Pith. sign in

REVIEW 4 major objections 4 minor 41 references

ClaritySpeech: Dementia Obfuscation in Speech

T0 review · 4 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read ClaritySpeech claims that dementia markers in speech can be obfuscated by transcribing, rewriting, and resynthesizing the audio in the speaker's own voice, reducing automatic dementia-detection F1 from 0.70 to 0.59 on ADReSS and from 0.63…

desk verdict A useful applied system with a real WER win, but the headline privacy gain is probably mostly TTS domain shift until a healthy-speech control and speaker-disjoint split are shown. read the letter →

arxiv 2507.09282 v1 pith:BOGN4SBI submitted 2025-07-12 cs.CL cs.CRcs.LGcs.SDeess.AS

Dominika Woszczyk , Ranya Aloufi , Soteris Demetriou This is my paper
classification cs.CLcs.CRcs.LGcs.SDeess.AS
keywords dementiaspeechobfuscationprivacypreservationzero-shottext-to-speechvoicecloningautomaticrecognitionAlzheimer'sdetectionaccessibility
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 establish that dementia-affected speech can be obfuscated by a three-stage pipeline: transcribing it, rewriting the transcript to remove cognitive-impairment markers, and resynthesizing it in the original voice with zero-shot text-to-speech. It reports that this drops the mean adversarial F1 score (a balanced measure of automatic detection accuracy) from 0.70 to 0.59 on ADReSS and from 0.63 to about 0.55–0.56 on ADReSSo, while keeping speaker similarity near 0.50 and improving ASR word error rate from 0.73 to 0.08. The paper's key claim is that zero-shot TTS acts as a normalizer: it transfers timbre but smooths out the atypical pauses, pitch instability, and disfluencies that leak dementia status. A sympathetic reader would care because it offers a privacy intervention for a vulnerable group that also improves the accessibility of speech technology.

What carries the argument

The load-bearing mechanism is what the paper calls the normalizing nature of zero-shot text-to-speech. A TTS model trained on large amounts of fluent, diverse speech can clone a new speaker's timbre from a short reference prompt, but it reproduces atypical speech poorly, so it smooths out dementia-related acoustic granularities such as pauses, unstable pitch, and irregular energy. The text obfuscator, DiDOTS, removes linguistic dementia markers before synthesis, and the ASR stage converts audio into the text that both modules need. The framework's hypothesis is that this weakness of TTS is exactly what makes the obfuscation work.

What would settle it

Resynthesize the original healthy-control speech through the same ClaritySpeech pipeline and measure the audio adversary's F1 on those synthetic healthy samples; if the pipeline already makes healthy speech look non-dementia, then the reported privacy drop is largely a synthesis artifact rather than dementia-marker removal. A separate check is to train a classifier only on synthesized versions of healthy and dementia speech and see whether the two classes remain separable at near-original accuracy.

Watch

Extended reading notes

Core claim

The central discovery the paper argues for is that a holistic pipeline, not any single module, can conceal dementia in speech: Whisper transcribes the disordered audio, a DiDOTS-derived text obfuscator paraphrases away linguistic dementia cues, and a zero-shot TTS model resynthesizes the cleaned text in the reference speaker's voice. Across static and adaptive adversaries and audio, text, and fusion modalities, the total mean F1 drops by 16% on ADReSS and 10% on ADReSSo while speaker similarity stays around 0.50 and the predicted speech-quality score UTMOS rises from about 1.65 to about 2.15. Ablations attribute the privacy gain to both text obfuscation and the normalizing TTS, with the TTS doing most of the acoustic work: removing text obfuscation raises text-modality leakage, and removing the TTS destroys speaker similarity.

Load-bearing premise

The privacy numbers assume the F1 drop comes from removing dementia markers rather than from the fact that any newly synthesized, paraphrased audio is out-of-distribution for classifiers trained on original recordings.

Editorial extensions

If this is right

  • People with dementia could post recordings, use voice banking, or interact with voice services with a lower risk of being automatically flagged as vulnerable.
  • ASR accessibility improves substantially: word error rate on dementia speech drops from 0.73 to 0.08 on ADReSS, making voice assistants and transcription tools more usable for atypical speech.
  • The pipeline works without per-speaker fine-tuning, which matters for low-resource clinical datasets where collecting hours of dementia speech is impractical.
  • Speaker identity is only partially preserved, at roughly 50% cosine similarity, so the framework is a privacy–utility trade-off rather than a full anonymizer.
  • The privacy gains partly survive an adaptive adversary who trains on obfuscated samples, but the remaining leakage is still substantial, with total mean F1 around 0.55–0.59.

