Pith. sign in

REVIEW 4 major objections 5 minor 37 references

Voice Conversion for Likability Control via Automated Rating of Speech Synthesis Corpora

T0 review · 4 major / 5 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 A voice conversion system that raises or lowers perceived likability of any speaker's voice while preserving who is speaking and what is said.

desk verdict A novel auto-annotated likability-conditioned VC system with a circular objective evaluation and an abstract that overreaches; worth a serious referee but needs major revision. read the letter →

arxiv 2507.01356 v1 pith:LHFKRDD3 submitted 2025-07-02 eess.AS cs.SD

Hitoshi Suda , Shinnosuke Takamichi , Satoru Fukayama This is my paper
classification eess.AScs.SD
keywords voiceconversionlikabilityparalinguisticcontrolspeechsynthesispredictionautomaticcorpusannotationFast2HuBERTdiscreteunits
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 sets out to establish that voice likability — the subjective likeability of a speaker's voice — can be treated as a controllable dimension of speech synthesis, alongside who is speaking and what is said. To make this practical, it trains an automatic likability predictor on an existing corpus of human ratings, then uses that predictor to label thousands of utterances in large Japanese speech corpora, removing the need for expensive manual ratings. On top of those pseudo-labels, a voice conversion model based on discrete speech units, speaker embeddings, and a target likability score synthesizes the same words in a more or less likable voice. Subjective listening tests with 100 participants confirm that listeners perceive the intended likability differences for three of the four test speakers, and objective checks show linguistic content and speaker identity are mostly preserved. If correct, the result is a scalable recipe for controlling a subjective voice quality without collecting new human ratings for every corpus.

What carries the argument

The load-bearing object is the automatic likability predictor, a single network that maps a log-Mel spectrogram through three time-delay neural network layers and a statistics-pooling layer (the x-vector-style backbone) to four time-invariant ratings, one per listener group defined by gender and age. A post-filtering step — a linear rescaling of the raw predictions to match the mean and variance of human ratings on the validation set — corrects for the predictor's tendency to regress toward the center of the rating scale. This predictor does double duty: it produces the pseudo-labels used to train the voice conversion model on the JVS and JTES corpora, and it is also the objective metric used to verify likability control on converted speech. The voice conversion model itself is a FastSpeech 2 TTS backbone, fed with compressed HuBERT cluster indices (k-means with k = 1000), an ECAPA-TDNN speaker embedding, and the target likability scalar; a trade-off multiplier s (1 at training, 2.5 at inference) scales the likability conditioning to balance control strength against speaker preservation.

What would settle it

Run a fresh listener panel on the existing converted audio: collect human likability ratings for utterances spanning the full target range (−2 to 2) across several speakers, and compare those ratings against both the target values and the predictor's outputs. If human ratings do not move monotonically with the target — or diverge from the predictor's scores while the predictor still tracks the targets — the control effect is an artifact of the auto-annotator echoing its own labels, and the central claim fails. A cheaper partial check is to verify whether the predictor's LCC of about 0.46 holds on a human-rated holdout drawn from JVS/JTES alone.

Watch

Extended reading notes

Core claim

The paper's central claim is that perceived voice likability is a controllable acoustic attribute that can be manipulated by conditioning a TTS-based voice converter on a single scalar likability target, while keeping the speaker and the words fixed. The discovery chain runs through two components that feed each other: a TDNN-based predictor, trained on the CocoNut-Humoresque corpus to output mean likability for four listener groups (gender × age), generalizes well enough (LCC = 0.46, p < 3×$10^{-17}$, 74% liked/disliked classification accuracy) to serve as an automatic annotator; and a FastSpeech 2 model conditioned on HuBERT discrete units, an ECAPA-TDNN speaker embedding, and the predicted likability rating learns to slide the output voice along the likability axis. The authors demonstrate the control objectively — predicted likability of converted speech tracks the target for all listener groups — and subjectively, with listeners significantly preferring the intended voice in paired comparisons for three of the four test speakers. They also report the expected trade-off: pushing likability far from the speaker's natural range degrades speaker identity and intelligibility, which they expose through a scalar multiplier that the user can tune at inference.

