Pith. sign in

REVIEW 2 major objections 6 minor 38 references

RapFlow-TTS: Rapid and High-Fidelity Text-to-Speech with Improved Consistency Flow Matching

T0 review · 2 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read RapFlow-TTS trains a text-to-speech acoustic model with velocity consistency constraints on a straightened flow-matching trajectory, claiming natural speech in only 2 solver steps.

desk verdict A genuinely useful application of consistency FM to TTS with credible two-step quality, but the paper over-attributes the speed-up to the consistency objective when the largest measured gains come from adversarial learning. read the letter →

arxiv 2506.16741 v1 pith:JSVJBD22 submitted 2025-06-20 eess.AS cs.AI

Hyun Joon Park , Jeongmin Liu , Jin Sob Kim , Jeong Yeol Yang , Sung Won Han , Eunwoo Song This is my paper · ORCID
classification eess.AScs.AI
keywords rapidtext-to-speechconsistencymodelflowmatchingvelocityadversariallearningfew-stepsynthesisacousticODE-basedgenerativemodels
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

RapFlow-TTS is a text-to-speech acoustic model that tries to break the usual trade-off between sound quality and generation speed in ODE-based neural speech synthesis. Its proposal is to train the flow-matching decoder with a consistency objective on the velocity field, so that points on a straightened ODE trajectory are mapped to the same endpoint, and then to generate speech with a two-step Euler solve. The paper argues that because flow-matching trajectories are already nearly straight, consistency training works far better than on the complicated trajectories of diffusion models. Experiments on LJSpeech and VCTK report naturalness comparable to strong many-step baselines at only two function evaluations, with the claimed 5- to 10-fold reduction in synthesis steps. The practical upshot is fast, high-fidelity speech synthesis at a speed comparable to single-step non-ODE TTS systems.

What carries the argument

The load-bearing object is the multi-segment velocity-consistency objective of Eq. (4). The time range $[0,1]$ is split into $S=2$ segments; within each segment, two losses are applied: a straight-flow loss $\mathcal{L}_{sf}$ comparing $f^i_\theta(t,x_t,\mu)$ with the same function evaluated at $t+\Delta t$, asking that the predicted endpoint of the segment be the same for nearby starting points, and a velocity-consistency loss $\mathcal{L}_{vc}$ that forces the velocity vectors themselves to agree. The prediction function is $f^i_\theta(t,x_t,\mu)=x_t+((i+1)/S-t)\,v^i_\theta(t,x_t,\mu)$, which extrapolates along the learned velocity to the segment endpoint. Training proceeds in two stages: first only the straight-flow loss to the ground-truth endpoint, then the full consistency objective, and finally adversarial learning on segment endpoints. The consistency constraint is what allows a single two-step Euler solve to approximate the whole trajectory.

What would settle it

Measure the consistency error directly: generate mel-spectrograms from two nearby starting times on the same trajectory and compute $\|f_\theta(t,x_t,\mu)-f_\theta(t+\Delta t,x_{t+\Delta t},\mu)\|$ for the trained model; if this distance does not shrink to near zero across the trajectory, the velocity-consistency objective is not doing the claimed work. A cheaper check is to retrain the model with the consistency loss removed while keeping adversarial learning and compare two-step MOS, since the paper's own ablation suggests the adversarial endpoint loss accounts for much of the quality gain.

Watch

Extended reading notes

Core claim

The central claim is that enforcing consistency in the velocity field along the flow-matching ODE lets an acoustic model produce mel-spectrograms whose quality no longer collapses when the number of sampling steps drops to two. Concretely, the paper uses the consistency flow matching objective with two segments, built on the conditional flow-matching architecture used by other fast TTS systems, and adds four techniques: encoder freezing, shared dropout, a pseudo-Huber metric, delta scheduling, and adversarial learning on segment endpoints. With all techniques, RapFlow-TTS reaches a mean opinion score of 4.01 on LJSpeech and a word error rate of 3.11 percent at NFE=2, compared with, for example, 3.32 MOS at NFE=2 and 3.83 MOS at NFE=10 for the standard flow-matching baseline. The authors interpret this as evidence that building a consistency model on a straight path, rather than on a curvy diffusion path, is the key to few-step quality.

