by artic- ulatory trajectory data collected using a rasterized X-ray mi- crobeam system tracking pellets placed midsagittally on the oral articulators. The dataset includes recor...

[5] [3]

It’s seam ore Sid

Results and discussion 3.1. Performance of the SI system Table 3 summarizes the performance of the proposed SI sys- tem on the XRMB and YU test sets, evaluated using Pear- son product–moment correlation (PPMC) scores, which mea- sure the alignment between estimated and ground-truth artic- ulatory parameters. On the XRMB test set, the proposed SI system ac...

[6] [4]

We developed an SI system capable of simultaneously estimating oral tract variables and three source features, and evaluated its performance on both English and non-English speech

Conclusions and future work In this paper, we investigated the performance of an SI sys- tem trained on English data when applied to non-English lan- guages, including French and Russian. We developed an SI system capable of simultaneously estimating oral tract variables and three source features, and evaluated its performance on both English and non-Engl...

[7] [5]

Generative AI Use Disclosure Generative AI tools were used solely for spelling and grammar correction

[8] [6]

K. N. Stevens,Acoustic phonetics. MIT press, 2000, vol. 30

2000