Load-bearing premise

The load-bearing premise is that the predictor's modest agreement with human raters (LCC = 0.46) is good enough that the pseudo-labels it stamps on the JVS and JTES corpora teach the conversion model the right thing, and that the same predictor is then a fair objective gauge of the converted speech.

Editorial extensions

If this is right

  • A user can take any utterance from a covered speaker and re-synthesize it at a higher or lower likability level, with speaker identity and linguistic content largely intact within the −1 to 1 target range.
  • Because the predictor requires no human raters at conversion time, the same pipeline can in principle be moved to any moderately sized multi-speaker corpus: predict, annotate, train.
  • The trade-off knob s gives an explicit, tunable dial between how much likability changes and how much of the original speaker's identity survives, making the method usable in applications where one of the two matters more.
  • The predictor itself can serve as a cheap, continuous substitute for some listening-test functions in future synthesis work, since its outputs correlate significantly with human ratings and it reports per-listener-group predictions.
  • The observed failure mode — weak control between targets 0 and 1, and speaker m49's converted speech becoming less likable at higher targets — implies that control quality depends on the base converter's naturalness and identity preservation, not just on the likability signal.

Reading between the lines

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

  • An immediate next test the authors do not run: check whether the likability predictor transfers across languages or recording conditions, since CocoNut-Humoresque, JVS, and JTES are all Japanese speech, leaving the transfer assumption untested outside this language and domain.
  • A practical application implied by the trade-off analysis: the scalar multiplier s could be set automatically by a speaker-verification-style threshold, so the system pushes likability as far as possible while keeping the converted voice inside a guaranteed identity-acceptance region.
  • Because the predictor outputs separate ratings for four listener groups, the same trained model could generate reference voices targeted at a specific demographic — e.g., maximally likable to women over 40 — a fine-grained control the paper demonstrates but does not pursue as an application.
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 / 5 minor

Summary. The paper proposes a pipeline for controlling perceived voice likability in voice conversion. A neural predictor (TDNN with statistics pooling) is trained on the CocoNut-Humoresque corpus to output likability ratings for four listener groups, with a linear post-filter that rescales predictions to match human rating variance. This predictor is then used to pseudo-label the JVS and JTES speech corpora, and a FastSpeech 2-based voice conversion model conditioned on HuBERT discrete units, speaker embeddings, and target likability ratings is trained on these pseudo-labels. The authors report objective evaluations of likability control (using the same predictor), content preservation (CER), and speaker identity (embedding cosine similarity), plus a subjective pairwise preference test on four speakers. The paper claims effective likability control while preserving speaker identity and linguistic content.

Significance. If the central claim were fully supported, the work would be a valuable contribution to controllable speech synthesis, since it offers a scalable route from subjective ratings to a voice conversion model without per-sample manual annotation. The predictor itself shows a significant correlation with human ratings (LCC = 0.46 for the combined listener group) and a reasonably good liked/disliked classification accuracy of 74%. The subjective listening test uses a nontrivial number of participants and provides some evidence that the model can shift perceived likability for three of four speakers in one comparison direction. However, the principal evidence for likability control is circular, and the independent subjective evidence is only partial. The paper also shows clear degradation in content preservation at the extreme target value and speaker-dependent failures. The strengths are the integration of an existing likability corpus, the multi-group listener modeling, and the explicit trade-off parameter for identity preservation; the weaknesses are the unvalidated transfer of the predictor to new corpora and the overstatement of the experimental results in the abstract and conclusion.