Load-bearing premise

The claim rests on the assumption that the two-segment velocity-consistency objective really makes nearby starting points on the straightened trajectory converge to the same final speech representation, so that a two-step Euler solve reaches the target distribution.

Editorial extensions

If this is right

  • If the central claim is right, high-quality ODE-based TTS no longer needs 10 to 25 solver steps; two-step synthesis reaches the quality of many-step baselines.
  • The two-stage recipe of straight-flow pretraining, then consistency training, then adversarial fine-tuning gives a reusable schedule for making consistency models work outside image generation.
  • Because consistency is built on already-straight flow-matching paths, the method should transfer to other conditional generation tasks that use optimal-transport flow matching with near-linear trajectories.
  • The reported real-time factor at NFE=2 is on the order of 0.03 on a single GPU, making the acoustic model's inference speed comparable to common single-step TTS systems.
  • The multi-segment design suggests a direct scaling path: more segments allow piecewise-linear fits to more complex distributions, at the cost of more network outputs per sample.

Reading between the lines

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

  • Editorial inference: because the paper's ablation shows the largest jump in naturalness comes from the adversarial endpoint loss, a plausible reading is that much of the two-step quality is driven by adversarial sharpening of segment endpoints rather than by the consistency objective alone.
  • Editorial inference: the paper never directly measures the consistency error it claims to enforce; a quantitative check of how close $f_\theta(t,x_t,\mu)$ and $f_\theta(t+\Delta t,x_{t+\Delta t},\mu)$ actually are along generated trajectories would separate the consistency mechanism from the other training techniques.
  • Editorial inference: the choice of $S=2$ segments is a compromise between trajectory flexibility and optimization difficulty; testing larger $S$ on larger multi-speaker corpora would reveal whether the piecewise-linear consistency approach continues to scale.
  • Editorial inference: if the consistency-on-straight-paths hypothesis is correct, the same approach could reduce sampling steps for other flow-matching-based generative models in audio and beyond, whenever the conditional distribution is complex enough to need more than a single straight line.
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

2 major / 6 minor

Summary. This paper proposes RapFlow-TTS, a text-to-speech acoustic model based on consistency flow matching (FM). The model is trained with a multi-segment consistency FM objective consisting of a straight-flow loss and a velocity-consistency loss, followed by several training techniques: encoder freezing, shared dropout, pseudo-Huber loss, delta scheduling, and multi-segment adversarial learning. Experiments on LJSpeech and VCTK compare RapFlow-TTS with Grad-TTS, VoiceFlow, Matcha-TTS, Comospeech, and FastSpeech2, using WER, MOS, and NISQA. The authors report that two-step generation with RapFlow-TTS achieves naturalness close to the ground truth and outperforms the baselines at equal step count, yielding a claimed 5- to 10-fold reduction in synthesis steps relative to prior ODE-based systems.

Significance. If the reported results hold, RapFlow-TTS would be a practical contribution: it demonstrates that a consistency-FM acoustic model with auxiliary training techniques can produce high-fidelity speech in two ODE steps, a regime where standard flow-matching and score-based baselines degrade. The paper's strengths include system-level evaluation against multiple external baselines, per-technique ablations, evaluation on both single- and multi-speaker corpora, and the inclusion of a demo/code link. The main caveat is that the marginal contribution of the velocity-consistency loss itself is not cleanly isolated; the largest ablation gain comes from the adversarial objective, and the explicit consistency term is weighted very small. This does not invalidate the system-level result, but it means the paper's mechanistic claim about consistency FM is not yet fully supported.

