Pith. sign in

REVIEW 4 major objections 4 minor 88 references

Luna-TTS Family Technical Report

T0 review · 4 major / 4 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read A 0.6B masked-diffusion TTS tops autoregressive and commercial rivals on all four Seed-TTS-Eval metrics, and a block-causal variant streams audio with a 41.6 ms first-block latency.

desk verdict A solid diffusion-TTS systems report whose architecture and streaming recipe are genuinely new and useful, but whose 'best on all four' benchmark headline rests on quoted baselines and thin margins; treat the numbers as conditional until same-protocol re-evaluation. read the letter →

arxiv 2608.11593 v1 pith:DKGOB72B submitted 2026-08-12 cs.SD eess.AS

Feng Yin , Shuai Shi , Junjie Zheng , Kechenying Zhou , Yiqiu Wang , Chenyang He , Qiuhua Jiang , Mengxiao Bi
show 14 more authors
Yanmin Qian Mingxin Chen Xun Gong Tianteng Gu Bing Han Peng Jiang Chenda Li Haiyang Sun Han Wang Wei Wang Yi Wang Leying Zhang Wangyou Zhang Chushu Zhou
This is my paper · ORCID
classification cs.SDeess.AS
keywords text-to-speechmaskeddiffusionnon-autoregressiveblock-causalstreamingzero-shotvoicecloningRVQtokengridlanguagemodelspeechediting
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 aims to show that TTS does not need an autoregressive codec language model: a 0.6B-parameter masked-diffusion model, pretrained on one million hours of Chinese, English, Japanese, and Korean speech, can match or beat leading autoregressive and commercial systems on standard zero-shot quality benchmarks, and a block-causal sibling can stream audio at interactive latency. If the reported results hold, the structural costs of AR decoding — latency growing with utterance length, error accumulation along a committed prefix, and the artificial left-to-right order imposed on the residual vector-quantization (RVQ) token grid — are not unavoidable. On Seed-TTS-Eval the fully parallel variant reports the best result on all four metrics (0.73 CER / 79.7 SIM for Mandarin, 1.49 WER / 76.8 SIM for English), while the streaming variant commits its first 1.28-second block in 41.6 ms with an end-to-end RTF of 0.0240. Because both variants share one tokenizer, data pipeline, and backbone, the family also provides a controlled comparison of fully parallel versus block-autoregressive diffusion decoding.

What carries the argument

The central machinery is the absorbing-state masked-diffusion objective over the RVQ token grid: each target grid position is independently replaced by a MASK token with probability t, and the model is trained to recover every masked position in parallel. Each frame's Q codebook embeddings are summed into one sequence position, with Q parallel classification heads predicting all codebooks at once, and unrestricted random masking during training lets confidence-based iterative unmasking commit tokens in any order across time and codebook depth. Block-causal attention, with per-block independent denoising and KV caching, converts the same objective into streaming generation. The progressive attention swap from causal to bidirectional to block-causal is what lets the family inherit text competence from a pretrained AR LLM rather than learning it from speech data alone.

What would settle it

Run the official Seed-TTS-Eval toolkit on the released Luna-TTS checkpoint and re-score every quoted baseline under identical prompts, duration handling, and scoring code; if any baseline then reaches or beats 0.73 CER / 79.7 SIM on Mandarin and 1.49 WER / 76.8 SIM on English under that single protocol, the paper's best-on-all-four claim is refuted.

Watch

Extended reading notes

Core claim

On its own terms, the paper claims that masked acoustic-token diffusion, scaled to production size, is at parity with or better than autoregressive codec language models for zero-shot synthesis and is practical for streaming. The load-bearing evidence is architectural: an RVQ tokenizer with eight codebooks at 25 Hz, whose first codebook is semantically anchored; a shared diffusion backbone that corrupts the whole T×Q token grid by random masking and predicts all masked positions in parallel; and a progressive adaptation path that converts a pretrained autoregressive text LLM into the bidirectional Luna-TTS and then into the block-causal Luna-TTS Realtime. As a result, zero-shot voice cloning and speech editing fall out as infilling, expressive control rides on text tokens, reinforcement-learning post-training is defined over the realized denoising trajectory, and the two decoding regimes are deployment choices rather than different systems.

Load-bearing premise

The headline 'best on all four metrics' rests on the assumption that numbers quoted from other systems' papers were produced under the same evaluation protocol — same scoring toolkit, test splits, and prompt settings — as the authors' own runs; if not, the ranking can be an artifact of protocol mismatch even if the model itself is genuinely strong.

Editorial extensions

If this is right

  • Quality no longer requires left-to-right decoding: with 32 parallel refinement steps, Luna-TTS reports 0.73 CER / 79.7 SIM on Mandarin and 1.49 WER / 76.8 SIM on English, best in its comparison on all four Seed-TTS-Eval metrics.
  • Latency stops growing with utterance length for the fully parallel variant, which reports an end-to-end RTF of 0.0211 at 16 steps; the streaming variant commits a 1.28s block in 41.6 ms and completes a 10.6s waveform in 254.0 ms at 8 steps with parallel classifier-free guidance.
  • Zero-shot voice cloning and speech editing are instances of infilling on the same any-order model, with no task-specific mechanisms.
  • The block-causal variant keeps most of the quality, trading roughly 0.3 CER/WER and 3 SIM points against Luna-TTS, and still beats several offline-evaluated systems on the four-language average of CV3-Eval.
  • GRPO-style reinforcement learning transfers to masked diffusion by computing policy ratios over realized denoising decisions, with a lexicographic reward that puts content fidelity before speaker similarity.