major comments (4)
  1. [Section 5.2, Figure 3] The objective likability evaluation is circular. The predictor used to score the converted speech is the same model that generated the pseudo-labels on which the conversion model was trained in Section 3. Consequently, Figure 3 primarily verifies that the conversion model has learned to match this particular predictor, not that human-perceived likability is controlled. The post-filter in Eq. (1) only aligns the marginal mean and variance of predictions; it cannot correct per-utterance ranking errors, and no human ratings on the JVS/JTES corpora are provided to validate that the CocoNut-Humoresque-trained predictor transfers to studio-recorded speech. To support the central claim, the authors should either provide held-out human likability ratings for converted JVS/JTES speech or validate the predictor on those corpora independently.
  2. [Section 5.3, Figure 6] The subjective evaluation provides only partial support for the control claim. The paper reports no significant difference between target values 0 and 1, and speaker m49 even shows a significant decrease in perceived likability from target -1 to 1, contradicting the intended control. Only comparisons involving the -1 target, and only for three of four speakers, support the effect. The abstract's statement that the method 'effectively controls voice likability' is too strong relative to this evidence. The claims should be tempered, or additional subjective tests covering a wider target range and more speakers should be provided.
  3. [Section 5.2, Figure 4 and Conclusion] The claim that linguistic content is preserved is inconsistent with the reported results. Figure 4 shows a marked increase in CER at target likability 2, especially for female speakers, and the text in Section 5.2 explicitly acknowledges that the conversion process resulted in a higher CER at that target. The abstract and conclusion nonetheless claim that linguistic content is preserved without qualification. The content-preservation claim should be restricted to the range for which the data provide support (approximately -2 to 1), and the conclusion should reflect the acknowledged limitation.
  4. [Section 5.2, Figure 5 and Abstract] The speaker-identity preservation claim is also overgeneralized. The text states that speaker identity was preserved 'within the target ratings of -1 to 1,' but Figure 5 shows degradation at the extreme target values, with the mean cosine similarity approaching or falling below the EER threshold at target 2 for some speakers. The abstract's unqualified claim that speaker identity is preserved therefore goes beyond the presented evidence. Please qualify the identity-preservation guarantee and coordinate the abstract, Section 5.2, and the conclusion.
minor comments (5)
  1. [Figure 2] The figure contains a typo: 'Mel-spetrogram Decoder' should be 'Mel-spectrogram Decoder.'
  2. [Section 5.3] The preference test does not state which statistical test was used to determine significance. Please specify the test and whether multiple-comparison corrections were applied.
  3. [Section 4.2, Table 2] The 'All' row aggregates the four listener groups, but the aggregation method is not described. Please clarify whether the rating is averaged over groups or over listeners.
  4. [Section 4.2] The paper reports the classification accuracy of the predictor on a liked/disliked dichotomy but does not define the threshold used to map the continuous rating to the binary class. Please provide this definition.
  5. [General] The manuscript does not mention code or data availability. Since the experiments rely on several open-source components and custom training pipelines, a reproducibility statement would be valuable.

Circularity Check

1 steps flagged · score 6.0 of 10

Objective likability evaluation reuses the predictor that generates the training labels, making the main quantitative evidence a consistency check; only the partial subjective test independently supports the claim.

  1. fitted input called prediction [Section 3 (training) and Section 5.2 (objective evaluation)]
    "To overcome this limitation, we utilize likability ratings that are automatically predicted by the likability predictor described in Section 2. ... First, we evaluated the likability of the synthesized speech using the likability predictor described in Section 4."

    The voice conversion model is trained on JVS/JTES pseudo-labels produced by the Section 2 predictor, so it is optimized to make that predictor output the target rating. Section 5.2 then measures success with the same predictor (Figure 3). The resulting 'predicted likability follows the target' is therefore a fit to the label generator, not an independent measure of perceived likability. The post-filtering (Eq. 1) only rescales mean/variance on the validation set and cannot fix per-utterance ranking errors. The predictor's only human validation is on the CocoNut-Humoresque corpus (Table 2, LCC=0.46); no held-out human ratings on JVS/JTES are provided to validate transfer. The independent subjective test (Section 5.3) is partial—three of four speakers, no significant 0 vs.