major comments (2)
  1. [Section 4.3, Table 2] The central claim that consistency FM enables the 2-step quality is not supported by a clean control. Model (G) (adversarial learning) yields the largest NISQA improvement (from 3.78 to 4.19), larger than the consistency FM gain from (A) to (B) (from 3.46 to 3.71), and no experiment trains a non-consistency FM baseline with the same auxiliary techniques. Additionally, the velocity-consistency term L_vc in Eq. (4) is weighted by alpha=1e-5, with no ablation of alpha. Please add a control (e.g., stage-1 straight flow or Matcha-TTS trained with the same multi-segment adversarial loss, delta scheduling, shared dropout, and Huber loss) and report an alpha sweep for L_vc.
  2. [Section 4.1, Tables 1-3] The MOS evaluation uses only 20 participants and 20 utterances per system, and no significance tests are reported for either MOS or WER. Several key comparisons have overlapping 95% confidence intervals, e.g., RapFlow-TTS† WER 3.11±0.76 vs. Matcha-TTS 10-step WER 3.28±0.78, and in Table 2 the NISQA difference between (G) and (H) is 4.19±0.04 vs. 4.25±0.04. Please report paired significance tests or effect sizes for the main comparisons to justify the use of the word "significant."
minor comments (6)
  1. [Section 3.1, Eq. (4)] The superscript i on f_i_theta and v_i_theta is never defined in the text; please clarify whether these are separate network heads per segment or a single network conditioned on segment index.
  2. [Section 4.3, Table 2] The row labeled "linear→exp schedule" is not described in the text; please define what exponential scheduling means in this context.
  3. [Section 4.1] The sentence "We additionally used 150 and 50 epochs for adversarial learning, respectively" is ambiguous because the antecedent of "respectively" is not explicit; please rephrase to state the per-dataset epochs clearly.
  4. [Throughout] There are typographical and formatting inconsistencies, including "Matha-TTS" in Section 4.2, "V oiceFlow" with extra spaces, and reference entries with irregular capitalization (e.g., Ref. [14] "K. FATRAS", Ref. [22] "marco cuturi"); please correct these.
  5. [Section 1] The claim that "none of which [the techniques] have been explored in the context of consistency FM" is strong and would benefit from softening unless a systematic literature search is reported.
  6. [Section 4.1, MOS test] The manuscript does not state whether the listening test was conducted under controlled conditions or whether participants were native English speakers; please provide these details for reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: RapFlow-TTS's speed/quality claims are externally benchmarked; its training objectives are regression and adversarial losses, and the only self-citation is incidental.

full rationale

RapFlow-TTS's central claim, that consistency-flow-matching training plus auxiliary losses yields high naturalness at NFE=2, is supported by external benchmarks (Matcha-TTS, VoiceFlow, Grad-TTS, Comospeech) and by MOS/WER measurements on LJSpeech and VCTK. The consistency objective in Eq. (4) is a training loss defined as regression between segment endpoint estimates and velocity estimates; it is not used to define the evaluation metric, so the reported improvement is not true by construction. The improved techniques (shared dropout, pseudo-Huber loss, delta scheduling, adversarial learning) are validated by ablations against NISQA/MOS, not by restating their definitions as predictions. The only self-citation is ref. [10] (DEX-TTS) in the introductory list of ODE-based TTS models; it is incidental and nothing in the derivation depends on it. The consistency-FM formulation is imported from external ref. [20], and the architecture from Matcha-TTS [19]. The absence of a non-consistency FM control trained with the same adversarial and scheduling techniques weakens mechanistic attribution, but that is an experimental support gap, not circularity. No equation reduces to its inputs, no fitted parameter is renamed as a prediction, and no load-bearing argument is carried by self-citation.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new entities. It relies on the prior flow matching and consistency model frameworks and on several hand-set hyperparameters. No prediction reduces to a fitted constant by the paper's own equations, but the central result depends on empirically chosen values for S, alpha, dropout, delta scheduling, and adversarial weights, several of which were selected based on NISQA, the same kind of metric used for the headline quality claims.