Reading between the lines

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

  • The normalizing-TTS mechanism likely extends beyond dementia: the same transcribe–obfuscate–resynthesize loop could improve ASR for stuttering, dysarthria, heavy accents, or elderly speech, though the medical-privacy claim would not automatically transfer.
  • The roughly 50% speaker similarity may not be enough for voice-biometric authentication, so real-world deployments would need to choose an operating point or accept a weaker identity guarantee.
  • Because the paper reports almost no change in jitter and shimmer, a future adversary that relies on those micro-prosodic features could retain detection ability; this is an untested boundary of the claim.
  • A natural testable extension is human evaluation: the reported gains are measured against automatic classifiers, and no evidence yet shows whether human listeners or medical professionals can still recognize dementia from the obfuscated samples.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper introduces ClaritySpeech, an end-to-end pipeline that uses Whisper ASR, a text obfuscator (DiDOTS), and zero-shot TTS to transform dementia-affected speech into a synthetic version intended to reduce automatic detection of dementia while preserving speaker identity. Experiments on ADReSS and ADReSSo report a mean adversarial F1 drop from 0.70 to 0.59 and from 0.63 to 0.55/0.56, improved WER (e.g., 0.73 to 0.08 on ADReSS), higher UTMOS, and speaker similarity around 0.5. The paper also provides ablations, prosody feature analysis, and latency measurements.

Significance. If the privacy gain reflects genuine removal of dementia markers, the framework is a valuable contribution to privacy-preserving speech processing for atypical speech, and its zero-shot, no-fine-tuning design is practically relevant. The paper's strengths include open-source code and demos, comparison of three TTS systems, inclusion of an adaptive adversary, and a component ablation. However, the current evaluation does not separate marker removal from generic domain shift, and the dataset split may confound the results, so the headline claims are not yet fully supported.

major comments (4)
  1. [Section 3.1] The 80/20 train/test split is described at the sentence-segment level, but no statement guarantees that speakers are disjoint between train and test. If segments from the same participant appear in both sets, the original F1 estimates in Table 1 may be inflated by speaker/session memorization, and the F1 drop after obfuscation could reflect removal of speaker-specific cues rather than dementia markers. Please confirm a speaker-disjoint split or rerun with a leave-one-speaker-out protocol.
  2. [Section 3.2 and Table 1] The headline privacy result is the drop in adversarial F1 between original and obfuscated samples, but every obfuscated sample passes through a paraphrasing TTS pipeline, so the drop conflates dementia-marker removal with generic domain shift. The adaptive adversary mitigates but does not eliminate this confound because the synthetic domain remains systematically different in acoustics and lexical content. Without a control that applies the full pipeline to healthy control speech, the F1 drop cannot be attributed to marker removal. Please add a healthy-speech control and report results separately for static and adaptive adversaries on both original and obfuscated data.
  3. [Section 2.2 and Section 3.2] The text obfuscator DiDOTS (ref. 24) and the BERT-based text adversary (refs. 27 and 28) are both from the same research group, and DiDOTS was specifically trained to remove dementia markers from text. Evaluating the obfuscator against an in-family detector may overstate text privacy gains. Please include at least one independently trained or external dementia detector, or otherwise discuss and quantify the potential circularity.
  4. [Section 3.4 and Results] The introduction states that the text obfuscator 'preserves the meaning and intent of the original text,' but no evaluation of semantic fidelity is reported in the paper. The WER and UTMOS improvements are consistent with radical simplification or content deletion. Please report a semantic similarity measure (e.g., BERTScore, BLEU, or human ratings) between original and obfuscated transcripts to support the accessibility/utility claim.
minor comments (4)
  1. [Section 3.2] The text says 'Under the static scenario, the adversary has only access to raw data while in the static scenario, the models have knowledge of the obfuscation mechanism used'; the second 'static scenario' should be 'adaptive scenario.'
  2. [Table 3] The caption states that statistically significant changes are highlighted, but the table as presented contains no highlighting or significance markers; please add them or correct the caption.
  3. [Table 2] The rows 'w/o ASR' and 'w/o ZS' are not explicitly defined in the text; please clarify in Section 4 what these ablations replace (e.g., ground-truth transcripts for 'w/o ASR', and a non-zero-shot TTS or different-speaker reference for 'w/o ZS').
  4. [Abstract] The abstract states 'maintaining 50% speaker similarity,' but Table 1 reports speaker similarity varying from 0.36 to 0.53 across systems and datasets; please qualify this claim or report the range.

