Pith. sign in

REVIEW 4 major objections 4 minor 35 references

Controlling your Attributes in Voice

T0 review · 4 major / 4 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read The paper claims that a two-module pipeline—a GAN-based speaker-representation variational autoencoder followed by a two-stage voice converter—can edit a speaker's apparent age and gender in speech while preserving source identity, using…

desk verdict Plausible architecture for speech-level attribute control, but identity preservation under gender modification is not established by the reported evidence. read the letter →

arxiv 2501.01674 v1 pith:CF2ZEABT submitted 2025-01-03 cs.SD eess.AS

Xuyuan Li , Zengqiang Shang.Li Wang , Pengyuan Zhang This is my paper · ORCID
classification cs.SDeess.AS
keywords speakerattributecontrolageandgendermodificationvoiceconversionrepresentationdisentanglementGAN-basedvariationalautoencoderflowmatchingzero-shotcyclicconsistencytraining
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

Age and gender are audible traits of a voice, but most speech-generation systems either clone the whole voice or ignore identity when changing attributes. This paper tries to establish that both traits can be edited independently of who is speaking: from a single unlabeled recording, the method rewrites the speaker's apparent age and gender into predefined target labels while keeping the voice recognizable as the same person. The authors build this as two modules: a GAN-based variational autoencoder that splits a speaker-recognition vector into age, gender, and identity embeddings, and a two-stage voice converter that turns those embeddings plus linguistic features into modified speech. If correct, the approach would give audiobook, dubbing, privacy, and synthetic-data applications a way to vary age and gender without parallel recordings or text transcripts.

What carries the argument

The central machinery is SRAVE's three-branch encoder, with one residual branch for age, one for gender, and one for identity, trained by classification losses on the first two, a contrastive cosine-similarity loss on the third, reconstruction and adversarial losses, and a cyclic-consistency step in which the encoder is frozen and the decoder is trained so that randomly re-labeled generated vectors pass the same classifier and contrastive checks as real vectors. The second machine is TSVC, an average generator using AdaIN that builds attribute-typical acoustics from attribute labels, SPPGs, and pitch, followed by a flow-matching ODE detail generator that maps those average features to speaker-specific features using transformer layers conditioned on the attribute and identity embeddings. Together they let attribute expression be shared across speakers: the average stage captures what an age-gender group sounds like, and the detail stage re-personalizes it via $z_{\text{identity}}$ without parallel data.

What would settle it

Train a linear probe on the SRAVE's identity embeddings $z_{\text{identity}}$ from held-out real speaker vectors to predict the seven age bands and gender; if the probe's accuracy is far above chance, then the identity channel still carries attribute information and the claimed disentanglement—and with it the promise of identity-preserving attribute editing—fails.

Watch

Extended reading notes

Core claim

In the paper's own terms, the central discovery is that speaker attributes can be controlled at the speech level rather than only at the representation level. A WavLM-based speaker-recognition vector from the source utterance is fed to the SRAVE, whose encoder produces three embeddings—$z_{\text{age}}$, $z_{\text{gender}}$, $z_{\text{identity}}$—and whose decoder reconstructs the vector from attribute labels and $z_{\text{identity}}$; a discriminator and a cyclic-consistency loop push these generated non-existent speaker vectors into the real speaker-vector space. The TSVC then takes sparse phonetic posteriorgrams (SPPGs) and pitch from the source, generates attribute-dependent average acoustic features with an AdaIN-based average generator, and uses a flow-matching ODE transformer to refine those features into speaker-specific acoustics conditioned on the attribute and identity embeddings. The authors report that the generated vectors are classified by the SRAVE's own attribute classifier at 98.0% age and 99.9% gender accuracy, that subjective gender agreement reaches 84.2% versus 88.5% for ground truth, and that objective speaker similarity on modified speech, while lower than on unmodified speech, remains well above the different-speaker baseline. Their explicit claim is that this is the first method to achieve age and gender control at the speech level while preserving speaker identity.

Load-bearing premise

The method assumes the speaker vector from the recognition model separates cleanly into age, gender, and identity channels, so that changing the age and gender labels leaves enough identity information intact to rebuild the same speaker's voice.