free parameters (5)
  • Consistency loss weight alpha = 1e-5
    Balances straight flow loss and velocity consistency loss in Eq. (4); set by hand and not swept.
  • Multi-segment count S = 2
    Number of equal time segments for the piecewise linear trajectory; affects how well the consistency path can represent complex distributions.
  • Delta scheduling start, end, and bin count = Delta_t from 0.1 to 0.001 over K=8 intervals
    Controls the bias-variance trade-off of the consistency objective during training; linear schedule chosen after comparison with exponential in Table 2.
  • Shared dropout ratio = 0.05 on LJSpeech, 0.1 on VCTK
    Dropout applied with the same random mask to both velocity branches; values differ per dataset and were set empirically.
  • Adversarial objective mixing ratio = 3:1:2 for Lcfm, Ladv, Lfm
    Weights for the final training stage; set by hand rather than derived.
assumptions (4)
  • standard math Flow matching with a linear interpolation probability path learns the true marginal vector field under standard regularity assumptions.
    Used implicitly in Sections 2 and 3.1 as the basis for the straight ODE objective.
  • domain assumption The Matcha-TTS text encoder and MAS alignment give conditioning features of sufficient quality, and freezing them during consistency training does not degrade the acoustic model.
    Section 3.1 and 3.2 adopt the Matcha-TTS network and encoder freeze; no analysis of alignment errors is provided.
  • ad hoc to paper A piecewise linear trajectory with S=2 segments can approximate the mel-spectrogram data distribution well enough for two-step generation.
    Eq. (4) assumes the optimal transport path can be segmented into two straight pieces; the paper gives no distributional or theoretical justification for S=2.
  • ad hoc to paper The consistency FM losses, after two-stage training, enforce sufficient global self-consistency along the trajectory.
    This is the load-bearing premise of the method in Section 3.1; Table 2 shows the largest quality gain comes from adversarial learning, so this premise is only weakly supported.

how reviews work

0 comments
Cite this review

Pith. "Pith review of RapFlow-TTS: Rapid and High-Fidelity Text-to-Speech with Improved Consistency Flow Matching." pith.science (2026). https://pith.science/paper/JSVJBD22

@misc{pith2026250616741,
  author       = {Pith},
  title        = {Pith review of: RapFlow-TTS: Rapid and High-Fidelity Text-to-Speech with Improved Consistency Flow Matching},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JSVJBD22}},
  note         = {Machine review of arXiv:2506.16741}
}
read the original abstract

We introduce RapFlow-TTS, a rapid and high-fidelity TTS acoustic model that leverages velocity consistency constraints in flow matching (FM) training. Although ordinary differential equation (ODE)-based TTS generation achieves natural-quality speech, it typically requires a large number of generation steps, resulting in a trade-off between quality and inference speed. To address this challenge, RapFlow-TTS enforces consistency in the velocity field along the FM-straightened ODE trajectory, enabling consistent synthetic quality with fewer generation steps. Additionally, we introduce techniques such as time interval scheduling and adversarial learning to further enhance the quality of the few-step synthesis. Experimental results show that RapFlow-TTS achieves high-fidelity speech synthesis with a 5- and 10-fold reduction in synthesis steps than the conventional FM- and score-based approaches, respectively.

Figures

Figures reproduced from arXiv: 2506.16741 by the authors.

Figure 1
Figure 1. Overview of RapFlow-TTS with multi-segment consistency flow matching. truth vector field is applied as follows: LFM = ||vθ(t, xt) − u(t, xt)||2 2, (2) where vθ(t, xt) denotes the learnable vector field network. In contrast to the FM, consistency FM trains the vector field by ensuring that any point on the trajectory reaches the same endpoint with the same velocity. Let the linearly interpolated noise sample at time … view at source ↗

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

38 extracted references · 30 canonical work pages

  1. [1]

    Among the various generative modeling approaches [4, 5, 6, 7, 8], ordinary differential equations (ODE)-based models [9, 10] have become strong solutions for outstanding TTS

    Introduction Text-to-speech (TTS), also known as speech synthesis, aims to synthesize high-fidelity speech, given an input text [1, 2, 3]. Among the various generative modeling approaches [4, 5, 6, 7, 8], ordinary differential equations (ODE)-based models [9, 10] have become strong solutions for outstanding TTS. One rep- resentative method involves diffus...