Reading between the lines

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

  • Because the paper's own table marks most baseline rows as quoted from other reports, the architecture claim that non-AR diffusion reaches production quality is separable from the exact ranking; a protocol-matched re-run against the quoted commercial systems would settle the ranking.
  • The matched tokenizer, data, and backbone lineage can serve as a reusable testbed: without retraining, one can sweep block size, step count, and CFG parallelization to map latency–quality frontiers for new languages or hardware.
  • The paper attributes the streaming variant's larger hard in-the-wild gap to irrevocable per-block commits and its learned end-of-speech decision; an observable prediction is that giving block-causal decoding the ability to revise committed blocks, or integrating duration prediction into the backbone, would shrink the hard-en gap from the reported 13.98 toward the offline 6.18.
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

4 major / 4 minor

Summary. The paper describes Luna-TTS Family, two diffusion-language-model TTS systems derived from a pretrained 0.6B text LLM (Qwen3-0.6B) by progressive adaptation of the attention pattern: Luna-TTS, a fully non-autoregressive masked-diffusion model that generates an entire RVQ token grid (25 Hz, 8 codebooks) in 32 parallel refinement steps, and Luna-TTS Realtime, a block-causal streaming variant that decodes 1.28-s blocks in parallel while being autoregressive across blocks. Both variants share a common tokenizer, data pipeline, and backbone lineage, are pretrained on approximately 1 million hours of Chinese, English, Japanese, and Korean speech, and are post-processed with annealed fine-tuning for emotion/NVV control and a GRPO-style RL stage over the denoising trajectory. The main empirical claims are: on Seed-TTS-Eval, Luna-TTS achieves the lowest CER/WER and highest SIM among compared systems (zh: 0.73 CER / 79.7 SIM; en: 1.49 WER / 76.8 SIM); on CV3-Eval it has the lowest Mandarin and English error rates in the comparison; and Luna-TTS Realtime achieves 41.6 ms first-block latency and an end-to-end RTF of 0.0240. The paper also reports expressive-control results against commercial systems and an internal Arena evaluation for dedicated voices.

Significance. If the reported zero-shot results are reproducible under a uniform protocol, the paper would be a significant data point: it would show that a 0.6B masked-diffusion TTS can reach production-level zero-shot quality, that block diffusion can serve streaming workloads at very low latency, and that progressive adaptation from an AR text LLM is a credible route to NAR speech models. The RL extension to masked diffusion is potentially useful to the community. The paper is also commendably explicit about several limitations, including the non-significance of the Arena top ranking and the weak Korean subset performance. However, the current evaluation does not yet establish the headline claims: most Seed-TTS-Eval baselines are quoted rather than re-measured, the RL reward models are undisclosed and may overlap with the evaluation metrics, the expressive-control tables lack uncertainty quantification, and the RL contribution is not isolated by an ablation. These issues are fixable and do not appear to invalidate the architectural contributions.

major comments (4)
  1. [§6.1, Table 8] The headline claim that Luna-TTS is best on all four Seed-TTS-Eval metrics is supported by a comparison in which most baselines are quoted from their original papers (marked ⋄) and only Qwen3-TTS-12Hz-1.7B-Base was re-evaluated under the same protocol. The closest margins (zh SIM 79.7 vs 79.6 for Seed-TTS; en SIM 76.8 vs 76.2; en WER 1.49 vs 1.54) are comparable to run-to-run or toolkit-version variation. Please re-evaluate the nearest competitors (Seed-TTS, Qwen-Audio-3.0-TTS, VoxCPM2, OmniVoice) with the same scoring code and prompt settings, and report per-run variance or bootstrap confidence intervals so that the claim can be assessed. Without this, the 'best on all four' conclusion rests on a protocol mismatch that the report itself acknowledges may exist.
  2. [§4.2 and §6.1] The RL post-training in §4.2 optimizes rewards defined by WER and speaker similarity against frozen ASR and SV models, and the evaluation in §6.1 reports Whisper/Paraformer WER/CER and WavLM SIM. The reward models are not named in §4.2. If the reward models are the same as the evaluation models, then the reported Seed-TTS-Eval gains could reflect overfitting to those specific scorers rather than generalizable quality improvements. Please disclose the reward ASR/SV models and add an evaluation with a held-out ASR model and a different speaker-embedding model to demonstrate that the improvements are not an artifact of matching the metric models.
  3. [§4.1] The RL surrogate in §4.1 computes importance ratios over the token distribution πθ(a|... , M_k) and omits the probability of the realized masking-position set M_k. Yet the paper's own decoding procedure (§2.5) selects positions using annealed Gumbel noise, which is a stochastic, θ-dependent process. Unless the position-selection distribution is treated as part of the environment (e.g., by fixing the noise), the surrogate L_RL is not an unbiased estimator of the policy gradient for the full generation process. Moreover, no ablation is shown that isolates the effect of the RL stage on the final Seed-TTS-Eval numbers. Please justify the approximation or modify the surrogate, and provide an with/without RL comparison.
  4. [§6.2] The expressive-control comparisons in Tables 10–13 do not report sample sizes, confidence intervals, or statistical tests. The human listening test is described only as 'randomized, blinded' with no listener count, stimulus count, or inter-rater reliability, and some reported differences (e.g., E-MOS 3.90 vs 3.82 in Table 13) are small. Given the abstract's claim of 'best results on most ... human-rated metrics', these summary statistics are insufficient to sustain the claim. Please provide the experimental design details and uncertainty quantification, or soften the claim accordingly.
minor comments (4)
  1. [Table 5] The 'Steps' column in Table 5 combines the step count and GPU count with a space (e.g., '32 1×H20'); consider separating these into two columns for clarity.
  2. [Abstract] The abstract says 'more than 40×real-time' but the actual RTF 0.0240 corresponds to about 41.7× real-time; the phrasing '40×' is fine but the missing space after the multiplication sign is a typo.
  3. [References] Several references contain an errant space in 'Voice' (e.g., refs [26] and [51]) and the text uses 'V oxCPM' in places; the correct term should be 'Voice' and 'VoxCPM'.
  4. [§2.5] The paper does not state the number of parameters in the duration predictor when claiming '0.6B' for the family; since the duration predictor is a separate Qwen3-0.6B, the total inference parameter count for Luna-TTS is larger than 0.6B. Clarify whether '0.6B' refers to the backbone only.

Circularity Check

1 steps flagged · score 2.0 of 10

No definitional circularity; only a mild training-objective/evaluation-metric overlap and a dependence on quoted baselines.