Editorial extensions

If this is right

  • From one source recording, a user can request a different age band and gender label, and the pipeline produces speech in that target group without parallel text or paired data.
  • Because the generated speaker vectors lie in the same representation space as real ones, downstream speaker-recognition and self-supervised models can consume synthetic age-gender-varied speakers directly.
  • Replacing the input semantic features with text would make the same two-stage design a text-to-speech system with controllable speaker attributes.
  • Removing either module degrades the result: the ablations show attribute consistency and intelligibility fall when TSVC is removed, and attribute consistency falls when SRAVE is removed.
  • Gender-modified speech shows the largest drop in same-speaker similarity, so the trade-off between attribute strength and identity retention is concentrated in how far pitch and timbre are moved.

Reading between the lines

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

  • Inference: because the paper never enforces independence between $z_{\text{identity}}$ and the attribute branches, a linear probe on $z_{\text{identity}}$ could reveal residual age or gender information; the reported identity metrics do not rule this out.
  • Inference: the objective speaker-similarity drop on gender-modified speech suggests that a practical version of this method would need a separate perceptual constraint tying fundamental-frequency contours to the same speaker, which the paper does not model.
  • Inference: cyclic consistency only certifies that generated vectors fool the SRAVE's own classifiers, so the method's attribute realism is only as convincing as those classifiers are; a stronger test would be whether a held-out age and gender predictor agrees on the final audio.
  • Inference: the authors' future-work idea that synthetic attribute-varied speech could improve speaker recognition is testable immediately, but the paper itself gives no evidence of downstream gains.
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 proposes a two-stage pipeline for age and gender control in speech without parallel data. First, SRAVE, a GAN-trained variational autoencoder, maps WavLM-derived speaker vectors into age, gender, and identity embeddings and can generate non-existent speaker vectors for target attribute labels. Second, TSVC, a two-stage voice converter consisting of an average generator and a flow-matching detail generator, conditions on those embeddings plus SPPGs and pitch to synthesize speech with the target attributes. Experiments on a VoxCeleb2/Age-Vox-Celeb subset report attribute classification accuracy, subjective age/gender consistency, MOS scores, and WavLM-based speaker similarity, with ablations showing that both proposed modules contribute. The paper claims this is the first method to achieve age and gender control at the speech level while preserving speaker identity.

Significance. If the results hold, the paper would be a useful early demonstration of independent age/gender and identity control in speech, with potential applications in dubbing, audiobooks, and privacy preservation. The manuscript includes a demo page, multiple ablations, and both objective and subjective evaluations, which are strengths. However, the current evidence does not fully establish the headline claim because key evaluations are internal and self-confirming, and the identity-preservation result under gender modification is not statistically supported. The significance is therefore moderate at this stage: the core idea is plausible, but the evidence base needs substantial strengthening before the claimed contribution can be accepted.