    work page

  2. [2]

    Consistency Flow Matching Here, we review the consistency FM [20] method for training the RapFlow-TTS. Given a time indext∈[0,1], sampled from a uniform distribution, FM learns a ground truth vector field ut to build the probability pathp t(xt)from the random noise x0∼p 0 to target datax 1∼p 1. The vector field is defined as an ODE for a flowϕ t, whereϕ t...

    work page Pith review arXiv 2025

  3. [3]

    TTS with Consistency Flow Matching For RapFlow-TTS, we follow the network design of Matcha- TTS [19] thanks to its fast and lightweight properties

    RapFlow-TTS 3.1. TTS with Consistency Flow Matching For RapFlow-TTS, we follow the network design of Matcha- TTS [19] thanks to its fast and lightweight properties. As shown in Figure 1, RapFlow-TTS consists of a text encoder, an aligner, and a flow matching decoder. The text encoder extracts the con- text representations from the input text, and the alig...

    work page

  4. [4]

    Experiment Setup DatasetTo verify RapFlow-TTS, we conducted experiments on the LJSpeech [27] and VCTK [28] datasets

    Experiments 4.1. Experiment Setup DatasetTo verify RapFlow-TTS, we conducted experiments on the LJSpeech [27] and VCTK [28] datasets. The LJSpeech is a single-speaker English dataset containing about 24 hours of audio clips. For the dataset split, we followed the settings of previous work in [2]. The VCTK is a multi-speaker English dataset composed of abo...

    work page

  5. [5]

    Based on consistency FM, RapFlow-TTS constructs a consistency model on a straight flow

    Conclusion In this study, we proposed RapFlow-TTS, the ODE-based TTS model capable of fast and high-fidelity speech synthesis. Based on consistency FM, RapFlow-TTS constructs a consistency model on a straight flow. By leveraging both the straight trajec- tory and the velocity-consistency property, high-quality speech synthesis was achievable in significan...

    work page

  6. [6]

    Leaders in INdustry-university Co- operation 3.0

    Acknowledgements This research was supported by V oice team, NA VER Cloud, Republic of Korea. This research was supported by the BK21 FOUR funded by the Ministry of Education of Korea and Na- tional Research Foundation of Korea. This research was also results of a study on the “Leaders in INdustry-university Co- operation 3.0” Project, supported by the Mi...

    work page

  7. [7]

    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,” inProc. ICLR, 2021

    work page 2021

  8. [8]

    Tacotron: Towards end-to-end speech synthesis,

    Y . Wang, R. Skerry-Ryan, D. Stanton, Y . Wu, R. J. Weiss, N. Jaitly, Z. Yang, Y . Xiao, Z. Chen, S. Bengio, Q. V . Le, Y . Agiomyrgiannakis, R. Clark, and R. A. Saurous, “Tacotron: Towards end-to-end speech synthesis,” inProc. Interspeech, 2017, pp. 4006–4010

    work page 2017

Show all 38 references
  1. [9]

    Natural TTS synthesis by conditioning Wavenet on mel spectrogram predic- tions,

    J. Shen, R. Pang, R. J. Weiss, M. Schuster, N. Jaitly, Z. Yang, Z. Chen, Y . Zhang, Y . Wang, R. Skerrv-Ryanet al., “Natural TTS synthesis by conditioning Wavenet on mel spectrogram predic- tions,” inProc. ICASSP, 2018, pp. 4779–4783

    2018

  2. [10]

    Glow-TTS: A generative flow for text-to-speech via monotonic alignment search,

    J. Kim, S. Kim, J. Kong, and S. Yoon, “Glow-TTS: A generative flow for text-to-speech via monotonic alignment search,” inProc. NeurIPS, vol. 33, 2020, pp. 8067–8077

    2020

  3. [11]

    FastSpeech: Fast, robust and controllable text to speech,

    Y . Ren, Y . Ruan, X. Tan, T. Qin, S. Zhao, Z. Zhao, and T.-Y . Liu, “FastSpeech: Fast, robust and controllable text to speech,” in Proc. NeurIPS, vol. 32, 2019