full rationale

The paper is mostly self-contained: the CocoNut-Humoresque dataset [11] is an external human-rated corpus, and the predictor's LCC=0.46 is a genuine correlation on a held-out CocoNut test set. The speaker-identity and linguistic-content evaluations use independent external tools (ECAPA-TDNN, HuBERT-based ASR). The circularity is confined to the likability-control objective evaluation: the same predictor that auto-annotates the training corpus (Section 3) is reused as the evaluation metric (Section 5.2). Figure 3 therefore demonstrates consistency between the VC model and the label generator rather than human-perceived likability control. The subjective listening test partially breaks the loop (significant -1 vs. 0 for three speakers, overall -1 vs. 1), but it is small and does not cover the full -2..2 target range, so the strong quantitative claim rests substantially on the circular objective measure. This warrants a 6 rather than a higher score because the subjective test provides some independent evidence and the CocoNut predictor validation is external.

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

The central claims rest on three domain assumptions about likability being learnable and transferable across corpora, plus several fitted calibration and hyperparameter choices. No new theoretical entities are introduced.

free parameters (4)
  • Post-filtering calibration statistics = µ, σ², µ_hat, σ_hat² per listener group, computed on validation set
    Eq. (1) linearly transforms predictions using validation-set mean/variance of human and predicted ratings; this is a data-fitted calibration that affects all downstream training labels.
  • Trade-off multiplier s = 2.5 during evaluation
    Section 5.1 sets s=2.5 by hand with no sensitivity analysis; it modulates the balance between likability control and speaker identity.
  • k (number of HuBERT cluster centroids) = 1000
    Section 5.1 chooses k=1000 via mini-batch k-means; no ablation is reported.
  • Listener group partition = 4 groups (gender x age)
    Section 2 partitions listeners into four groups based on gender and age; this choice shapes the predictor targets and downstream conditioning.
assumptions (4)
  • domain assumption Likability is predictable from log-Mel spectrograms with a TDNN
    Section 2 assumes the CocoNut-Humoresque ratings can be learned from spectrograms; supported only by LCC=0.46 on the test set.
  • domain assumption Likability ratings from CocoNut-Humoresque generalize to JVS/JTES speakers for pseudo-labeling
    Section 3 uses the predictor trained on CocoNut-Humoresque to label the JVS/JTES corpus; no validation that predicted ratings align with human ratings on these corpora is provided.
  • domain assumption Pseudo-labels are reliable enough to train a voice conversion model
    Section 3 conditions FastSpeech 2 on pseudo-labels; if the predictor is noisy, the control signal is noisy and the model may overfit to predictor artifacts.
  • standard math Pretrained components (HuBERT, ECAPA-TDNN, HiFi-GAN, FastSpeech 2) work as expected
    The approach relies on prior published models and implementations [12,19,20,21,28] without re-verifying their behavior in-house.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Voice Conversion for Likability Control via Automated Rating of Speech Synthesis Corpora." pith.science (2026). https://pith.science/paper/LHFKRDD3

@misc{pith2026250701356,
  author       = {Pith},
  title        = {Pith review of: Voice Conversion for Likability Control via Automated Rating of Speech Synthesis Corpora},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LHFKRDD3}},
  note         = {Machine review of arXiv:2507.01356}
}
read the original abstract

Perceived voice likability plays a crucial role in various social interactions, such as partner selection and advertising. A system that provides reference likable voice samples tailored to target audiences would enable users to adjust their speaking style and voice quality, facilitating smoother communication. To this end, we propose a voice conversion method that controls the likability of input speech while preserving both speaker identity and linguistic content. To improve training data scalability, we train a likability predictor on an existing voice likability dataset and employ it to automatically annotate a large speech synthesis corpus with likability ratings. Experimental evaluations reveal a significant correlation between the predictor's outputs and human-provided likability ratings. Subjective and objective evaluations further demonstrate that the proposed approach effectively controls voice likability while preserving both speaker identity and linguistic content.