  1. fitted input called prediction [§4.2 (RL reward) and §6.1/Table 8 (Seed-TTS-Eval results)]
    "Each generated waveform is evaluated by frozen models for automatic speech recognition and speaker verification. ... The two signals are converted into a prompt-local ordinal reward, Rg,i = Ψg(−WERg,i, SIMg,i). ... The Luna-TTS Family rows report the results after RL post-training of §4."

    The headline Seed-TTS-Eval metrics (WER/CER and SIM) are the same constructs that the RL stage explicitly optimized as its utterance-level reward. This is not an equation-level reduction: Seed-TTS-Eval is a held-out benchmark, the official toolkit aggregation differs from the prompt-local ordinal reward, and the report does not state that the frozen ASR/SV models used in RL are identical to the evaluation ones. Still, the four headline numbers are partially self-referential: the model was trained to minimize WER and maximize SIM, and is then scored on those very quantities. The independence of those specific metric values is therefore reduced, though the architecture claims and the CV3-Eval comparisons retain independent content.

full rationale

The paper is an empirical technical report rather than a derivation chain. The progressive-adaptation recipe (Qwen3-0.6B -> bidirectional masked diffusion -> block-causal diffusion) follows external, cited prior work ([37], [41], [42], [43], [48]); no load-bearing step is justified by a self-citation or by an author-imported uniqueness theorem. The closest thing to self-citation is the use of WavLM [54], where the corresponding author is a co-author of WavLM, but WavLM is a public, externally validated model used both for codec distillation and for SIM evaluation; this creates a minor metric-alignment caveat rather than circularity. The headline 'best on all four Seed-TTS-Eval metrics' rests mostly on baseline numbers quoted from other papers (marked with a diamond), which the paper itself concedes 'may differ in evaluation details'; that is a comparability threat, not a circularity. The only genuine self-dependence is the RL/evaluation overlap described in the step: optimizing WER/SIM rewards and then reporting WER/SIM. Because the benchmark is external and held out, the results are not forced by construction, so the appropriate finding is no significant circularity, with a score of 2 reflecting the mild training-objective/evaluation-metric overlap.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

The system's quality claims are empirical. The main formal assumptions are the masked-diffusion upper bound used for training, the transferability of AR text-LLM weights to diffusion, and the validity of ASR, speaker-verification, and model-as-judge metrics as proxies for TTS quality. Hyperparameters such as block size, step count, codebook shape, and learning rate are hand-chosen engineering choices rather than fitted physical constants. No new physical or formal entities are postulated.

free parameters (7)
  • block_size_B = 32 codec frames (1.28 s)
    Hand-chosen streaming granularity; directly trades first-block latency, KV-cache size, and per-block quality in Sections 2.6 and 5.2.
  • refinement_steps_S = 32 for Luna-TTS; 16 or 8 for Realtime
    Inference-time quality-speed dial; fixed by the authors, not derived from theory.
  • codebook_config = Q=8 codebooks, 2048 entries, 25 Hz
    Luna-Codec design choice defining the token grid; reconstruction quality and LM complexity depend on it.
  • peak_learning_rate = 2.1e-4, cosine annealed to 0.1x
    Training hyperparameter from Section 3.2; not fit by theory.
  • cfg_guidance = text-conditioning dropout 0.1 during training
    Classifier-free guidance drop rate chosen to sharpen text adherence at low step counts (Section 2.5).
  • masking_and_decoding_schedule = uniform t in (0,1]; cosine confidence schedule with top-k, temperature, Gumbel annealing
    Decoding policy family under which RL importance ratios are computed; exact settings not made public.
  • grpo_hyperparameters = group size G>=2; sigma_min; epsilon clipping; lexicographic reward ranking
    RL objective details in Section 4; exact constants and the full reward composition are not specified in the report.
assumptions (5)
  • standard math The masked-diffusion loss in Eq. (2) upper-bounds the negative log-likelihood of the data.
    Cited to Sahoo et al. [34]; used to justify the training objective in Section 2.3.
  • domain assumption AR-pretrained LLM weights transfer to bidirectional and block-diffusion decoders and train substantially faster than from scratch.
    Used in Sections 2.4 and 2.6, citing [37, 42, 43] and block diffusion [41]. This is an empirical claim inherited from prior work.
  • domain assumption ASR and speaker-verification benchmarks (Whisper-large-v3, Paraformer-zh, WavLM, ERes2Net) are valid proxies for TTS quality.
    The entire zero-shot evaluation in Section 6.1 relies on these frozen models.
  • domain assumption Gemini 3.1 Pro model-as-a-judge ratings are valid perceptual measures for NVV and emotion control.
    Used for LALM-based evaluations in Section 6.2; no human calibration of the judge scores is reported.
  • domain assumption The cross-ASR data filtering pipeline selects high-quality training data without systematically biasing the model toward evaluation benchmarks.
    Section 3.1 rejects cuts with cross-ASR disagreement; the effect of this filter on downstream benchmark scores is not separately measured.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Luna-TTS Family Technical Report." pith.science (2026). https://pith.science/paper/DKGOB72B

@misc{pith2026260811593,
  author       = {Pith},
  title        = {Pith review of: Luna-TTS Family Technical Report},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DKGOB72B}},
  note         = {Machine review of arXiv:2608.11593}
}
read the original abstract