    2019

  4. [12]

    Flow- TTS: A non-autoregressive network for text to speech based on flow,

    C. Miao, S. Liang, M. Chen, J. Ma, S. Wang, and J. Xiao, “Flow- TTS: A non-autoregressive network for text to speech based on flow,” inProc. ICASSP, 2020, pp. 7209–7213

    2020

  5. [13]

    Neural speech synthesis with Transformer network,

    N. Li, S. Liu, Y . Liu, S. Zhao, M. Liu, and M. Zhou, “Neural speech synthesis with Transformer network,” inProc. AAAI, 2019, pp. 6706–6713

    2019

  6. [14]

    Improving and generalizing flow-based generative models with minibatch optimal transport,

    A. Tong, K. FATRAS, N. Malkin, G. Huguet, Y . Zhang, J. Rector- Brooks, G. Wolf, and Y . Bengio, “Improving and generalizing flow-based generative models with minibatch optimal transport,” Transactions on Machine Learning Research, 2024

    2024

  7. [15]

    Diff-TTS: A denoising diffusion model for text-to-speech,

    M. Jeong, J. Kim, J. Kong, and J. Son, “Diff-TTS: A denoising diffusion model for text-to-speech,” inProc. Interspeech, 2021, pp. 3605–3609

    2021

  8. [16]

    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,” inProc. ICML, 2021, pp. 8599–8608

    2021

  9. [17]

    DEX-TTS: Diffusion-based EXpressive Text-to-Speech with style modeling on time variability,

    H. J. Park, J. S. Kim, W. Shin, and S. W. Han, “DEX-TTS: Diffusion-based EXpressive Text-to-Speech with style modeling on time variability,”arXiv preprint arXiv:2406.19135, 2024

    2024 arXiv

  10. [18]

    Score-based generative modeling through stochas- tic differential equations,

    Y . Song, J. Sohl-Dickstein, D. P. Kingma, A. Kumar, S. Ermon, and B. Poole, “Score-based generative modeling through stochas- tic differential equations,” inProc. ICLR, 2021

    2021

  11. [19]

    Elucidating the design space of diffusion-based generative models,

    T. Karras, M. Aittala, T. Aila, and S. Laine, “Elucidating the design space of diffusion-based generative models,” inProc. NeurIPS, vol. 35, 2022, pp. 26 565–26 577

    2022

  12. [20]

    Flow matching for generative modeling,

    Y . Lipman, R. T. Q. Chen, H. Ben-Hamu, M. Nickel, and M. Le, “Flow matching for generative modeling,” inProc. ICLR, 2023

    2023

  13. [21]

    Improved techniques for training con- sistency models,

    Y . Song and P. Dhariwal, “Improved techniques for training con- sistency models,” inProc. ICLR, 2024

    2024

  14. [22]

    V oiceflow: Efficient text-to-speech with rectified flow matching,

    Y . Guo, C. Du, Z. Ma, X. Chen, and K. Yu, “V oiceflow: Efficient text-to-speech with rectified flow matching,” inProc. ICASSP, 2024, pp. 11 121–11 125

    2024

  15. [23]

    Flow straight and fast: Learning to generate and transfer data with rectified flow,

    X. Liu, C. Gong, and qiang liu, “Flow straight and fast: Learning to generate and transfer data with rectified flow,” inProc. ICLR, 2023

    2023

  16. [24]

    Consistency models,

    Y . Song, P. Dhariwal, M. Chen, and I. Sutskever, “Consistency models,” inProc. ICML, 2023, pp. 32 211–32 252

    2023

  17. [25]

    Furthermore, we extend it to multi-segment adversarial learning for consistency FM

    on the mel-spectrogram levels using a Conv2d discrimina- tor [26]. Furthermore, we extend it to multi-segment adversarial learning for consistency FM. Letx i andˆxi be the ground-truth and estimated endpoint at each segment, where each is obtained by(i+ 1)/S×x 1 + (1−(i+ 1)/S)...