Figures

Figures reproduced from arXiv: 2507.01356 by the authors.

Figure 1
Figure 1. Architecture of the proposed voice likability predictor. The predictor outputs likability ratings ranging from −1 to 1 for four listener groups defined by age and gender: males under 40, males 40 or older, females under 40, and females 40 or older. T denotes the number of time frames, and D denotes the number of dimensions of the intermediate features. the center (i.e., near 0 when normalized to the range [−1, 1]). … view at source ↗
Figure 2
Figure 2. Architecture of the proposed voice conversion model for likability control. The model is based on FastSpeech 2 [21] and utilizes HuBERT-based discrete units, speaker embeddings extracted with ECAPA-TDNN [19], and target likability ratings. This approach entails a trade-off between likability control and speaker identity preservation: excessive control may com￾promise speaker identity, and vice versa. To improve usab… view at source ↗
Figure 3
Figure 3. Predicted likability ratings of utterances with con￾trolled likability. Thin lines represent the ratings for each gen￾der–age listener group, while the thick line represents the aver￾age rating across all four listener groups. 5. Experiment 2: Voice likability control 5.1. Experimental setup We utilized two speech corpora: the JVS corpus, comprising 14997 utterances from 100 speakers [24], and the JTES corpus, compr… view at source ↗
Figures from the paper (2 more)
Figure 6
Figure 6. Figure 6: Likability preference rates for synthesized utterances with different target likability ratings. Each row corresponds to a different speaker, while the “all” row represents the overall ratings. Error bars indicate the 95% confidence intervals. speaker threshold of 0.25…
Figure 5
Figure 5. Figure 5: Cosine similarity between speaker embeddings ex￾tracted from reference and synthesized speech. The leftmost col￾umn shows the results for natural speech. The gray dashed line indicates the similarity threshold corresponding to the equal er￾ror rate (EER) computed on th…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

37 extracted references · 34 canonical work pages

  1. [1]

    Numerous studies have explored methods for controlling paralinguistic and non-linguistic fea- tures in synthesized speech

    Introduction Speech conveys three types of information: (1) linguistic in- formation, which is represented by sequences of discrete sym- bols; (2) paralinguistic information, such as speaking styles, which can be intentionally controlled by speakers; and (3) non- linguistic information, such as speaker identity, which is typi- cally beyond their control [...

    work page

  2. [2]

    Voice Conversion for Likability Control via Automated Rating of Speech Synthesis Corpora

    Automatic prediction of voice likability This section describes our approach to automatically predict- ing voice likability. Figure 1 shows the architecture of the proposed likability predictor. The model accepts a log-Mel spectrogram as input and employs time-delay neural networks (TDNNs) along with a statistics pooling layer, similar to the x- vector ar...

    work page Pith review arXiv 2025

  3. [3]

    Figure 2 depicts the archi- tecture of the method

    V oice conversion for likability control This section describes the proposed voice conversion method that controls voice likability while preserving both speaker identity and linguistic information. Figure 2 depicts the archi- tecture of the method. This method is based on a voice conver- sion approach that utilizes sequences of discrete speech units extr...

    work page

  4. [4]

    liked” or “disliked

    Experiment 1: V oice likability prediction 4.1. Experimental setup We utilized the CocoNut-Humoresque corpus [11] and normal- ized the ratings to the range [ −1, 1]. The corpus is divided Table 1: Time contexts and the number of dimensions for each layer in the likability prediction architecture. Ts denotes the total number of signal samples, T denotes th...

    work page

  5. [5]

    Experiment 2: V oice likability control 5.1. Experimental setup We utilized two speech corpora: the JVS corpus, comprising 14997 utterances from 100 speakers [24], and the JTES corpus, comprising 20000 utterances from 100 speakers in four emo- tional styles (neutral, angry, joyful, and sad) [25]. We reserved ten neutral sentences (sentences 41–50) uttered...

    work page

  6. [6]

    Conclusion This paper presents a method for controlling voice likability for any speaker by extending a voice conversion approach that uses discrete speech units. To improve the scalability of training data, we constructed a likability predictor based on a voice lik- ability corpus and used it to automatically annotate speech syn- thesis corpora with lika...

    work page

  7. [7]

    R&D on Generative AI Foundation Models for the Physical Domain

    Acknowledgements This work was supported by JSPS KAKENHI Grant Numbers 21H04900, 23K20017, and 23K24895, and JST FOREST Pro- gram, Grant Number JPMJFR226V . This research was partially supported by the AIST policy-based budget project “R&D on Generative AI Foundation Models for the Physical Domain.” The authors would like to acknowledge Mr. Takizawa (AIST...

    work page

  8. [8]

    Prosody, models, and spontaneous speech,

    H. Fujisaki, “Prosody, models, and spontaneous speech,” in Com- puting Prosody. Springer US, 1997, pp. 27–42

    work page 1997

Show all 37 references
  1. [9]

    Style tokens: Un- supervised style modeling, control and transfer in end-to-end speech synthesis,

    Y . Wang, D. Stanton, Y . Zhang, R.-S. Ryan, E. Battenberg, J. Shor, Y . Xiao, Y . Jia, F. Ren, and R. A. Saurous, “Style tokens: Un- supervised style modeling, control and transfer in end-to-end speech synthesis,” in Proc. 35th International Conference on Ma- chine Learning, ...

    2018

  2. [10]

    Deep learning-based expressive speech synthesis: A systematic review of approaches, challenges, and resources,

    H. Barakat, O. Turk, and C. Demiroglu, “Deep learning-based expressive speech synthesis: A systematic review of approaches, challenges, and resources,” EURASIP Journal on Audio, Speech, and Music Processing, vol. 2024, no. 1, pp. 1–34, 2024

    2024

  3. [11]

    V oice transformation: A survey,

    Y . Stylianou, “V oice transformation: A survey,” in Proc. 2009 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2009, pp. 3585–3588

    2009

  4. [12]

    PromptTTS: Controllable text-to-speech with text descriptions,

    Z. Guo, Y . Leng, Y . Wu, S. Zhao, and X. Tan, “PromptTTS: Controllable text-to-speech with text descriptions,” in Proc. 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2023, pp. 1–5

    2023

  5. [13]

    Weiss, J

    B. Weiss, J. Trouvain, M. Barkat-Defradas, and J. J. Ohala, Eds., Voice Attractiveness: Studies on Sexy, Likable, and Charismatic Speakers. Springer Nature Singapore, 2020

    2020

  6. [14]

    V ocal and visual attractiveness are related in women,

    S. A. Collins and C. Missing, “V ocal and visual attractiveness are related in women,” Animal behaviour , vol. 65, no. 5, pp. 997– 1004, 2003

    2003

  7. [15]

    V ocal characteristics of presidential candidates can predict the outcome of actual elec- tions,

    I. Pavela Banai, B. Banai, and K. Bovan, “V ocal characteristics of presidential candidates can predict the outcome of actual elec- tions,” Evolution and human behavior: official journal of the Hu- man Behavior and Evolution Society, vol. 38, no. 3, pp. 309–314, 2017

    2017

  8. [16]

    Body height, immu- nity, facial and vocal attractiveness in young men,

    I. Skrinda, T. Krama, S. Kecko, F. R. Moore, A. Kaasik, L. Meija, V . Lietuvietis, M. J. Rantala, and I. Krams, “Body height, immu- nity, facial and vocal attractiveness in young men,”Die Naturwis- senschaften, vol. 101, no. 12, pp. 1017–1025, 2014

    2014

  9. [17]

    “Would you buy a car from me?