Modern text-to-speech (TTS) is dominated by autoregressive (AR) codec language models, whose left-to-right decoding brings latency that grows with utterance length, error accumulation along the committed prefix, and an artificial generation order imposed on the Residual Vector Quantization (RVQ) token grid. We propose Luna-TTS Family, diffusion-language-model-based TTS systems pretrained on 1 million hours of speech across Chinese, English, Japanese, and Korean. The family is built by progressive adaptation of a pretrained AR text LLM, from causal to bidirectional and finally to block-causal attention, and comprises two variants sharing a single tokenizer, data pipeline, and 0.6B backbone lineage. Luna-TTS is fully non-autoregressive: it generates the entire RVQ token grid in a fixed number of parallel refinement steps, with zero-shot voice cloning and speech editing arising natively as infilling. Luna-TTS Realtime, derived by continual training, is autoregressive over blocks of 32 codec frames (1.28s) while denoising each block in parallel; it supports KV-cached blockwise generation and incremental audio delivery, achieving an end-to-end RTF of 0.0240 and 41.6 ms local first-block latency under the warmed serving protocol. An annealed fine-tuning stage adds explicit control over emotion and non-verbal vocalizations (NVVs), and a reinforcement-learning stage applies GRPO with policy ratios computed over the realized denoising trajectory. On Seed-TTS-Eval, Luna-TTS achieves the best results on all four metrics among compared open-source and commercial systems (0.73 CER / 79.7 SIM on test-zh, 1.49 WER / 76.8 SIM on test-en); on the harder in-the-wild CV3-Eval, it posts the lowest Mandarin and English error rates in our comparison. Against leading commercial systems, it achieves the best results on most objective, model-based, and human-rated metrics for NVV and emotion control.

Figures

Figures reproduced from arXiv: 2608.11593 by the authors.

Figure 1
Figure 1. Overview of Luna-TTS Family. (a) Luna-Codec, a causal RVQ codec (25 Hz, Q = 8 codebooks) whose first codebook is semantically anchored by WavLM distillation. (b) The shared diffusion LM backbone: text, prompt token grid, and partially masked target token grid form one sequence; per-frame codebook embeddings are summed at the input, and Q parallel heads predict all masked positions in parallel. (c) The two decoding r… view at source ↗
Figure 2
Figure 2. Four-stage multilingual data-processing pipeline. Raw recordings are converted into speaker-consistent, [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

88 extracted references · 35 canonical work pages

  1. [1]

    SoundStream: An end- to-end neural audio codec.IEEE/ACM Transactions on Audio, Speech, and Language Processing, 30:495–507, 2022

    Neil Zeghidour, Alejandro Luebs, Ahmed Omran, Jan Skoglund, and Marco Tagliasacchi. SoundStream: An end- to-end neural audio codec.IEEE/ACM Transactions on Audio, Speech, and Language Processing, 30:495–507, 2022

    2022

  2. [2]

    High fidelity neural audio compression

    Alexandre Défossez, Jade Copet, Gabriel Synnaeve, and Yossi Adi. High fidelity neural audio compression. arXiv preprint arXiv:2210.13438, 2022

    arXiv 2022

  3. [3]

    High-fidelity audio compression with improved RVQGAN

    Rithesh Kumar, Prem Seetharaman, Alejandro Luebs, Ishaan Kumar, and Kundan Kumar. High-fidelity audio compression with improved RVQGAN. InAdvances in Neural Information Processing Systems, 2023

    2023

  4. [4]

    Better speech synthesis through scaling.arXiv preprint arXiv:2305.07243, 2023

    James Betker. Better speech synthesis through scaling.arXiv preprint arXiv:2305.07243, 2023

    arXiv 2023

  5. [5]

    Neural codec language models are zero-shot text to speech synthesizers.arXiv preprint arXiv:2301.02111, 2023

    Chengyi Wang, Sanyuan Chen, Yu Wu, Ziqiang Zhang, Long Zhou, Shujie Liu, Zhuo Chen, Yanqing Liu, Huam- ing Wang, Jinyu Li, Lei He, Sheng Zhao, and Furu Wei. Neural codec language models are zero-shot text to speech synthesizers.arXiv preprint arXiv:2301.02111, 2023

    arXiv 2023

  6. [6]

    Seed-TTS: A family of high-quality versatile speech generation models.arXiv preprint arXiv:2406.02430, 2024

    Philip Anastassiou, Jiawei Chen, Jitong Chen, Yuanzhe Chen, Zhuo Chen, Ziyi Chen, et al. Seed-TTS: A family of high-quality versatile speech generation models.arXiv preprint arXiv:2406.02430, 2024. * Equal contribution. † Corresponding author. ‡ Listed alphabetically by family name. 17 Luna-TTS Family Technical Report

    arXiv 2024

  7. [7]

    CosyV oice: A scalable multilingual zero-shot text-to-speech synthesizer based on supervised semantic tokens.arXiv preprint arXiv:2407.05407, 2024

    Zhihao Du, Qian Chen, Shiliang Zhang, Kai Hu, Heng Lu, Yexin Yang, Hangrui Hu, Siqi Zheng, Yue Gu, Ziyang Ma, Zhifu Gao, and Zhijie Yan. CosyV oice: A scalable multilingual zero-shot text-to-speech synthesizer based on supervised semantic tokens.arXiv preprint arXiv:2407.05407, 2024

    arXiv 2024

  8. [8]

    CosyV oice 2: Scalable streaming speech synthesis with large language models.arXiv preprint arXiv:2412.10117, 2024

    Zhihao Du, Yuxuan Wang, Qian Chen, Xian Shi, Xiang Lv, Tianyu Zhao, Zhifu Gao, Yexin Yang, et al. CosyV oice 2: Scalable streaming speech synthesis with large language models.arXiv preprint arXiv:2412.10117, 2024

    arXiv 2024

Show all 88 references
  1. [9]

    CosyV oice 3: Towards in-the-wild speech generation via scaling-up and post-training.arXiv preprint arXiv:2505.17589, 2025

    Zhihao Du, Changfeng Gao, Yuxuan Wang, Fan Yu, Tianyu Zhao, Hao Wang, Xiang Lv, Hui Wang, et al. CosyV oice 3: Towards in-the-wild speech generation via scaling-up and post-training.arXiv preprint arXiv:2505.17589, 2025

    2025 arXiv

  2. [10]

    MiniMax-Speech: Intrinsic zero-shot text-to-speech with a learnable speaker encoder.arXiv preprint arXiv:2505.07916, 2025

    Bowen Zhang, Congchao Guo, Geng Yang, Hang Yu, Haozhe Zhang, et al. MiniMax-Speech: Intrinsic zero-shot text-to-speech with a learnable speaker encoder.arXiv preprint arXiv:2505.07916, 2025