  18. [26]

    Como- speech: One-step speech and singing voice synthesis via consis- tency model,

    Z. Ye, W. Xue, X. Tan, J. Chen, Q. Liu, and Y . Guo, “Como- speech: One-step speech and singing voice synthesis via consis- tency model,” inProc. ACM Multimedia, 2023, pp. 1831–1839

    2023

  19. [27]

    Matcha-TTS: A fast tts architecture with conditional flow match- ing,

    S. Mehta, R. Tu, J. Beskow, ´E. Sz ´ekely, and G. E. Henter, “Matcha-TTS: A fast tts architecture with conditional flow match- ing,” inProc. ICASSP, 2024, pp. 11 341–11 345

    2024

  20. [28]

    Consistency flow matching: Defining straight flows with velocity consistency,

    L. Yang, Z. Zhang, Z. Zhang, X. Liu, M. Xu, W. Zhang, C. Meng, S. Ermon, and B. Cui, “Consistency flow matching: Defining straight flows with velocity consistency,”arXiv preprint arXiv:2407.02398, 2024

    2024 arXiv

  21. [29]

    Simple reflow: Improved techniques for fast flow models,

    B. Kim, Y .-G. Hsieh, M. Klein, marco cuturi, J. C. Ye, B. Kawar, and J. Thornton, “Simple reflow: Improved techniques for fast flow models,” inProc. ICLR, 2025

    2025

  22. [30]

    Improving the training of rectified flows,

    S. Lee, Z. Lin, and G. Fanti, “Improving the training of rectified flows,” inProc. NeurIPS, 2024

    2024

  23. [31]

    Least squares generative adversarial networks,

    X. Mao, Q. Li, H. Xie, R. Y . Lau, Z. Wang, and S. Paul Smolley, “Least squares generative adversarial networks,” inProc. ICCV, 2017, pp. 2794–2802

    2017

  24. [32]

    Improved techniques for training gans,

    T. Salimans, I. Goodfellow, W. Zaremba, V . Cheung, A. Radford, and X. Chen, “Improved techniques for training gans,” inProc. NeurIPS, vol. 29, 2016

    2016

  25. [33]

    StyleTTS: A style-based generative model for natural and diverse text-to-speech synthesis,

    Y . A. Li, C. Han, and N. Mesgarani, “StyleTTS: A style-based generative model for natural and diverse text-to-speech synthesis,” arXiv preprint arXiv:2205.15439, 2022

    2022 arXiv

  26. [34]

    The LJ speech dataset,

    K. Ito and L. Johnson, “The LJ speech dataset,” https://keithito. com/LJ-Speech-Dataset/, 2017

    2017

  27. [35]

    CSTR VCTK cor- pus: English multi-speaker corpus for CSTR voice cloning toolkit (version 0.92),

    J. Yamagishi, C. Veaux, and K. MacDonald, “CSTR VCTK cor- pus: English multi-speaker corpus for CSTR voice cloning toolkit (version 0.92),” 2019

    2019

  28. [36]

    HiFi-GAN: Generative adversarial networks for efficient and high fidelity speech synthesis,

    J. Kong, J. Kim, and J. Bae, “HiFi-GAN: Generative adversarial networks for efficient and high fidelity speech synthesis,” inProc. NeurIPS, vol. 33, 2020, pp. 17 022–17 033

    2020

  29. [37]

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

    A. Radford, J. W. Kim, T. Xu, G. Brockman, C. McLeavey, and I. Sutskever, “Robust speech recognition via large-scale weak su- pervision,” inProc. ICML, 2023, pp. 28 492–28 518

    2023

  30. [38]

    NISQA: A deep CNN-self-attention model for multidimensional speech quality prediction with crowdsourced datasets,

    G. Mittag, B. Naderi, A. Chehadi, and S. M ¨oller, “NISQA: A deep CNN-self-attention model for multidimensional speech quality prediction with crowdsourced datasets,” inProc. Inter- speech, 2021, pp. 2127–2131

    2021

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.