    F. Burkhardt, B. Schuller, B. Weiss, and F. Weninger, ““Would you buy a car from me?” – On the likability of telephone voices,” in Proc. Interspeech 2011, 2011, pp. 1557–1560

    2011

  10. [18]

    Who finds this voice attractive? A large-scale experiment using in-the-wild data,

    H. Suda, A. Watanabe, and S. Takamichi, “Who finds this voice attractive? A large-scale experiment using in-the-wild data,” in Proc. Interspeech 2024, 2024, pp. 3165–3169

    2024

  11. [19]

    X-vectors: Robust DNN embeddings for speaker recogni- tion,

    D. Snyder, D. Garcia-Romero, G. Sell, D. Povey, and S. Khudan- pur, “X-vectors: Robust DNN embeddings for speaker recogni- tion,” in Proc. 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2018, pp. 5329–5333

    2018

  12. [20]

    AutoMOS: Learning a non-intrusive as- sessor of naturalness-of-speech,

    B. Patton, Y . Agiomyrgiannakis, M. Terry, K. Wilson, R. A. Saurous, and D. Sculley, “AutoMOS: Learning a non-intrusive as- sessor of naturalness-of-speech,” in Proc. NIPS 2016 End-to-end Learning for Speech and Audio Processing Workshop, 2016

    2016

  13. [21]

    UTMOS: UTokyo-SaruLab system for V oice- MOS Challenge 2022,

    T. Saeki, D. Xin, W. Nakata, T. Koriyama, S. Takamichi, and H. Saruwatari, “UTMOS: UTokyo-SaruLab system for V oice- MOS Challenge 2022,” in Proc. Interspeech 2022 , 2022, pp. 4521–4525

    2022

  14. [22]

    Predicting likability of speakers with Gaussian processes,

    D. Lu and F. Sha, “Predicting likability of speakers with Gaussian processes,” in Proc. Interspeech 2012, 2012, pp. 286–289

    2012

  15. [23]

    Likability of human voices: A feature analysis and a neural network regression approach to automatic likability estimation,

    F. Eyben, F. Weninger, E. Marchi, and B. Schuller, “Likability of human voices: A feature analysis and a neural network regression approach to automatic likability estimation,” in Proc. 14th Inter- national Workshop on Image Analysis for Multimedia Interactive Services (WIAMIS...

    2013

  16. [24]

    Any-to-one sequence- to-sequence voice conversion using self-supervised discrete speech representations,

    W.-C. Huang, Y .-C. Wu, and T. Hayashi, “Any-to-one sequence- to-sequence voice conversion using self-supervised discrete speech representations,” in Proc. 2021 IEEE International Con- ference on Acoustics, Speech and Signal Processing (ICASSP) , 2021, pp. 5944–5948

    2021

  17. [25]

    A comparison of discrete and soft speech units for improved voice conversion,

    B. van Niekerk, M.-A. Carbonneau, J. Za ¨ıdi, M. Baas, H. Seut ´e, and H. Kamper, “A comparison of discrete and soft speech units for improved voice conversion,” in Proc. 2022 IEEE Interna- tional Conference on Acoustics, Speech and Signal Processing (ICASSP), 2022, pp. 6562–6566

    2022

  18. [26]

    ECAPA- TDNN: Emphasized channel attention, propagation and aggrega- tion in TDNN based speaker verification,

    B. Desplanques, J. Thienpondt, and K. Demuynck, “ECAPA- TDNN: Emphasized channel attention, propagation and aggrega- tion in TDNN based speaker verification,” in Proc. Interspeech 2020, 2020, pp. 3830–3834

    2020

  19. [27]

    HuBERT: Self-supervised speech representation learning by masked prediction of hidden units,

    W.-N. Hsu, B. Bolte, Y .-H. H. Tsai, K. Lakhotia, R. Salakhut- dinov, and A. Mohamed, “HuBERT: Self-supervised speech representation learning by masked prediction of hidden units,” IEEE/ACM Transactions on Audio, Speech, and Language Pro- cessing, vol. 29, pp. 3451–3460, 2021