    2025 arXiv

  3. [11]

    Llasa: Scaling train-time and inference-time compute for Llama-based speech synthesis.arXiv preprint arXiv:2502.04128, 2025

    Zhen Ye, Xinfa Zhu, Chi-Min Chan, Xinsheng Wang, Xu Tan, Jiahe Lei, Yi Peng, et al. Llasa: Scaling train-time and inference-time compute for Llama-based speech synthesis.arXiv preprint arXiv:2502.04128, 2025

    2025 arXiv

  4. [12]

    GLM-TTS technical report.arXiv preprint arXiv:2512.14291, 2025

    Jiayan Cui, Zhihan Yang, Naihan Li, Jiankun Tian, Xingyu Ma, Yi Zhang, Guangyu Chen, Runxuan Yang, Zijian Huang, Yuqing Cheng, Yizhi Zhou, Guochen Yu, Xiaotao Gu, and Jie Tang. GLM-TTS technical report.arXiv preprint arXiv:2512.14291, 2025

    2025

  5. [13]

    Qwen3-TTS technical report.arXiv preprint arXiv:2601.15621, 2026

    Hangrui Hu, Xinfa Zhu, Ting He, Dake Guo, Bin Zhang, et al. Qwen3-TTS technical report.arXiv preprint arXiv:2601.15621, 2026

    2026 arXiv

  6. [14]

    Qwen-Audio-3.0-TTS: Freely controllable and highly robust speech synthesis with multi-stage training paradigm.arXiv preprint arXiv:2607.23938, 2026

    Bajian Xiang, Cheng Wen, Han Zhao, Hao Wang, Haoxu Wang, et al. Qwen-Audio-3.0-TTS: Freely controllable and highly robust speech synthesis with multi-stage training paradigm.arXiv preprint arXiv:2607.23938, 2026

    2026 arXiv

  7. [15]

    Fish audio S2 technical report.arXiv preprint arXiv:2603.08823, 2026

    Shijia Liao, Yuxuan Wang, Songting Liu, Yifan Cheng, et al. Fish audio S2 technical report.arXiv preprint arXiv:2603.08823, 2026

    2026

  8. [16]

    MOSS-TTS technical report.arXiv preprint arXiv:2603.18090, 2026

    SII-OpenMOSS Team. MOSS-TTS technical report.arXiv preprint arXiv:2603.18090, 2026

    2026

  9. [17]

    Spark-TTS: An efficient LLM-based text-to-speech model with single-stream decoupled speech tokens.arXiv preprint arXiv:2503.01710, 2025

    Xinsheng Wang, Mingqi Jiang, Ziyang Ma, Ziyu Zhang, Songxiang Liu, Linqin Li, et al. Spark-TTS: An efficient LLM-based text-to-speech model with single-stream decoupled speech tokens.arXiv preprint arXiv:2503.01710, 2025

    2025 arXiv

  10. [18]

    IndexTTS2: A breakthrough in emotionally expressive and duration-controlled auto-regressive zero-shot text-to-speech.arXiv preprint arXiv:2506.21619, 2025

    Siyi Zhou, Yiquan Zhou, Yi He, Xun Zhou, Jinchao Wang, Wei Deng, and Jingchen Shu. IndexTTS2: A breakthrough in emotionally expressive and duration-controlled auto-regressive zero-shot text-to-speech.arXiv preprint arXiv:2506.21619, 2025

    2025 arXiv

  11. [19]

    FireRedTTS: A foundation text-to-speech framework for industry-level generative speech applications.arXiv preprint arXiv:2409.03283, 2024

    Hao-Han Guo, Kun Liu, Fei-Yu Shen, Yi-Chen Wu, Feng-Long Xie, Kun Xie, and Kai-Tuo Xu. FireRedTTS: A foundation text-to-speech framework for industry-level generative speech applications.arXiv preprint arXiv:2409.03283, 2024

    2024 arXiv

  12. [20]

    MiMo-Audio: Audio language models are few-shot learners.arXiv preprint arXiv:2512.23808, 2025

    LLM-Core Xiaomi. MiMo-Audio: Audio language models are few-shot learners.arXiv preprint arXiv:2512.23808, 2025

    2025

  13. [21]

    Step-Audio 2 technical report.arXiv preprint arXiv:2507.16632, 2025

    StepFun Audio Team. Step-Audio 2 technical report.arXiv preprint arXiv:2507.16632, 2025

    2025 arXiv

  14. [22]

    Simple and controllable music generation

    Jade Copet, Felix Kreuk, Itai Gat, Tal Remez, David Kant, Gabriel Synnaeve, Yossi Adi, and Alexandre Défossez. Simple and controllable music generation. InAdvances in Neural Information Processing Systems, 2023

    2023

  15. [23]

    Moshi: A speech-text foundation model for real-time dialogue.arXiv preprint arXiv:2410.00037, 2024

    Alexandre Défossez, Laurent Mazaré, Manu Orsini, Amélie Royer, Patrick Pérez, Hervé Jégou, Edouard Grave, and Neil Zeghidour. Moshi: A speech-text foundation model for real-time dialogue.arXiv preprint arXiv:2410.00037, 2024

    2024 arXiv

  16. [24]

    DiSTAR: Diffusion over a scalable token autoregressive representation for speech generation.arXiv preprint arXiv:2510.12210, 2025

    Yakun Song, Xiaobin Zhuang, Jiawei Chen, et al. DiSTAR: Diffusion over a scalable token autoregressive representation for speech generation.arXiv preprint arXiv:2510.12210, 2025

    2025

  17. [25]

    Yaron Lipman, Ricky T. Q. Chen, Heli Ben-Hamu, Maximilian Nickel, and Matt Le. Flow matching for genera- tive modeling. InInternational Conference on Learning Representations, 2023

    2023

  18. [26]

    V oicebox: Text-guided multilingual universal speech generation at scale

    Matthew Le, Apoorv Vyas, Bowen Shi, Brian Karrer, Leda Sari, Rashel Moritz, Mary Williamson, Vimal Manohar, Yossi Adi, Jay Mahadeokar, and Wei-Ning Hsu. V oicebox: Text-guided multilingual universal speech generation at scale. InAdvances in Neural Information Processing Systems, 2023

    2023

  19. [27]

    E2 TTS: Embarrassingly easy fully non-autoregressive zero-shot TTS

    Sefik Emre Eskimez, Xiaofei Wang, Manthan Thakker, Canrun Li, Chung-Hsien Tsai, Zhen Xiao, Hemin Yang, Zirun Zhu, Min Tang, Xu Tan, Yanqing Liu, Sheng Zhao, and Naoyuki Kanda. E2 TTS: Embarrassingly easy fully non-autoregressive zero-shot TTS. InIEEE Spoken Language Technology...