major comments (4)
  1. [§IV-B, Table II] The identity-preservation claim is not supported by adequately rigorous evidence for gender-modified speech. For the proposed system, SIMS-gender is 0.707 versus SIMD 0.605, a margin of only 0.102, and no confidence intervals, significance tests, EER, or d-prime values are reported. In addition, the SR model used for scoring appears to be from the same WavLM family used to create the input speaker vectors, so it is not an independent identity test. The subjective AB test in Fig. 3 is reported only as an aggregate, with no separate result for the gender-modified subset. Please add per-condition subjective identity results and an independent speaker-verification evaluation with uncertainty estimates, or restrict the identity-preservation claim accordingly.
  2. [§IV-A, Table I] The vector-level evaluation is self-confirming. Attribute accuracy is computed with the classifier inside the SRAVE encoder, and the identity similarity is computed with the contrastive branch of the same encoder that was trained to produce these embeddings. This can only show that the decoder fits the encoder's own latent geometry; it does not establish that the generated vectors carry attribute or identity information usable by external models. Please evaluate generated speaker vectors using independent pretrained age/gender classifiers and a general speaker-verification model trained on real speech.
  3. [§IV-B, age evaluation] The age evaluation procedure is potentially biased because the age intervals were selected after inspecting human predictions. The paper states that the intervals 0-12, 12-25, 25-55, and >55 were determined based on the prediction results on real speech; evaluating modified speech with the same fitted intervals, with the original speech as a reference, makes the reported age-consistency numbers favorable by construction rather than testing against a predefined label set. Please define the age intervals a priori, report sensitivity to interval boundaries, and provide per-interval counts or confidence intervals for the confusion matrix.
  4. [§II-B, Algorithm 1] The disentanglement that the method relies on is not demonstrated. The paper explicitly avoids gradient reversal and mutual information minimization, and the cyclic consistency step only guides generated vectors into the real-vector domain; nothing prevents z_identity from encoding age or gender. This could explain why identity similarity drops more for gender-modified speech in Table II. Please provide a direct test of attribute leakage in z_identity, for example by classifying age and gender from z_identity alone or by showing that z_identity is invariant under changes to the attribute labels.
minor comments (4)
  1. [§II-A] The abbreviation 'TSCV' appears in the text and should be 'TSVC'; additionally, 'SRAVE' and 'SRV AE' are used inconsistently throughout the manuscript.
  2. [Algorithm 1] The control flow of Algorithm 1 is unclear: the lines 'if train discriminator then' and 'if train consistency then' appear at the same indentation, but the intended nesting and the execution order of the update steps are not obvious. Please reformat the pseudocode.
  3. [§III-B] No explicit train/validation/test speaker split is described; please clarify whether the evaluation speakers are disjoint from the training speakers and how the test utterances were selected.
  4. [§I] The claim that the method is 'the first' to achieve age and gender control at the speech level while preserving identity should be supported by a more systematic comparison with existing attribute-editing and voice-conversion work; the current related-work discussion is brief and does not fully justify the novelty claim.

Circularity Check

2 steps flagged · score 4.0 of 10

Supporting vector-level evaluation reuses the SRV AE's own classifier and contrastive losses; the central speech-level claim still rests on independent listening tests and an external SR similarity model.

  1. fitted input called prediction [Sec. IV-A, Table I; Algorithm 1 (Cyclic Consistency Training)]
    "In Algorithm 1: 'Calculate classifier loss Lce, contrastive loss Lcl, and adversarial loss Ladv: Lφ =Lce +Lcl +Ladv.' Sec. IV-A: 'We assessed the extracted attribute embeddings by measuring their classification accuracy on the classifier in the SRV AE encoder.'"

    The consistency-training step in Alg. 1 minimizes Lce (classifier loss) and Lcl (contrastive loss) on the decoder's generated vectors with the encoder frozen. Table I's ACC-age/gender on generated vectors is therefore the value of the training loss Lce, not an independent prediction. The paper's inference that similar accuracies 'suggest that the non-existent speaker vectors generated by SRV AE reside in the same representation space as real speaker vectors' is circular: the decoder was explicitly trained to make the frozen classifier assign the sampled attribute labels, so high classifier agreement is by construction.

  2. fitted input called prediction [Sec. IV-A, Table I (identity rows); Algorithm 1 (Cyclic Consistency Training)]
    "Sec. IV-A: 'As for the extracted identity embedding, we compared the cosine similarity of those from the same speaker (CSIS) with the cosine similarity of those from different speakers (CSID).' ... 'the gap for the generated speaker vectors is 0.633, a slight decrease, yet still sufficient to indicate that the proposed SRV AE is capable of extracting accurate identity embeddings from the modified non-existent speaker vectors.'"

    Lcl is precisely the same-speaker cosine-similarity objective that Table I reports as CSIS/CSID. Since the identity encoder used for scoring is the frozen encoder that serves as the training target during consistency training, the reported 0.633 gap is the optimized contrastive objective, not external evidence of identity preservation. This supports only that the loss was optimized, not that age/gender-modified speaker vectors preserve identity in an independent sense.

full rationale

The central speech-level claim does not reduce to its inputs: gender and age consistency are measured by listener predictions, speech quality by QCMOS/ICMOS, and identity by an AB listening test and a pretrained WavLM SR similarity model that is not trained by the authors. These are external, falsifiable benchmarks. The only genuinely circular component is the Section IV-A evaluation of generated speaker vectors, which scores with the same classifier and contrastive branch used as training objectives in SRV AE's cyclic-consistency step. Because the vector-level evaluation is supporting rather than the main claim, and because the speech-level evidence is independent, the overall circularity is partial and moderate. The self-citation to the authors' own SF-Speech [18] for the transformer backbone is an architectural choice, not a load-bearing derivation.

Assumptions & free parameters 3 free parameters · 3 assumptions · 2 invented entities

The central claim rests on the assumption that WavLM speaker vectors can be decomposed into independent attribute and identity parts, that attribute embeddings can be shared across speakers despite being speaker-dependent, and that flow matching over average-to-specific acoustic features preserves identity. No external measurement independently validates the attribute and identity decomposition.

free parameters (3)
  • age band boundaries = 7 training bands plus evaluation intervals re-chosen as 0-12, 12-25, 25-55, >55
    The training age bands are arbitrary and the perceived-age intervals used for evaluation were determined after inspecting human predictions (Sec. IV-B), so attribute consistency results depend on these choices.
  • model capacities and training steps = 512 hidden units, 3 residual blocks, 8 transformer layers, 1024 hidden units, SRAVE 800K steps, TSVC 300K steps
    These hyperparameters are chosen by hand and no sensitivity analysis is provided; the central results may depend on them.
  • equal loss weighting in SRAVE = implicit weight of 1 for Lce, Lcl, Lmse, and Ladv
    Algorithm 1 sums the losses with implicit equal weights, and no ablation of loss weights is reported.
assumptions (3)
  • domain assumption The WavLM speaker vector contains sufficient and linearly separable age, gender, and identity information.
    The entire SRAVE design assumes three independent branches can decompose the extracted speaker vector; no analysis verifies separability or sufficiency beyond training losses (Sec. II-B).
  • domain assumption Attribute embeddings are speaker-dependent yet can be shared across similar speakers by TSVC.
    The paper states this belief in Sec. II-C ('We believe these embeddings can help...'), but the sharing mechanism is not formally modeled or independently demonstrated.
  • standard math Flow matching with X_t = (1 - t)(X0 + xi) + t X_T defines a valid ODE path from average to speaker-specific acoustic features.
    The method relies on rectified flow theory [32], assumed valid for the chosen acoustic feature spaces; no failure analysis is provided.
invented entities (2)
  • SRAVE attribute embeddings (z_age, z_gender)
    purpose: Carry age and gender information from a generated speaker vector into the voice conversion conditioner.
    No external probe validates what these embeddings encode; the classifier used to score them is part of the same encoder, and the paper states they are speaker-dependent (Sec. II-C). Speech-level listening results provide indirect evidence but not independent measurement.
  • Non-existent speaker vectors
    purpose: Intermediate speaker representation with target attributes and source identity, used to condition TSVC.
    These vectors are only evaluated by re-encoding them with the same SRAVE encoder (Table I), so their standing in the real speaker-vector distribution is not externally verified.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Controlling your Attributes in Voice." pith.science (2026). https://pith.science/paper/CF2ZEABT

@misc{pith2026250101674,
  author       = {Pith},
  title        = {Pith review of: Controlling your Attributes in Voice},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CF2ZEABT}},
  note         = {Machine review of arXiv:2501.01674}
}
read the original abstract

Attribute control in generative tasks aims to modify personal attributes, such as age and gender while preserving the identity information in the source sample. Although significant progress has been made in controlling facial attributes in image generation, similar approaches for speech generation remain largely unexplored. This letter proposes a novel method for controlling speaker attributes in speech without parallel data. Our approach consists of two main components: a GAN-based speaker representation variational autoencoder that extracts speaker identity and attributes from speaker vector, and a two-stage voice conversion model that captures the natural expression of speaker attributes in speech. Experimental results show that our proposed method not only achieves attribute control at the speaker representation level but also enables manipulation of the speaker age and gender at the speech level while preserving speech quality and speaker identity.

Figures

Figures reproduced from arXiv: 2501.01674 by the authors.

Figure 1
Figure 1. Overall workflow of our proposed method (a), detailed [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Confusion matrix for subjective prediction of speak [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

35 extracted references · 29 canonical work pages

  1. [1]

    Heterogeneous face recognition via face synthesis with identity- JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 5 attribute disentanglement,

    Ziming Y ang, Jian Liang, Chaoyou Fu, Mandi Luo, and Xiao- Y u Zhang, “Heterogeneous face recognition via face synthesis with identity- JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 5 attribute disentanglement,” IEEE Transactions on Information F orensics and Security , vol. 17, pp. 1344–1358, 2022

    work page 2015

  2. [2]

    Cont rollable and guided face synthesis for unconstrained face recognition,

    Feng Liu, Minchul Kim, Anil Jain, and Xiaoming Liu, “Cont rollable and guided face synthesis for unconstrained face recognition, ” in European Conference on Computer Vision . Springer, 2022, pp. 701–719

    work page 2022

  3. [3]

    Privacy-preserving a nnotation of face images through attribute-preserving face synthesis,

    Sola Shirai and Jacob Whitehill, “Privacy-preserving a nnotation of face images through attribute-preserving face synthesis,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Reco gnition W orkshops, 2019, pp. 0–0

    work page 2019

  4. [4]

    Is synthetic data all we nee d? benchmarking the robustness of models trained with synthet ic images,

    Krishnakant Singh, Thanush Navaratnam, Jannik Holmer, Simone Schaub-Meyer, and Stefan Roth, “Is synthetic data all we nee d? benchmarking the robustness of models trained with synthet ic images,” in Proceedings of the IEEE/CVF Conference on Computer Vision a nd Pattern Recognition, 2024, pp. 2505–2515

    work page 2024

  5. [5]

    Attgan: Facial attribute editing by only changing wh at you want,

    Zhenliang He, Wangmeng Zuo, Meina Kan, Shiguang Shan, an d Xilin Chen, “Attgan: Facial attribute editing by only changing wh at you want,” IEEE transactions on image processing , vol. 28, no. 11, pp. 5464–5478, 2019

    work page 2019

  6. [6]

    Guidedstyle: Attribute knowledge guided style manipulation for semantic face editing,

    Xianxu Hou, Xiaokang Zhang, Hanbang Liang, Linlin Shen, Zhihui Lai, and Jun Wan, “Guidedstyle: Attribute knowledge guided style manipulation for semantic face editing,” Neural Networks , vol. 145, pp. 209–220, 2022

    work page 2022

  7. [7]

    Icgnet: An intensity-controllable generation ne twork based on covering learning for face attribute synthesis,

    Xin Ning, Feng He, Xiaoli Dong, Weijun Li, Fayadh Alenezi , and Prayag Tiwari, “Icgnet: An intensity-controllable generation ne twork based on covering learning for face attribute synthesis,” Information Sciences , vol. 660, pp. 120130, 2024

    work page 2024

  8. [8]

    Speech fluency: effe ct of age, gender and context,

    Linda H Leeper and Richard Culatta, “Speech fluency: effe ct of age, gender and context,” F olia phoniatrica et logopaedica , vol. 47, no. 1, pp. 1–14, 1995

    work page 1995

Show all 35 references
  1. [9]

    Ag e- related effects on speech production: A review,

    Linda Mortensen, Antje S Meyer, and Glyn W Humphreys, “Ag e- related effects on speech production: A review,” Language and Cognitive Processes, vol. 21, no. 1-3, pp. 238–290, 2006

    2006

  2. [10]

    A novel method for classifying body mass index on the basis of s peech signals for future clinical applications: a pilot study,

    Bum Ju Lee, Boncho Ku, Jun-Su Jang, and Jong Y eol Kim, “A novel method for classifying body mass index on the basis of s peech signals for future clinical applications: a pilot study,” Evidence-Based Complementary and Alternative Medicine , vol. 2013, no. 1, pp. 150265, 2013

    2013

  3. [11]

    Evolution of voices in french audiovisual media across gen ders and age in a diachronic perspective,

    Albert Rilliard, David Doukhan, R´ emi Uro, and Simon De vauchelle, “Evolution of voices in french audiovisual media across gen ders and age in a diachronic perspective,” arXiv preprint arXiv:2404.16104 , 2024

    2024 arXiv

  4. [12]

    Learning utterance-l evel represen- tations for speech emotion and age/gender recognition usin g deep neural networks,

    Zhong-Qiu Wang and Ivan Tashev, “Learning utterance-l evel represen- tations for speech emotion and age/gender recognition usin g deep neural networks,” in 2017 IEEE international conference on acoustics, speech and signal processing (ICASSP) . IEEE, 2017, pp. 5150–5154

    2017

  5. [13]

    Age and gender recognition using a convolutional neural ne twork with a specially designed multi-attention module through speec h spectro- grams,

    Anvarjon Tursunov, Mustaqeem, Joon Y eon Choeh, and Soo nil Kwon, “Age and gender recognition using a convolutional neural ne twork with a specially designed multi-attention module through speec h spectro- grams,” Sensors, vol. 21, no. 17, pp. 5892, 2021

    2021

  6. [14]

    Speech-based age and gender predicti on with transformers,

    Felix Burkhardt, Johannes Wagner, Hagen Wierstorf, Fl orian Eyben, and Bj¨ orn Schuller, “Speech-based age and gender predicti on with transformers,” in Speech Communication; 15th ITG Conference . VDE, 2023, pp. 46–50

    2023

  7. [15]

    Privacy-oriented manipulation of speaker representatio ns,

    Francisco Teixeira, Alberto Abad, Bhiksha Raj, and Isa bel Trancoso, “Privacy-oriented manipulation of speaker representatio ns,” IEEE Ac- cess, 2024

    2024

  8. [16]

    Investigating the contribution of speaker attributes to speaker separability using disent angled speaker representations,

    Chau Luu, Steve Renals, and Peter Bell, “Investigating the contribution of speaker attributes to speaker separability using disent angled speaker representations,” in Interspeech 2022 . ISCA, 2022, pp. 610–614

    2022

  9. [17]

    Adversarial-fr ee speaker identity-invariant representation learning for automati c dysarthric speech classification.,

    Parvaneh Janbakhshi and Ina Kodrasi, “Adversarial-fr ee speaker identity-invariant representation learning for automati c dysarthric speech classification.,” in INTERSPEECH, 2022, pp. 2138–2142

    2022

  10. [18]

    Sf-speech: Straightened flow for zero-shot voice clone on small-scale dataset,

    Xuyuan Li, Zengqiang Shang, Hua Hua, Peiyang Shi, Chen Y ang, Li Wang, and Pengyuan Zhang, “Sf-speech: Straightened flow for zero-shot voice clone on small-scale dataset,” arXiv preprint arXiv:2410.12399, 2024

    2024 arXiv

  11. [19]

    Cosyvoice : A scalable multilingual zero-shot text-to-speech synthes izer based on supervised semantic tokens,

    Zhihao Du, Qian Chen, Shiliang Zhang, Kai Hu, Heng Lu, Y e xin Y ang, Hangrui Hu, Siqi Zheng, Y ue Gu, Ziyang Ma, et al., “Cosyvoice : A scalable multilingual zero-shot text-to-speech synthes izer based on supervised semantic tokens,” arXiv preprint arXiv:2407.05407 , 2024

    2024 arXiv

  12. [20]

    Seed-tts: A family of high-quality versatile speech gener ation models,

    Philip Anastassiou, Jiawei Chen, Jitong Chen, Y uanzhe Chen, Zhuo Chen, Ziyi Chen, Jian Cong, Lelai Deng, Chuang Ding, Lu Gao, e t al., “Seed-tts: A family of high-quality versatile speech gener ation models,” arXiv preprint arXiv:2406.02430 , 2024

    2024 arXiv

  13. [21]

    A b enchmark for multi-speaker anonymization,

    Xiaoxiao Miao, Ruijie Tao, Chang Zeng, and Xin Wang, “A b enchmark for multi-speaker anonymization,” arXiv preprint arXiv:2407.05608 , 2024

    2024 arXiv

  14. [22]

    Speaker anonymization using neural audio codec language models,

    Michele Panariello, Francesco Nespoli, Massimiliano Todisco, and Nicholas Evans, “Speaker anonymization using neural audio codec language models,” in ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) . IEEE, 2024, pp. 4725–4729

    2024

  15. [23]

    Distinctive and natural speaker anonymization via singul ar value transformation-assisted matrix,

    Jixun Y ao, Qing Wang, Pengcheng Guo, Ziqian Ning, and Le i Xie, “Distinctive and natural speaker anonymization via singul ar value transformation-assisted matrix,” IEEE/ACM Transactions on Audio, Speech, and Language Processing , vol. 32, pp. 2944–2956, 2024

    2024

  16. [24]

    Synthe-sees: Face based text-to-speech for virtual speak er,

    Jae Hyun Park, Joon-Gyu Maeng, TaeJun Bak, and Y oung-Su n Joo, “Synthe-sees: Face based text-to-speech for virtual speak er,” in ICASSP 2024-2024 IEEE International Conference on Acoustics, Spe ech and Signal Processing (ICASSP) . IEEE, 2024, pp. 10321–10325

    2024

  17. [25]

    Fvtts: Fac e based voice synthesis for text-to-speech,

    Minyoung Lee, Eunil Park, and Sungeun Hong, “Fvtts: Fac e based voice synthesis for text-to-speech,” in Proc. Interspeech 2024 , 2024, pp. 4953–4957

    2024

  18. [26]

    Dreamvoice: Text-guided voice conversion,

    Jiarui Hai, Karan Thakkar, Helin Wang, Zengyi Qin, and M ounya Elhilali, “Dreamvoice: Text-guided voice conversion,” arXiv preprint arXiv:2406.16314, 2024

    2024 arXiv

  19. [27]

    Generating speakers by prompting listener im pressions for pre-trained multi-speaker text-to-speech systems,

    Zhengyang Chen, Xuechen Liu, Erica Cooper, Junichi Y am agishi, and Y anmin Qian, “Generating speakers by prompting listener im pressions for pre-trained multi-speaker text-to-speech systems,” arXiv preprint arXiv:2406.08812, 2024

    2024 arXiv

  20. [28]

    Fine-grained and interpretable neural speech editing,

    Max Morrison, Cameron Churchwell, Nathan Pruyne, and B ryan Pardo, “Fine-grained and interpretable neural speech editing,” Interspeech 2024, 2024

    2024

  21. [29]

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

    Sanyuan Chen, Chengyi Wang, Zhengyang Chen, Y u Wu, Shuj ie Liu, Zhuo Chen, Jinyu Li, Naoyuki Kanda, Takuya Y oshioka, Xiong X iao, et al., “Wavlm: Large-scale self-supervised pre-training for full stack speech processing,” IEEE Journal of Selected Topics in Signal Process- in...

    2022

  22. [30]

    Disen tangled information bottleneck,

    Ziqi Pan, Li Niu, Jianfu Zhang, and Liqing Zhang, “Disen tangled information bottleneck,” in Proceedings of the AAAI Conference on Artificial Intelligence , 2021, vol. 35, pp. 9285–9293

    2021

  23. [31]

    Arbitrary style transfe r in real-time with adaptive instance normalization,

    Xun Huang and Serge Belongie, “Arbitrary style transfe r in real-time with adaptive instance normalization,” in Proceedings of the IEEE international conference on computer vision , 2017, pp. 1501–1510

    2017

  24. [32]

    Flow strai ght and fast: Learning to generate and transfer data with rectified fl ow,

    Xingchao Liu, Chengyue Gong, and Qiang Liu, “Flow strai ght and fast: Learning to generate and transfer data with rectified fl ow,” arXiv preprint arXiv:2209.03003, 2022

    2022 arXiv

  25. [33]

    V oxceleb2: Deep speaker recognition,

    Joon Son Chung, Arsha Nagrani, and Andrew Zisserman, “V oxceleb2: Deep speaker recognition,” Interspeech 2018 , 2018

    2018

  26. [34]

    Age-vox-celeb: Multi-modal corpus for facial a nd speech estimation,

    Naohiro Tawara, Atsunori Ogawa, Y uki Kitagishi, and Ho sana Kamiyama, “Age-vox-celeb: Multi-modal corpus for facial a nd speech estimation,” in ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) . IEEE, 2021, pp. 6963–6967

    2021

  27. [35]

    Hifi-gan : Generative adversarial networks for efficient and high fidelity speech s ynthesis,

    Jungil Kong, Jaehyeon Kim, and Jaekyoung Bae, “Hifi-gan : Generative adversarial networks for efficient and high fidelity speech s ynthesis,” Advances in neural information processing systems , vol. 33, pp. 17022– 17033, 2020

    2020

Pith tools

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