Circularity Check

1 steps flagged · score 4.0 of 10

Text-modality privacy gain is measured with an in-family BERT detector for a same-group obfuscator; audio and utility results remain independent.

  1. self citation load bearing [Section 2.2 (Dementia Obfuscation in Text) and Section 3.2 (Adversarial Models), with results in Tables 1 and 2]
    "We base our system on DiDOTS [24], which was designed to modify the content of the transcribed text to reduce or eliminate linguistic markers linked with dementia. ... For the text-based detection model, we re-implement a BERT model based on [27] and [28] with a learning rate of 1e-6, 10 epochs with early stopping (patience =1 on validation loss), gradient clipping of 1, input length limit of 256 tokens, and batch size of 8, trained on augmented data through back translation."

    DiDOTS [24] is prior work by two of the present authors and is the framework's 'core privacy-preserving step'; it was explicitly designed to remove the same dementia markers whose removal is then measured. The text adversary used to measure that removal is a re-implemented BERT detector whose stated basis includes [28], a dementia-detection paper co-authored by the same group. The measured text-modality privacy gain (e.g., ADReSS text F1 from 0.72 to 0.59 in Table 1, and 'the text obfuscation module effectively reduces the privacy leakage in Text by 10%' in Section 4) is therefore an in-family consistency result rather than independent evidence that dementia-specific markers, rather than generic text changes, were removed.

full rationale

ClaritySpeech's central end-to-end claim is not reduced to its inputs by definition: the pipeline combines external Whisper ASR, third-party zero-shot TTS models, and a same-group text obfuscator, and the audio privacy, speaker similarity (ECAPA-TDNN), UTMOS, and WER results are measured with external tools and would stand or fall independently of the DiDOTS citation. The one partially circular link is the text-modality privacy claim: DiDOTS was built by two of the present authors specifically to eliminate dementia markers, and the text adversary derives from the same group's earlier dementia-detection work, so the text-F1 drop is an in-family check. The missing healthy-speech control and the segment-level split without an explicit speaker-disjoint guarantee are genuine confounds for the causal 'marker removal' interpretation, but they are validity threats rather than circular reductions. The WER comparison also deserves scrutiny because the same ASR model that produced the transcript is used on audio synthesized from that transcript, but the paper does not specify the WER reference transcript, so I do not count it as a demonstrated circular step. Overall, one load-bearing self-citation in one modality supports score 4 rather than a higher score, because the framework retains independent content in its audio privacy and utility measurements.

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

The central claim rests on several domain assumptions about TTS behavior, classifier-based privacy metrics, and dataset splits; no free parameters are fitted in this paper, and no new entities are introduced.

assumptions (4)
  • domain assumption Zero-shot TTS models trained on fluent public speech normalize atypical dementia prosody while retaining enough speaker timbre.
    Stated as the core hypothesis in Section 2.2 and asserted again from Table 1's SS and WER columns; if false, the framework's privacy and identity arguments collapse.
  • domain assumption A drop in adversarial classifier F1 measures a privacy gain against realistic adversaries.
    Used throughout Section 4 to interpret Table 1; static and adaptive adversaries bracket the threat, but the metric remains tied to the specific classifiers chosen, and domain shift is not separated from marker removal.
  • domain assumption ADReSS and ADReSSo segment splits are representative and speaker-independent.
    Section 3.1 describes an 80/20 segment split but does not state that participants are disjoint; dementia detection benchmarks normally require speaker-disjoint splits to avoid identity memorization.
  • domain assumption Whisper large v3 transcribes dementia speech accurately enough for the text obfuscator to act on the intended content.
    Relied on in Section 2.2 and evaluated indirectly through WER; ASR errors could change meaning before obfuscation, and no end-to-end semantic metric is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of ClaritySpeech: Dementia Obfuscation in Speech." pith.science (2026). https://pith.science/paper/BOGN4SBI

@misc{pith2026250709282,
  author       = {Pith},
  title        = {Pith review of: ClaritySpeech: Dementia Obfuscation in Speech},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BOGN4SBI}},
  note         = {Machine review of arXiv:2507.09282}
}
read the original abstract

Dementia, a neurodegenerative disease, alters speech patterns, creating communication barriers and raising privacy concerns. Current speech technologies, such as automatic speech transcription (ASR), struggle with dementia and atypical speech, further challenging accessibility. This paper presents a novel dementia obfuscation in speech framework, ClaritySpeech, integrating ASR, text obfuscation, and zero-shot text-to-speech (TTS) to correct dementia-affected speech while preserving speaker identity in low-data environments without fine-tuning. Results show a 16% and 10% drop in mean F1 score across various adversarial settings and modalities (audio, text, fusion) for ADReSS and ADReSSo, respectively, maintaining 50% speaker similarity. We also find that our system improves WER (from 0.73 to 0.08 for ADReSS and 0.15 for ADReSSo) and speech quality from 1.65 to ~2.15, enhancing privacy and accessibility.

Figures

Figures reproduced from arXiv: 2507.09282 by the authors.

Figure 1
Figure 1. End-to-end ClaritySpeech framework. The input speech is transcribed through ASR, obfuscated via a text ob￾fuscator and fed to the zero-shot TTS system together with the reference speech, which outputs the obfuscated sample in the reference voice. We present an end-to-end framework for dementia obfusca￾tion in speech, ClaritySpeech, consisting of three components: an automatic speech recognition (ASR) module, a text … view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

41 extracted references · 30 canonical work pages

  1. [1]

    Introduction As dementia develops, it alters speech patterns, introducing dis- fluencies, pauses, lexical swaps, and convoluted grammar [1, 2, 3]. These changes not only create communication challenges but also become identifiable markers of the individual’s health condition, raising privacy risks and limiting access to speech- operated services [4, 5]. I...

    work page Pith review arXiv 2025

  2. [2]

    weakness

    ClaritySpeech 2.1. Threat and Defense Models We consider an adversary with access to a target user’s recorded speech with the goal of detecting whether the target user suffers from dementia. The adversary can operate in two settings: (a) a static setting and (b) an adaptive setting. A static adversary (A) is obfuscation-oblivious and has direct access to ...

    work page

  3. [3]

    Datasets We evaluate our systems on the ADReSS (ADR) [25] and ADReSSo (ADRo) [25] datasets

    Experimental Setup 3.1. Datasets We evaluate our systems on the ADReSS (ADR) [25] and ADReSSo (ADRo) [25] datasets. We split our datasets into sentence-level segments. For ADReSS, manual transcripts are provided and we split each sample into segments using the sentence-level timestamps from the transcripts. We split the segments with an 80-20% split for t...

    work page

  4. [4]

    We find that all systems provide significant pri- vacy and utility gain regardless of the TTS system 6

    Results Privacy & Utility Evaluation We evaluate our proposed sys- tem for privacy and utility with different zero-shot TTS sys- tems and report the results for the ADR and ADRo datasets in Table 1. We find that all systems provide significant pri- vacy and utility gain regardless of the TTS system 6. However, they exhibit tradeoffs which showcase the imp...

    work page

  5. [5]

    We also find that the impact of the transcription error is mitigated by the text obfuscation mechanism

    Discussion & Limitations We find that the proposed framework successfully reduces pri- vacy leaks while improving downstream utility, and zero-shot TTS can indeed correct atypicality in this setting. We also find that the impact of the transcription error is mitigated by the text obfuscation mechanism. Limitations of this work include the small, low-diver...

    work page

  6. [6]

    We propose an end-to-end framework that transforms both text and audio to conceal dementia markers while maintaining con- tent and speaker similarity

    Conclusion This work addresses the privacy leakage of dementia in speech. We propose an end-to-end framework that transforms both text and audio to conceal dementia markers while maintaining con- tent and speaker similarity. Our approach leverages zero-shot TTS, which filters certain acoustic features and smooths out dementia traits, enhancing downstream ...

    work page

  7. [7]

    Identifying mild cognitive impairment and mild alzheimer’s disease based on spontaneous speech using asr and linguistic features,

    G. Gosztolya, V . Vincze, L. T ´oth, M. P ´ak´aski, J. K ´alm´an, and I. Hoffmann, “Identifying mild cognitive impairment and mild alzheimer’s disease based on spontaneous speech using asr and linguistic features,” Computer Speech & Language , vol. 53, pp. 181–197, 2019

    work page 2019

  8. [8]

    Automatic speech analysis for the assessment of patients with predementia and alzheimer’s disease,

    A. K ¨onig, A. Satt, A. Sorin, R. Hoory, O. Toledo-Ronen, A. Der- reumaux, V . Manera, F. Verhey, P. Aalten, P. H. Robert et al. , “Automatic speech analysis for the assessment of patients with predementia and alzheimer’s disease,” Alzheimer’s & Dementia: Diagnosis, Assessment & Disease Monitoring , vol. 1, no. 1, pp. 112–124, 2015

    work page 2015

Show all 41 references
  1. [9]

    Con- nected speech as a marker of disease progression in autopsy- proven alzheimer’s disease,

    S. Ahmed, A.-M. F. Haigh, C. A. de Jager, and P. Garrard, “Con- nected speech as a marker of disease progression in autopsy- proven alzheimer’s disease,” Brain, vol. 136, no. 12, pp. 3727– 3737, 2013

    2013

  2. [10]

    Quantifying the effect of speech pathology on automatic and human speaker verification,

    B. M. Halpern, T. Tienkamp, W.-C. Huang, L. P. Violeta, T. Rebernik, S. de Visscher, M. Witjes, M. Wieling, D. Abur, and T. Toda, “Quantifying the effect of speech pathology on automatic and human speaker verification,” arXiv preprint arXiv:2406.06208, 2024

    2024 arXiv

  3. [11]

    Artificial in- telligence, speech, and language processing approaches to mon- itoring alzheimer’s disease: a systematic review,

    S. de la Fuente Garcia, C. W. Ritchie, and S. Luz, “Artificial in- telligence, speech, and language processing approaches to mon- itoring alzheimer’s disease: a systematic review,” Journal of Alzheimer’s Disease, vol. 78, no. 4, pp. 1547–1574, 2020

    2020

  4. [12]

    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

  5. [13]

    Robust speech recognition via large-scale weak su- pervision,

    A. Radford, J. W. Kim, T. Xu, G. Brockman, C. McLeavey, and I. Sutskever, “Robust speech recognition via large-scale weak su- pervision,” in International Conference on Machine Learning . PMLR, 2023, pp. 28 492–28 518

    2023

  6. [14]

    A survey of automatic speech recognition for dysarthric speech,

    Z. Qian and K. Xiao, “A survey of automatic speech recognition for dysarthric speech,” Electronics, vol. 12, no. 20, p. 4278, 2023

    2023

  7. [15]

    Rethinking evalu- ation in asr: Are our models robust enough?

    T. Likhomanenko, Q. Xu, V . Pratap, P. Tomasello, J. Kahn, G. Avidov, R. Collobert, and G. Synnaeve, “Rethinking evalu- ation in asr: Are our models robust enough?” arXiv preprint arXiv:2010.11745, 2020

    2010 arXiv

  8. [16]

    Xtts: a massively multilingual zero-shot text-to-speech model,

    E. Casanova, K. Davis, E. G ¨olge, G. G ¨oknar, I. Gulea, L. Hart, A. Aljafari, J. Meyer, R. Morais, S. Olayemi et al. , “Xtts: a massively multilingual zero-shot text-to-speech model,” arXiv preprint arXiv:2406.04904, 2024

    2024 arXiv

  9. [17]

    V oicecraft: Zero-shot speech editing and text-to-speech in the wild,

    P. Peng, P.-Y . Huang, D. Li, A. Mohamed, and D. Harwath, “V oicecraft: Zero-shot speech editing and text-to-speech in the wild,” arXiv preprint arXiv:2403.16973, 2024

    2024 arXiv

  10. [18]

    Quality of life of patients after total laryngectomy: the struggle against stigmatization and social exclusion using speech synthesis,

    J. Mertl, E. ˇZ´aˇckov´a, and B. ˇRepov´a, “Quality of life of patients after total laryngectomy: the struggle against stigmatization and social exclusion using speech synthesis,”Disability and Rehabili- tation: Assistive Technology, vol. 13, no. 4, pp. 342–352, 2018

    2018

  11. [19]

    Adversarial disentanglement of speaker representation for attribute-driven privacy preservation,

    P.-G. No ´e, M. Mohammadamini, D. Matrouf, T. Parcollet, A. Nautsch, and J.-F. Bonastre, “Adversarial disentanglement of speaker representation for attribute-driven privacy preservation,” arXiv preprint arXiv:2012.04454, 2020

    2012 arXiv

  12. [20]

    Differentially private adversarial auto- encoder to protect gender in voice biometrics,

    O. Chouchane, M. Panariello, O. Zari, I. Kerenciler, I. Chihaoui, M. Todisco, and M. ¨Onen, “Differentially private adversarial auto- encoder to protect gender in voice biometrics,” in Proceedings of the 2023 ACM Workshop on Information Hiding and Multimedia Security, 2023, pp...

    2023

  13. [21]

    Privacy-preserving voice analysis via disentangled representations,

    R. Aloufi, H. Haddadi, and D. Boyle, “Privacy-preserving voice analysis via disentangled representations,” in Proceedings of the 2020 ACM SIGSAC Conference on Cloud Computing Security Workshop, 2020, pp. 1–14

    2020

  14. [22]

    Fluentspeech: Stutter-oriented automatic speech editing with context-aware diffusion models,

    Z. Jiang, Q. Yang, J. Zuo, Z. Ye, R. Huang, Y . Ren, and Z. Zhao, “Fluentspeech: Stutter-oriented automatic speech editing with context-aware diffusion models,” arXiv preprint arXiv:2305.13612, 2023

    2023 arXiv

  15. [23]

    Accurate synthesis of dysarthric speech for asr data augmenta- tion,

    M. Soleymanpour, M. T. Johnson, R. Soleymanpour, and J. Berry, “Accurate synthesis of dysarthric speech for asr data augmenta- tion,” Speech Communication, vol. 164, p. 103112, 2024

    2024

  16. [24]

    Data augmentation using healthy speech for dysarthric speech recognition,

    B. Vachhani, C. Bhat, and S. K. Kopparapu, “Data augmentation using healthy speech for dysarthric speech recognition,” in Inter- speech 2018, 2018, pp. 471–475

    2018

  17. [25]

    Few-shot dysarthric speech recog- nition with text-to-speech data augmentation,

    E. Hermann and M. M. Doss, “Few-shot dysarthric speech recog- nition with text-to-speech data augmentation,” in Proc. INTER- SPEECH 2023, 2023, pp. 156–160

    2023

  18. [26]

    Using hmm-based speech synthesis to reconstruct the voice of individuals with degenerative speech disorders

    C. Veaux, J. Yamagishi, and S. King, “Using hmm-based speech synthesis to reconstruct the voice of individuals with degenerative speech disorders.” in Interspeech, 2012, pp. 967–970

    2012

  19. [27]

    Building personalised synthetic voices for individuals with severe speech impairment,

    S. Creer, S. Cunningham, P. Green, and J. Yamagishi, “Building personalised synthetic voices for individuals with severe speech impairment,” Computer Speech & Language , vol. 27, no. 6, pp. 1178–1193, 2013

    2013

  20. [28]

    A preliminary study of a two-stage paradigm for preserving speaker identity in dysarthric voice con- version,

    W.-C. Huang, K. Kobayashi, Y .-H. Peng, C.-F. Liu, Y . Tsao, H.-M. Wang, and T. Toda, “A preliminary study of a two-stage paradigm for preserving speaker identity in dysarthric voice con- version,” arXiv preprint arXiv:2106.01415, 2021

    2021 arXiv

  21. [29]

    Learning explicit prosody models and deep speaker embeddings for atypical voice conversion,

    D. Wang, S. Liu, L. Sun, X. Wu, X. Liu, and H. Meng, “Learning explicit prosody models and deep speaker embeddings for atypical voice conversion,”arXiv preprint arXiv:2011.01678, 2020

    2011 arXiv

  22. [30]

    Didots: Knowledge distilla- tion from large-language-models for dementia obfuscation in tran- scribed speech,

    D. Woszczyk and S. Demetriou, “Didots: Knowledge distilla- tion from large-language-models for dementia obfuscation in tran- scribed speech,” Proceedings on Privacy Enhancing Technolo- gies, 2025

    2025

  23. [31]

    Alzheimer’s dementia recognition through spontaneous speech: The adress challenge,

    S. Luz, F. Haider, S. de la Fuente, D. Fromm, and B. MacWhin- ney, “Alzheimer’s dementia recognition through spontaneous speech: The adress challenge,” arXiv preprint arXiv:2004.06833, 2020

    2004 arXiv

  24. [32]

    Silero vad: pre-trained enterprise-grade voice activity detector (vad), number detector and language classifier,

    S. Team, “Silero vad: pre-trained enterprise-grade voice activity detector (vad), number detector and language classifier,” https:// github.com/snakers4/silero-vad, 2024

    2024

  25. [33]

    Dis- fluencies and fine-tuning pre-trained language models for detec- tion of alzheimer’s disease

    J. Yuan, Y . Bian, X. Cai, J. Huang, Z. Ye, and K. Church, “Dis- fluencies and fine-tuning pre-trained language models for detec- tion of alzheimer’s disease.” in Interspeech, vol. 2020, 2020, pp. 2162–6

    2020

  26. [34]

    Data augmentation for dementia detection in spo- ken language,

    D. Woszczyk, A. Hl ´edikov´a, A. Akman, S. Demetriou, and B. Schuller, “Data augmentation for dementia detection in spo- ken language,” Interspeech. 2022., 2022

    2022

  27. [35]

    Artificial intelligence-enabled end-to- end detection and assessment of alzheimer’s disease using voice,

    F. Agbavor and H. Liang, “Artificial intelligence-enabled end-to- end detection and assessment of alzheimer’s disease using voice,” Brain sciences, vol. 13, no. 1, p. 28, 2022

    2022

  28. [36]

    Hifi-gan: Generative adversarial net- works for efficient and high fidelity speech synthesis,

    J. Kong, J. Kim, and J. Bae, “Hifi-gan: Generative adversarial net- works for efficient and high fidelity speech synthesis,” Advances in neural information processing systems , vol. 33, pp. 17 022– 17 033, 2020

    2020

  29. [37]

    Styletts 2: Towards human-level text-to-speech through style dif- fusion and adversarial training with large speech language mod- els,

    Y . A. Li, C. Han, V . Raghavan, G. Mischler, and N. Mesgarani, “Styletts 2: Towards human-level text-to-speech through style dif- fusion and adversarial training with large speech language mod- els,” Advances in Neural Information Processing Systems, vol. 36, 2024

    2024

  30. [38]

    Hierspeech++: Bridging the gap between semantic and acoustic representation of speech by hierarchical variational inference for zero-shot speech synthesis,

    S.-H. Lee, H.-Y . Choi, S.-B. Kim, and S.-W. Lee, “Hierspeech++: Bridging the gap between semantic and acoustic representation of speech by hierarchical variational inference for zero-shot speech synthesis,” arXiv preprint arXiv:2311.12454, 2023

    2023 arXiv

  31. [39]

    Mp-senet: A speech enhance- ment model with parallel denoising of magnitude and phase spec- tra,

    Y .-X. Lu, Y . Ai, and Z.-H. Ling, “Mp-senet: A speech enhance- ment model with parallel denoising of magnitude and phase spec- tra,” arXiv preprint arXiv:2305.13686, 2023

    2023 arXiv

  32. [40]

    Utmos: Utokyo-sarulab system for voicemos challenge 2022,

    T. Saeki, D. Xin, W. Nakata, T. Koriyama, S. Takamichi, and H. Saruwatari, “Utmos: Utokyo-sarulab system for voicemos challenge 2022,” arXiv preprint arXiv:2204.02152, 2022

    2022 arXiv

  33. [41]

    Ecapa- tdnn: Emphasized channel attention, propagation and ag- gregation in tdnn based speaker verification,

    B. Desplanques, J. Thienpondt, and K. Demuynck, “Ecapa- tdnn: Emphasized channel attention, propagation and ag- gregation in tdnn based speaker verification,” arXiv preprint arXiv:2005.07143, 2020

    2005 arXiv

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.