    2024

  20. [28]

    F5-TTS: A fairytaler that fakes fluent and faithful speech with flow matching.arXiv preprint arXiv:2410.06885, 2024

    Yushen Chen, Zhikang Niu, Ziyang Ma, Keqi Deng, Chunhui Wang, Jian Zhao, Kai Yu, and Xie Chen. F5-TTS: A fairytaler that fakes fluent and faithful speech with flow matching.arXiv preprint arXiv:2410.06885, 2024. 18 Luna-TTS Family Technical Report

    2024 arXiv

  21. [29]

    SoundStorm: Efficient parallel audio generation.arXiv preprint arXiv:2305.09636, 2023

    Zalán Borsos, Matt Sharifi, Damien Vincent, Eugene Kharitonov, Neil Zeghidour, and Marco Tagliasacchi. SoundStorm: Efficient parallel audio generation.arXiv preprint arXiv:2305.09636, 2023

    2023 arXiv

  22. [30]

    NaturalSpeech 3: Zero-shot speech synthesis with factorized codec and diffusion models

    Zeqian Ju, Yuancheng Wang, Kai Shen, Xu Tan, Detai Xin, Dongchao Yang, Yanqing Liu, et al. NaturalSpeech 3: Zero-shot speech synthesis with factorized codec and diffusion models. InInternational Conference on Machine Learning, 2024

    2024

  23. [31]

    MaskGCT: Zero-shot text-to-speech with masked generative codec transformer

    Yuancheng Wang, Haoyue Zhan, Liwei Liu, Ruihong Zeng, Haotian Guo, Jiachen Zheng, Qiang Zhang, Xueyao Zhang, Shunsi Zhang, and Zhizheng Wu. MaskGCT: Zero-shot text-to-speech with masked generative codec transformer. InInternational Conference on Learning Representations, 2025

    2025

  24. [32]

    Johnson, Jonathan Ho, Daniel Tarlow, and Rianne van den Berg

    Jacob Austin, Daniel D. Johnson, Jonathan Ho, Daniel Tarlow, and Rianne van den Berg. Structured denoising diffusion models in discrete state-spaces. InAdvances in Neural Information Processing Systems, 2021

    2021

  25. [33]

    Discrete diffusion modeling by estimating the ratios of the data distribution

    Aaron Lou, Chenlin Meng, and Stefano Ermon. Discrete diffusion modeling by estimating the ratios of the data distribution. InInternational Conference on Machine Learning, 2024

    2024

  26. [34]

    Chiu, Alexan- der Rush, and V olodymyr Kuleshov

    Subham Sekhar Sahoo, Marianne Arriola, Yair Schiff, Aaron Gokaslan, Edgar Marroquin, Justin T. Chiu, Alexan- der Rush, and V olodymyr Kuleshov. Simple and effective masked diffusion language models. InAdvances in Neural Information Processing Systems, volume 37, pages 130136–1...

    2024

  27. [35]

    Jiaxin Shi, Kehang Han, Zhe Wang, Arnaud Doucet, and Michalis K. Titsias. Simplified and generalized masked diffusion for discrete data. InAdvances in Neural Information Processing Systems, 2024

    2024

  28. [36]

    Large language diffusion models.arXiv preprint arXiv:2502.09992, 2025

    Shen Nie, Fengqi Zhu, Zebin You, Xiaolu Zhang, Jingyang Ou, Jun Hu, Jun Zhou, Yankai Lin, Ji-Rong Wen, and Chongxuan Li. Large language diffusion models.arXiv preprint arXiv:2502.09992, 2025

    2025 arXiv

  29. [37]

    Dream 7B: Diffusion large language models.arXiv preprint arXiv:2508.15487, 2025

    Jiacheng Ye, Zhihui Xie, Lin Zheng, Jiahui Gao, Zirui Wu, Xin Jiang, Zhenguo Li, and Lingpeng Kong. Dream 7B: Diffusion large language models.arXiv preprint arXiv:2508.15487, 2025

    2025 arXiv

  30. [38]

    Mercury: Ultra-fast language models based on diffusion.arXiv preprint arXiv:2506.17298, 2025

    Inception Labs, Samar Khanna, Siddhant Kharbanda, Shufan Li, et al. Mercury: Ultra-fast language models based on diffusion.arXiv preprint arXiv:2506.17298, 2025

    2025 arXiv

  31. [39]

    Seed diffusion: A large-scale diffusion language model with high-speed inference.arXiv preprint arXiv:2508.02193, 2025

    Yuxuan Song, Zheng Zhang, Cheng Luo, Pengyang Gao, Fan Xia, et al. Seed diffusion: A large-scale diffusion language model with high-speed inference.arXiv preprint arXiv:2508.02193, 2025

    2025 arXiv

  32. [40]

    LLaDA2.0: Scaling up diffusion language models to 100B

    Tiwei Bie, Maosong Cao, Kun Chen, Lun Du, et al. LLaDA2.0: Scaling up diffusion language models to 100B. arXiv preprint arXiv:2512.15745, 2025

    2025 arXiv

  33. [41]

    Chiu, Zhihan Yang, Zhixuan Qi, Jiaqi Han, Subham Sekhar Sahoo, and V olodymyr Kuleshov

    Marianne Arriola, Aaron Gokaslan, Justin T. Chiu, Zhihan Yang, Zhixuan Qi, Jiaqi Han, Subham Sekhar Sahoo, and V olodymyr Kuleshov. Block diffusion: Interpolating between autoregressive and diffusion language models. InInternational Conference on Learning Representations, 2025

    2025

  34. [42]

    Scaling diffusion language models via adaptation from autoregressive models

    Shansan Gong, Shivam Agarwal, Yizhe Zhang, Jiacheng Ye, Lin Zheng, Mukai Li, Chenxin An, et al. Scaling diffusion language models via adaptation from autoregressive models. InInternational Conference on Learning Representations, 2025

    2025

  35. [43]

    Fast-dLLM v2: Efficient block-diffusion LLM

    Chengyue Wu, Hao Zhang, Shuchen Xue, Shizhe Diao, et al. Fast-dLLM v2: Efficient block-diffusion LLM. arXiv preprint arXiv:2509.26328, 2025

    2025

  36. [44]

    Sequential diffusion language models.arXiv preprint arXiv:2509.24007, 2025

    Yangzhou Liu, Yue Cao, Hao Li, Gen Luo, Zhe Chen, et al. Sequential diffusion language models.arXiv preprint arXiv:2509.24007, 2025

    2025

  37. [45]

    StepAudio 2.5 technical report.arXiv preprint arXiv:2605.23463, 2026

    StepFun-Audio Team. StepAudio 2.5 technical report.arXiv preprint arXiv:2605.23463, 2026

    2026 arXiv

  38. [46]

    LLaDA-TTS: Unifying speech synthesis and zero-shot editing via masked diffusion modeling.arXiv preprint arXiv:2603.26364, 2026

    Xiaoyu Fan, Huizhi Xie, Wei Zou, and Yunzhang Chen. LLaDA-TTS: Unifying speech synthesis and zero-shot editing via masked diffusion modeling.arXiv preprint arXiv:2603.26364, 2026

    2026

  39. [47]

    DiffuSpeech: Silent thought, spoken answer via unified speech- text diffusion.arXiv preprint arXiv:2601.22889, 2026

    Yuxuan Lou, Ziming Wu, Yaochen Wang, et al. DiffuSpeech: Silent thought, spoken answer via unified speech- text diffusion.arXiv preprint arXiv:2601.22889, 2026

    2026

  40. [48]

    OmniV oice: Towards omnilingual zero-shot text-to-speech with diffu- sion language models.arXiv preprint arXiv:2604.00688, 2026

    Han Zhu, Lingxuan Ye, Wei Kang, et al. OmniV oice: Towards omnilingual zero-shot text-to-speech with diffu- sion language models.arXiv preprint arXiv:2604.00688, 2026

    2026 arXiv

  41. [49]

    Chatterbox-flash: Prior-calibrated block diffusion for streaming zero-shot TTS.arXiv preprint arXiv:2605.30748, 2026

    Deokjin Seo, Gangin Park, and Kihyun Nam. Chatterbox-flash: Prior-calibrated block diffusion for streaming zero-shot TTS.arXiv preprint arXiv:2605.30748, 2026

    2026 arXiv

  42. [50]

    Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, Y . K. Li, Y . Wu, and Daya Guo. DeepSeekMath: Pushing the limits of mathematical reasoning in open language models, 2024. URLhttps://arxiv.org/abs/2402.03300. arXiv preprint ...

    2024 arXiv

  43. [51]

    V oxCPM: Tokenizer-free TTS for context-aware speech genera- tion and true-to-life voice cloning.arXiv preprint arXiv:2509.24650, 2025

    Yixuan Zhou, Guoyang Zeng, Xin Liu, Xiang Li, Renjie Yu, Ziyang Wang, Runchuan Ye, Weiyue Sun, Jiancheng Gui, Kehan Li, Zhiyong Wu, and Zhiyuan Liu. V oxCPM: Tokenizer-free TTS for context-aware speech genera- tion and true-to-life voice cloning.arXiv preprint arXiv:2509.24650...

    2025

  44. [52]

    Codec does matter: Exploring the semantic shortcoming of codec for audio language model.arXiv preprint arXiv:2408.17175, 2024

    Zhen Ye, Peiwen Sun, Jiahe Lei, Hongzhan Lin, Xu Tan, Zheqi Dai, Qiuqiang Kong, Jianyi Chen, Jiahao Pan, Qifeng Liu, Yike Guo, and Wei Xue. Codec does matter: Exploring the semantic shortcoming of codec for audio language model.arXiv preprint arXiv:2408.17175, 2024

    2024 arXiv

  45. [53]

    SpeechTokenizer: Unified speech tokenizer for speech large language models

    Xin Zhang, Dong Zhang, Shimin Li, Yaqian Zhou, and Xipeng Qiu. SpeechTokenizer: Unified speech tokenizer for speech large language models. InInternational Conference on Learning Representations, 2024

    2024

  46. [54]

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

    Sanyuan Chen, Chengyi Wang, Zhengyang Chen, Yu Wu, Shujie Liu, Zhuo Chen, Jinyu Li, Naoyuki Kanda, Takuya Yoshioka, Xiong Xiao, Jian Wu, Long Zhou, Shuo Ren, Yanmin Qian, Yao Qian, Jian Wu, Michael Zeng, Xiangzhan Yu, and Furu Wei. WavLM: Large-scale self-supervised pre-traini...

    2022

  47. [55]

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

    Jungil Kong, Jaehyeon Kim, and Jaekyoung Bae. HiFi-GAN: Generative adversarial networks for efficient and high fidelity speech synthesis. InAdvances in Neural Information Processing Systems, 2020

    2020

  48. [56]

    BigVGAN: A universal neural vocoder with large-scale training

    Sang-gil Lee, Wei Ping, Boris Ginsburg, Bryan Catanzaro, and Sungroh Yoon. BigVGAN: A universal neural vocoder with large-scale training. InInternational Conference on Learning Representations, 2023

    2023

  49. [57]

    Higgs audio v2: Text-audio foundation model.https://github.com/boson-ai/higgs-audio,

    Boson AI. Higgs audio v2: Text-audio foundation model.https://github.com/boson-ai/higgs-audio,

  50. [58]

    Huiwen Chang, Han Zhang, Lu Jiang, Ce Liu, and William T. Freeman. MaskGIT: Masked generative image transformer. InIEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022

    2022

  51. [59]

    d1: Scaling reasoning in diffusion large language models via reinforcement learning

    Siyan Zhao, Devaansh Gupta, Qinqing Zheng, and Aditya Grover. d1: Scaling reasoning in diffusion large language models via reinforcement learning. InAdvances in Neural Information Processing Systems, volume 38, pages 56729–56762, 2025. URLhttps://proceedings.neurips.cc/paper_f...

    2025

  52. [60]

    Training dif- fusion models with reinforcement learning

    Kevin Black, Michael Janner, Yilun Du, Ilya Kostrikov, and Sergey Levine. Training dif- fusion models with reinforcement learning. InInternational Conference on Learning Repre- sentations, 2024. URLhttps://proceedings.iclr.cc/paper_files/paper/2024/file/ 14f75513f0f1ca01de1e82...

    2024

  53. [61]

    Mask-aware policy gradients for diffusion language models.arXiv preprint arXiv:2607.15200, 2026

    Haran Raajesh, Kulin Shah, Adam Klivans, and Philipp Krähenbühl. Mask-aware policy gradients for diffusion language models.arXiv preprint arXiv:2607.15200, 2026. doi: 10.48550/arXiv.2607.15200. Accepted at COLM 2026

    2026 doi arXiv

  54. [62]

    Classifier-free diffusion guidance.arXiv preprint arXiv:2207.12598, 2022

    Jonathan Ho and Tim Salimans. Classifier-free diffusion guidance.arXiv preprint arXiv:2207.12598, 2022. doi: 10.48550/arXiv.2207.12598

    2022 doi arXiv

  55. [63]

    Proximal policy optimization algorithms.arXiv preprint arXiv:1707.06347, 2017

    John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. Proximal policy optimization algorithms.arXiv preprint arXiv:1707.06347, 2017. doi: 10.48550/arXiv.1707.06347

    2017 doi arXiv

  56. [64]

    vLLM-Omni: Fully disaggregated serving for any-to-any multimodal models.arXiv preprint arXiv:2602.02204, 2026

    Peiqi Yin, Jiangyun Zhu, Han Gao, Chenguang Zheng, Yongxiang Huang, Taichang Zhou, Ruirui Yang, Weizhi Liu, Weiqing Chen, Canlin Guo, Didan Deng, Zifeng Mo, Cong Wang, James Cheng, Roger Wang, and Hong- sheng Liu. vLLM-Omni: Fully disaggregated serving for any-to-any multimoda...

    2026 doi

  57. [65]

    Accelerating flow- matching-based text-to-speech via empirically pruned step sampling.arXiv preprint arXiv:2505.19931, 2025

    Qixi Zheng, Yushen Chen, Zhikang Niu, Ziyang Ma, Xiaofei Wang, Kai Yu, and Xie Chen. Accelerating flow- matching-based text-to-speech via empirically pruned step sampling.arXiv preprint arXiv:2505.19931, 2025. doi: 10.48550/arXiv.2505.19931

    2025 doi arXiv

  58. [66]

    Zipvoice: Fast and high-quality zero-shot text-to-speech with flow matching.arXiv preprint arXiv:2506.13053, 2025

    Han Zhu, Wei Kang, Zengwei Yao, Liyong Guo, Fangjun Kuang, Zhaoqing Li, Weiji Zhuang, Long Lin, and Daniel Povey. Zipvoice: Fast and high-quality zero-shot text-to-speech with flow matching.arXiv preprint arXiv:2506.13053, 2025. doi: 10.48550/arXiv.2506.13053

    2025 doi arXiv

  59. [67]

    V oxCPM2: Production deployment and inference performance.https://github.com/OpenBMB/ VoxCPM, 2026

    OpenBMB. V oxCPM2: Production deployment and inference performance.https://github.com/OpenBMB/ VoxCPM, 2026. Accessed: 2026-08-10

    2026

  60. [68]

    Spark-TTS: Nvidia triton inference serving.https://github.com/SparkAudio/Spark-TTS,

    SparkAudio. Spark-TTS: Nvidia triton inference serving.https://github.com/SparkAudio/Spark-TTS,

  61. [69]

    SGLang-Omni: High-performance multi-stage pipeline framework for omni models.https: //github.com/sgl-project/sglang-omni, 2026

    SGLang Team. SGLang-Omni: High-performance multi-stage pipeline framework for omni models.https: //github.com/sgl-project/sglang-omni, 2026. Accessed: 2026-08-10

    2026

  62. [70]

    Accessed: 2026-08-10

    2026

  63. [71]

    Models: Eleven flash v2.5.https://elevenlabs.io/docs/overview/models, 2026

    ElevenLabs. Models: Eleven flash v2.5.https://elevenlabs.io/docs/overview/models, 2026. Ac- cessed: 2026-08-10. 20 Luna-TTS Family Technical Report

    2026

  64. [72]

    Sonic 3.5 self-hosted hardware selection and latency.https://docs.cartesia.ai/self-hosted/ hardware-selection, 2026

    Cartesia. Sonic 3.5 self-hosted hardware selection and latency.https://docs.cartesia.ai/self-hosted/ hardware-selection, 2026. Accessed: 2026-08-10

    2026

  65. [73]

    Text-to-speech models: Play 3.0 mini.https://docs.play.ht/reference/models, 2026

    PlayHT. Text-to-speech models: Play 3.0 mini.https://docs.play.ht/reference/models, 2026. Ac- cessed: 2026-08-10

    2026

  66. [74]

    Text-to-speech: Octave 2.https://dev.hume.ai/docs/text-to-speech-tts/overview, 2026

    Hume AI. