Pith. sign in

REVIEW 5 major objections 6 minor 37 references

EME-TTS: Unlocking the Emphasis and Emotion Link in Speech Synthesis

T0 review · 5 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read EME-TTS keeps stressed words clear across five emotions.

desk verdict Plausible new combination of emphasis control and emotional TTS, but the unvalidated pseudo-labels and small subjective effects make the central claim conditional. read the letter →

arxiv 2507.12015 v1 pith:UBITUMJO submitted 2025-07-16 cs.SD eess.AS

Haoxun Li , Leyuan Qu , Jiaxi Hu , Taihao Li This is my paper · ORCID
classification cs.SDeess.AS
keywords EmotionalSpeechSynthesisEmphasisControlEmotionExpressivenessWeaklySupervisedLabelingVariance-BasedProsodyAttentionModulationExpressiveTTS
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

The paper sets out to show that emotional tone and word emphasis can be controlled in a single speech synthesis model instead of being handled separately. It builds an emotional text-to-speech system trained with automatically generated emphasis labels and prosodic variance features, and adds an emphasis-perception block that strengthens the attention given to designated words. The reported experiments indicate that this combination makes emotional speech more expressive and keeps listeners able to identify the intended stressed word across angry, happy, sad, surprise, and neutral conditions. If the claim holds, a user could specify both the emotion and the words to stress, and the model would deliver natural speech without emphasis being drowned out by the emotion.

What carries the argument

The load-bearing mechanism is the Emphasis Perception Enhancement (EPE) block, a transformer-style block that re-weights attention in the regions marked as emphasized. Its Emphasis Adapter adds a scaled mask to the attention weights at the start and end of the emphasized word, so the model can enhance perceivable prominence through attention rather than by directly shaping pitch or energy. The same pipeline also uses variance-based pitch and duration features, computed as local averages in emphasized regions minus sentence-level averages, which tell the acoustic model how far to push prosody above the sentence norm.

What would settle it

Hand-label a small sample of the emotional test set for emphasis, compute per-emotion agreement between human labels and the pseudo-labeler, and check whether emphasis recognition accuracy drops when the model is trained with human labels instead of pseudo-labels; if the pseudo-labels disagree sharply on surprise or happy speech, the reported emphasis control partly measures pitch-range artifacts rather than true prominence.

Watch

Extended reading notes

Core claim

The central claim is that there is a learnable, controllable link between emphasis and emotion, and that EME-TTS exploits it. The framework takes emotion labels and emphasized-word positions as two conditioning signals. During training, emphasis positions are obtained as pseudo-labels from a pre-trained emphasis recognizer; the model then predicts pitch and duration variance features that highlight the emphasized region, while the Emphasis Perception Enhancement (EPE) block re-weights its attention with a mask over those positions and integrates emotion through conditional normalization. At inference, a large language model is used to choose the emphasis positions from text and emotion. The paper reports that the resulting speech is perceived as more emotionally expressive, especially with surrounding context, and that target emphasis remains recognizable across all five emotions, with the largest gains on surprise, where pitch rises at sentence end otherwise mislead listeners.

Load-bearing premise

The training pipeline assumes that the pre-trained emphasis recognizer's automatic labels on the emotional speech database are accurate enough to act as ground truth for where emphasis actually falls.

Editorial extensions

If this is right

  • Listeners can identify the intended emphasized word from synthesized speech across all five emotions, with mean recognition accuracy rising from 0.73 without the EPE block to 0.78 with it.
  • Emphasis control improves perceived emotion accuracy: subjective emotion recognition averaged 0.67 for EME-TTS versus 0.58 for the base emotional TTS model.
  • The expressiveness gain is strongest when the generated sentence appears with surrounding context, suggesting emphasis needs semantic support to enhance emotion.
  • Emphasis does not degrade synthesis quality: the model's MOS and NISQA scores sit above the base model and above the version without the EPE block.
  • The increased duration of emphasized regions is credited with improving objective recognition of sad emotion, which rose to 0.61 from 0.54 for the base model.

Reading between the lines

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

  • The paper fixes the emphasis strength at 0.2; a natural extension would be to expose that scalar as a user dial and measure whether emphasis clarity and emotion accuracy trade off as it varies.
  • Because the LLM's emphasis predictions are not separately evaluated, one could test how sensitive the emotion outcomes are to prediction errors by comparing LLM-chosen positions against human-chosen positions on the same test sentences.
  • The same weakly supervised pipeline could be applied to other expressive dimensions, such as contrastive focus or question focus, if a suitable recognizer supplies pseudo-labels.
  • An interesting open question is whether emphasis positions themselves differ by emotion in natural speech; the current design treats them as an input rather than learning emotion-specific emphasis tendencies.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 6 minor

Summary. The paper proposes EME-TTS, an emotional text-to-speech system built on EmoSpeech, which adds variance-based emphasis features and an Emphasis Perception Enhancement (EPE) block. Emphasis positions are obtained as pseudo-labels by running the EmphaClass recognizer on the ESD dataset, and the model is conditioned on both emotion labels and emphasis positions. At inference, an unspecified large language model predicts emphasis positions from text and emotion. The experiments compare EME-TTS with and without the EPE block, EmoSpeech, and CosyVoice2 on emphasis recognition accuracy, emotion recognition accuracy, expressiveness preference, and MOS/NISQA. The central claim is that EME-TTS, combined with LLM-based emphasis prediction, yields more natural emotional speech while preserving stable and distinguishable target emphasis across emotions.

Significance. The interaction between emphasis control and emotion in TTS is a timely and underexplored problem, and the proposed two-question framing is useful. If the claims were fully supported, EME-TTS would provide a practical control interface that couples emphasis positions and emotion labels in a single acoustic model, and the EPE block would be a concrete contribution. The manuscript also has several strengths: a multi-task evaluation protocol, inclusion of objective and subjective measures, and a public demo page with synthesized samples. However, the current evidence is not yet sufficient to establish the central claim: the pseudo-labeling step is unvalidated on the target corpus, the LLM component is unspecified, the listener-level statistics are missing, and no emphasis-aware baseline is included. These gaps are load-bearing because they affect the validity of the emphasis and emotion accuracy results.

major comments (5)
  1. [§2.2, Eqs. (3)-(6), Eq. (10)] Section 2.2 asserts that EmphaClass yields 'highly reliable emphasis pseudo-labels' on ESD, but no validation on ESD is reported. These labels are used as hard supervision for the variance targets in Eqs. (3)-(6) and for the EPE mask in Eq. (10), and they also define the expected answers in the Emphasis Accuracy Test. ESD is acted emotional speech, so prosodic events such as surprise pitch rises or phrase-final lengthening could be systematically misclassified as emphasis by a recognizer trained on EmphAssess. Please report per-emotion agreement between EmphaClass and human annotation on a held-out ESD subset, including inter-annotator statistics and error analysis; without this, the supervision and the EAT evaluation rest on an unverified transfer assumption.
  2. [§3.2.2 and Abstract] The paper's central claim is that EME-TTS 'when combined with large language models for emphasis position prediction' produces more natural emotional speech, but the LLM component is not specified: no prompt template, model version, or inference procedure is given, and no evaluation of the predicted emphasis positions is reported. As a result, the EAT and EEPT results cannot be cleanly attributed to the acoustic model; they may reflect LLM quality rather than the proposed architecture. Please provide the exact LLM prompt and a quantitative evaluation of the predicted emphasis positions against human judgments or a held-out EmphaClass reference.
  3. [Tables 1 and 3, §3.2.1-3.2.2] The main support for the emphasis and emotion claims consists of small differences in accuracy scores measured with only 11 listeners and 80 items per task, but the paper reports no confidence intervals, significance tests, or listener-level variance. For example, the EAT mean is 0.78 vs. 0.73, and the EAT-EMO mean is 0.67 vs. 0.58, with per-emotion values that fluctuate considerably (e.g., happy accuracy 0.32 in Table 3). Please report paired significance tests (e.g., Wilcoxon signed-rank or bootstrap) and confidence intervals for each table, and state the chance level and the number of candidate words in the EAT task.
  4. [§3.2.1 and Tables 1-3] No emphasis-aware baseline is included. The comparison to EmoSpeech and CosyVoice2 addresses emotion accuracy, but the paper's central emphasis-control claim is supported only by the ablation of the EPE block. To isolate the contribution of the EPE block and the variance-based emphasis modeling, the EAT should also be run on an existing emphasis-controllable TTS system (e.g., the variance-feature approach of [19] adapted to the same emotion-conditioned backbone) under identical conditions. Without such a baseline, the reader cannot tell whether the reported emphasis accuracy comes from the proposed components or from the underlying FastSpeech2/EmoSpeech backbone.
  5. [§2.3, §2.4, §3.1] The emphasis intensity control relies on two free parameters: the strength value in the Emphasis Adapter (set to 0.2) and the normalization range [0,2] for variance features. No sensitivity analysis or justification is given for either choice. Since these parameters directly control the magnitude of emphasis modulation, a small sweep (or at least a rationale based on preliminary listening) is necessary to establish that the reported gains over the w/o EPE variant are not an artifact of a particular parameter setting.
minor comments (6)
  1. [Table 1] The per-emotion numbers in Table 1 are visually misaligned; the column boundaries are unclear, making the table hard to read.
  2. [§3.2.2] The abbreviations EAT and EAT-EMO are easily confused; renaming the emotion task (e.g., ERT for emotion recognition test) would improve clarity.
  3. [§3.2.4] The MOS and NISQA results are reported only as overall means; given the paper's emphasis on per-emotion effects, a per-emotion breakdown would be more informative.
  4. [§4] The sentence 'EME-TTS demonstrate its effectiveness' contains a subject-verb agreement error; it should be 'demonstrates'.
  5. [General] The paper does not release code or trained models; the online demo page is useful but not sufficient for reproducibility of the reported quantitative results.
  6. [References] Reference [29] is the GPT-4 technical report, but the text never states which LLM and version was actually used for inference; this should be stated explicitly.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's conditions, losses, and evaluations are anchored to external data, external recognizers, and independent listener judgments rather than to the claims being tested.

full rationale

The derivation chain is self-contained relative to the claims. Emphasis positions come from EmphaClass [22], an external recognizer trained on EmphAssess, not from the proposed model or from the authors' own prior work; emotion labels are fixed dataset labels from ESD [23]. The variance features in Eqs. (1)-(2) are computed from target pitch/duration statistics inside pseudo-labeled emphasis regions, and the losses in Eqs. (3)-(6) are ordinary supervised regression losses: the targets P_tar, D_tar, PV_tar, and DV_tar are derived from data, not from the model's own output. The EPE adjustment in Eq. (10) is a conditioning mechanism, not a quantity claimed as a prediction. Evaluation is independent: EAT asks human listeners to identify emphasized words in shuffled samples, EAT-EMO uses both listener judgments and the external Emotion2vec-plus-large classifier, and MOS/NISQA measure quality. No equation defines its output as the evaluation metric, no fitted parameter is renamed as a prediction, no load-bearing result rests on a self-citation, and no uniqueness theorem is imported from the authors' prior work. The lack of validation of EmphaClass pseudo-labels on ESD and the unspecified LLM prompt are genuine evidence-quality and correctness concerns, but they concern whether the inputs are accurate, not whether the claimed outputs are equivalent to the inputs by construction. Under the given rules, that is not circularity, so the score is 0.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

No new physical or perceptual entities are introduced. All components are architectural blocks or external models. The central dependence is on external annotations (EmphaClass and LLM) and hand-set scalar parameters, not on a new entity.

free parameters (2)
  • EA strength = 0.2
    Manual scalar added to attention weights in Equation (10); no sensitivity analysis or ablation over strength values is reported.
  • variance feature normalization range = [0,2]
    Pitch and duration variance features are normalized to [0,2] based on data distribution (Section 2.3); this choice affects loss scale and emphasis intensity, with no ablation.
assumptions (3)
  • domain assumption Pitch and duration are the primary acoustic indicators of emphasis and energy can be ignored
    Adopted from [19] in Section 2.3; controls which variance features are modeled and which are omitted.
  • domain assumption EmphaClass predictions transfer to ESD with high accuracy
    Section 2.2 states that pseudo-labels are highly reliable, but no validation, agreement rate, or error analysis on ESD is provided.
  • domain assumption An LLM can infer appropriate emphasis positions from text and emotion label
    Used at inference in Sections 1 and 3.2.2; no prompt details or accuracy evaluation for the LLM are reported.

how reviews work

0 comments
Cite this review

Pith. "Pith review of EME-TTS: Unlocking the Emphasis and Emotion Link in Speech Synthesis." pith.science (2026). https://pith.science/paper/UBITUMJO

@misc{pith2026250712015,
  author       = {Pith},
  title        = {Pith review of: EME-TTS: Unlocking the Emphasis and Emotion Link in Speech Synthesis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UBITUMJO}},
  note         = {Machine review of arXiv:2507.12015}
}
read the original abstract

In recent years, emotional Text-to-Speech (TTS) synthesis and emphasis-controllable speech synthesis have advanced significantly. However, their interaction remains underexplored. We propose Emphasis Meets Emotion TTS (EME-TTS), a novel framework designed to address two key research questions: (1) how to effectively utilize emphasis to enhance the expressiveness of emotional speech, and (2) how to maintain the perceptual clarity and stability of target emphasis across different emotions. EME-TTS employs weakly supervised learning with emphasis pseudo-labels and variance-based emphasis features. Additionally, the proposed Emphasis Perception Enhancement (EPE) block enhances the interaction between emotional signals and emphasis positions. Experimental results show that EME-TTS, when combined with large language models for emphasis position prediction, enables more natural emotional speech synthesis while preserving stable and distinguishable target emphasis across emotions. Synthesized samples are available on-line.

Figures

Figures reproduced from arXiv: 2507.12015 by the authors.

Figure 1
Figure 1. The entire framework of our proposed model. (a) is the overall architecture diagram. (b) and (c) show the detailed structure of variance adapter and Emphasis Perception Enhancement (EPE) block, respectively. 2. Proposed Method 2.1. Overview The overall architecture of EME-TTS is shown in Figure 1a. We use EmoSpeech [9] as the base architecture of the acoustic model, using the embedding of emotions and the position o… view at source ↗
Figure 2
Figure 2. Effect of Emphasis on Emotional Expressiveness in TTS Models. 3.2.4. Speech Quality and Naturalness We assess the quality and naturalness of the synthesized speech through both objective and subjective evaluations. Objectively, we utilize the NISQA library [31] to predict naturalness scores ratings on a 5-point scale. Subjectively, participants completed task 4 (MOS Rating), in which they rated 100 randomly shuf￾fle… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

37 extracted references · 33 canonical work pages

  1. [19]

    Daft- exprt: Cross-speaker prosody transfer on any text for expressive speech synthesis,

    J. Za ¨ıdi, H. Seut´e, B. van Niekerk, and M.-A. Carbonneau, “Daft- exprt: Cross-speaker prosody transfer on any text for expressive speech synthesis,” in Interspeech 2022, 2022, pp. 4591–4595

    work page 2022

  2. [1]

    EME-TTS: Unlocking the Emphasis and Emotion Link in Speech Synthesis

    Introduction With the advancement of deep learning, Text-to-Speech (TTS) systems have significantly improved in terms of quality, clarity, and naturalness, leveraging architectures such as Transformers [1, 2], normalizing flows [3, 4], and diffusion models [5, 6]. However, conventional TTS systems often struggle with ex- pressiveness, producing monotonal ...

    work page Pith review arXiv 2025

  3. [2]

    Overview The overall architecture of EME-TTS is shown in Figure 1a

    Proposed Method 2.1. Overview The overall architecture of EME-TTS is shown in Figure 1a. We use EmoSpeech [9] as the base architecture of the acoustic model, using the embedding of emotions and the position of the emphasis as conditions. These conditions are obtained through the emotion label and the emphasis pseudo label respectively. EME-TTS consists of...

    work page

  4. [3]

    Experiments 3.1. Experimental Setup For our experiments, EME-TTS utilizes the English portion of the Emotional Speech Database (ESD) [23], which com- prises recordings from 10 speakers across five emotions: an- gry, happy, sad, surprise, and neutral. Each speaker contributes 350 utterances per emotion, resulting in approximately 1,750 utterances and 1.2 h...

    work page

  5. [4]

    Conclusion This paper presents EME-TTS, a framework that explores how emphasis enhances emotional expressiveness and how to main- tain its perceptual clarity and stability across emotions. By leveraging variance-based emphasis features, weakly super- vised learning, and EPE, EME-TTS demonstrate its effective- ness in generating emotionally expressive spee...

    work page

  6. [5]

    LQN25F020001), and in part by the Key R&D Program of Zhejiang (2025C01104)

    Acknowledgements This work was supported in part by the Scientific Research Staring Foundation of Hangzhou Institute for Advanced Study (2024HIASC2001), in part by Zhejiang Provincial Natural Sci- ence Foundation of China (No. LQN25F020001), and in part by the Key R&D Program of Zhejiang (2025C01104)

    work page

  7. [6]

    Naturalspeech: End-to-end text-to- speech synthesis with human-level quality,

    X. Tan, J. Chen, H. Liu, J. Cong, C. Zhang, Y . Liu, X. Wang, Y . Leng, Y . Yi, L. Heet al., “Naturalspeech: End-to-end text-to- speech synthesis with human-level quality,”IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024

    work page 2024

  8. [7]

    Attention is all you need,

    A. Waswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. Gomez, L. Kaiser, and I. Polosukhin, “Attention is all you need,” in NIPS, 2017

    work page 2017

Show all 37 references
  1. [8]

    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. Liu, “Fastspeech 2: Fast and high-quality end-to-end text to speech,” in 9th International Conference on Learning Representations, ICLR 2021, Virtual Event, Austria, May 3-7, 2021 . OpenReview.net, 2021

    2021

  2. [9]

    Glow-tts: A genera- tive flow for text-to-speech via monotonic alignment search,

    J. Kim, S. Kim, J. Kong, and S. Yoon, “Glow-tts: A genera- tive flow for text-to-speech via monotonic alignment search,” Ad- vances in Neural Information Processing Systems , vol. 33, pp. 8067–8077, 2020

    2020

  3. [10]

    Conditional variational autoencoder with adversarial learning for end-to-end text-to-speech,

    J. Kim, J. Kong, and J. Son, “Conditional variational autoencoder with adversarial learning for end-to-end text-to-speech,” in Inter- national Conference on Machine Learning . PMLR, 2021, pp. 5530–5540

    2021

  4. [11]

    Grad-tts: A diffusion probabilistic model for text-to-speech,

    V . Popov, I. V ovk, V . Gogoryan, T. Sadekova, and M. Kudinov, “Grad-tts: A diffusion probabilistic model for text-to-speech,” in International Conference on Machine Learning . PMLR, 2021, pp. 8599–8608

    2021

  5. [12]

    Msemotts: Multi-scale emotion transfer, prediction, and control for emotional speech synthesis,

    Y . Lei, S. Yang, X. Wang, and L. Xie, “Msemotts: Multi-scale emotion transfer, prediction, and control for emotional speech synthesis,” IEEE/ACM Transactions on Audio, Speech, and Lan- guage Processing, vol. 30, pp. 853–864, 2022

    2022

  6. [13]

    K. Lee. ”Expressive-FastSpeech2”. [Online]. Available: https: //github.com/keonlee9420/Expressive-FastSpeech2

  7. [14]

    Exploring transfer learning for low resource emotional tts,

    N. Tits, K. El Haddad, and T. Dutoit, “Exploring transfer learning for low resource emotional tts,” in Intelligent Systems and Appli- cations: Proceedings of the 2019 Intelligent Systems Conference (IntelliSys) V olume 1. Springer, 2020, pp. 52–60

    2019

  8. [15]

    Emospeech: guiding fastspeech2 to- wards emotional text to speech,

    D. Diatlova and V . Shutov, “Emospeech: guiding fastspeech2 to- wards emotional text to speech,” in 12th ISCA Speech Synthesis Workshop (SSW2023), 2023, pp. 106–112

    2023

  9. [16]

    Emodiff: Intensity con- trollable emotional text-to-speech with soft-label guidance,

    Y . Guo, C. Du, X. Chen, and K. Yu, “Emodiff: Intensity con- trollable emotional text-to-speech with soft-label guidance,” in ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) . IEEE, 2023, pp. 1–5

    2023

  10. [17]

    Controllable emotion trans- fer for end-to-end speech synthesis,

    T. Li, S. Yang, L. Xue, and L. Xie, “Controllable emotion trans- fer for end-to-end speech synthesis,” in 2021 12th International Symposium on Chinese Spoken Language Processing (ISCSLP) . IEEE, 2021, pp. 1–5

    2021

  11. [18]

    Supervised and unsupervised approaches for controlling narrow lexical focus in sequence-to-sequence speech synthesis,

    S. Shechtman, R. Fernandez, and D. Haws, “Supervised and unsupervised approaches for controlling narrow lexical focus in sequence-to-sequence speech synthesis,” in 2021 IEEE Spoken Language Technology Workshop (SLT) . IEEE, 2021, pp. 431– 437

    2021

  12. [20]

    The acoustics of word stress in en- glish as a function of stress level and speaking style,

    A. Eriksson and M. Heldner, “The acoustics of word stress in en- glish as a function of stress level and speaking style,” in 16th An- nual Conference of the International Speech Communication As- sociation (INTERSPEECH 2015), Dresden, Germany, September 6-10, 2015, 2015, pp. 41–45

    2015

  13. [21]

    The acoustics of lexical stress in italian as a func- tion of stress level and speaking style,

    A. Eriksson, P. M. Bertinetto, M. Heldner, R. Nodari, and G. Lenoci, “The acoustics of lexical stress in italian as a func- tion of stress level and speaking style,” inInterspeech 2016. The International Speech Communication Association (ISCA), 2016, pp. 1059–1063

    2016

  14. [22]

    Lex- ical stress perception as a function of acoustic properties and the native language of the listener,

    A. Eriksson, R. Nodari, J. ˇSimko, A. Suni, and M. Vainio, “Lex- ical stress perception as a function of acoustic properties and the native language of the listener,” in The 10th International Conference on Speech Prosody: Communicative and Interactive Prosody. ISCA, 2020, pp. 449–453

    2020

  15. [23]

    Emphatic speech generation with conditioned input layer and bidirectional lstms for expressive speech synthesis,

    R. Li, Z. Wu, Y . Huang, J. Jia, H. Meng, and L. Cai, “Emphatic speech generation with conditioned input layer and bidirectional lstms for expressive speech synthesis,” in 2018 IEEE Interna- tional Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2018, pp....

    2018

  16. [24]

    to compute prominence scores via Continuous Wavelet Transform (CWT) using pitch, energy, and duration signals, we leverage the EmphAssess dataset and the EmphaClass empha- sis recognizer [22]. EmphaClass fine-tunes a pre-trained Self- Supervised Learning (SSL) speech model for...

  17. [25]

    Emphasis control for parallel neural tts,

    S. Seshadri, T. Raitio, D. Castellani, and J. Li, “Emphasis control for parallel neural tts,” inInterspeech 2022, 2022, pp. 3378–3382

    2022

  18. [26]

    Ee-tts: Emphatic expressive tts with linguistic information,

    Y . Zhong, C. Zhang, X. Liu, C. Sun, W. Deng, H. Hu, and Z. Sun, “Ee-tts: Emphatic expressive tts with linguistic information,” in Interspeech 2023, 2023, pp. 4873–4877

    2023

  19. [27]

    A survey of large language models,

    W. X. Zhao, K. Zhou, J. Li, T. Tang, X. Wang, Y . Hou, Y . Min, B. Zhang, J. Zhang, Z. Dong et al., “A survey of large language models,” arXiv preprint arXiv:2303.18223, 2023

    2023 arXiv

  20. [28]

    Em- phAssess : a prosodic benchmark on assessing emphasis transfer in speech-to-speech models,

    M. de Seyssel, A. D’Avirro, A. Williams, and E. Dupoux, “Em- phAssess : a prosodic benchmark on assessing emphasis transfer in speech-to-speech models,” in Proceedings of the 2024 Con- ference on Empirical Methods in Natural Language Processing , 2024, pp. 495–507

    2024

  21. [29]

    Emotional voice con- version: Theory, databases and esd,

    K. Zhou, B. Sisman, R. Liu, and H. Li, “Emotional voice con- version: Theory, databases and esd,”Speech Communication, vol. 137, pp. 1–18, 2022

    2022

  22. [30]

    Hierarchical rep- resentation and estimation of prosody using continuous wavelet transform,

    A. Suni, J. ˇSimko, D. Aalto, and M. Vainio, “Hierarchical rep- resentation and estimation of prosody using continuous wavelet transform,” Computer Speech & Language , vol. 45, pp. 123–136, 2017

    2017

  23. [31]

    Adaspeech 4: Adaptive text to speech in zero-shot scenar- ios,

    Y . Wu, X. Tan, B. Li, L. He, S. Zhao, R. Song, T. Qin, and T.-Y . Liu, “Adaspeech 4: Adaptive text to speech in zero-shot scenar- ios,” in Interspeech 2022, 2022, pp. 2568–2572

    2022

  24. [32]

    istftnet: Fast and lightweight mel-spectrogram vocoder incorporating inverse short-time fourier transform,

    T. Kaneko, K. Tanaka, H. Kameoka, and S. Seki, “istftnet: Fast and lightweight mel-spectrogram vocoder incorporating inverse short-time fourier transform,” in ICASSP 2022-2022 IEEE Inter- national Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2022, pp. ...

    2022

  25. [33]

    Adam: A method for stochastic optimization,

    P. K. Diederik, “Adam: A method for stochastic optimization,” ICLR, 2015

    2015

  26. [34]

    Cosyvoice 2: Scalable stream- ing speech synthesis with large language models,

    Z. Du, Y . Wang, Q. Chen, X. Shi, X. Lv, T. Zhao, Z. Gao, Y . Yang, C. Gao, H. Wanget al., “Cosyvoice 2: Scalable stream- ing speech synthesis with large language models,” arXiv preprint arXiv:2412.10117, 2024

    2024 arXiv

  27. [35]

    GPT-4 technical report,

    J. Achiam, S. Adler, S. Agarwal, L. Ahmad, I. Akkaya, F. L. Ale- man, D. Almeida, J. Altenschmidt, S. Altman, S. Anadkat et al., “GPT-4 technical report,”arXiv preprint arXiv:2303.08774, 2023

    2023 arXiv

  28. [36]

    emotion2vec: Self-supervised pre-training for speech emotion representation,

    Z. Ma, Z. Zheng, J. Ye, J. Li, Z. Gao, S. Zhang, and X. Chen, “emotion2vec: Self-supervised pre-training for speech emotion representation,” in Findings of the Association for Computational Linguistics: ACL 2024 , 2024, pp. 15 747–15 760

    2024

  29. [37]

    Deep learning based assessment of syn- thetic speech naturalness,

    G. Mittag and S. M ¨oller, “Deep learning based assessment of syn- thetic speech naturalness,” in Interspeech 2020, 2020, pp. 1748– 1752

    2020

Pith tools

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