    2021

  20. [28]

    FastSpeech 2: Fast and high-quality end-to-end text to speech,

    Y . Ren, C. Hu, X. Tan, T. Qin, S. Zhao, Z. Zhao, and T.-Y . Liu, “FastSpeech 2: Fast and high-quality end-to-end text to speech,” in Proc. International Conference on Learning Representations , 2021

    2021

  21. [29]

    Coco-Nut: Corpus of Japanese utterance and voice characteristics description for prompt-based control,

    A. Watanabe, S. Takamichi, Y . Saito, W. Nakata, D. Xin, and H. Saruwatari, “Coco-Nut: Corpus of Japanese utterance and voice characteristics description for prompt-based control,” in Proc. 2023 IEEE Automatic Speech Recognition and Understand- ing Workshop (ASRU), 2023

    2023

  22. [30]

    Statistics of natural reverberation enable perceptual separation of sound and space,

    J. Traer and J. H. McDermott, “Statistics of natural reverberation enable perceptual separation of sound and space,” Proc. National Academy of Sciences of the United States of America , vol. 113, no. 48, pp. E7856–E7865, 2016

    2016

  23. [31]

    JVS corpus: free Japanese multi-speaker voice corpus,

    S. Takamichi, K. Mitsui, Y . Saito, T. Koriyama, N. Tanji, and H. Saruwatari, “JVS corpus: free Japanese multi-speaker voice corpus,” arXiv [cs.SD] 1908.06248, 2019

    1908 arXiv

  24. [32]

    Construction and anal- ysis of phonetically and prosodically balanced emotional speech database,

    E. Takeishi, T. Nose, Y . Chiba, and A. Ito, “Construction and anal- ysis of phonetically and prosodically balanced emotional speech database,” in Proc. 2016 Conference of The Oriental Chapter of International Committee for Coordination and Standardization of Speech Databases ...

    2016

  25. [33]

    Web-scale k-means clustering,

    D. Sculley, “Web-scale k-means clustering,” in Proc. 19th Inter- national Conference on World Wide Web, 2010

    2010

  26. [34]

    Auto- matic speech recognition of Japanese dialects using large-scale self-supervised learning models (in Japanese),

    D. Takizawa, T. Nakamura, H. Suda, and S. Fukayama, “Auto- matic speech recognition of Japanese dialects using large-scale self-supervised learning models (in Japanese),” in Proc. 2025 Spring Meeting of the Acoustical Society of Japan, 2025

    2025

  27. [35]

    SpeechBrain: A general-purpose speech toolkit,

    M. Ravanelli, T. Parcollet, P. Plantinga, A. Rouhe, S. Cornell, L. Lugosch, C. Subakan, N. Dawalatabad, A. Heba, J. Zhong, J.-C. Chou, S.-L. Yeh, S.-W. Fu, C.-F. Liao, E. Rastorgueva, F. Grondin, W. Aris, H. Na, Y . Gao, R. D. Mori, and Y . Ben- gio, “SpeechBrain: A general-pu...

    2021 arXiv

  28. [36]

    Conformer: Convolution-augmented transformer for speech recognition,

    A. Gulati, J. Qin, C.-C. Chiu, N. Parmar, Y . Zhang, J. Yu, W. Han, S. Wang, Z. Zhang, Y . Wu, and R. Pang, “Conformer: Convolution-augmented transformer for speech recognition,” in Proc. Interspeech 2020, 2020, pp. 5036–5040

    2020

  29. [37]

    Construction of a large-scale Japanese ASR corpus on TV recordings,

    S. Ando and H. Fujihara, “Construction of a large-scale Japanese ASR corpus on TV recordings,” in Proc. 2021 IEEE Interna- tional Conference on Acoustics, Speech and Signal Processing (ICASSP), 2021, pp. 6948–6952

    2021

Pith tools

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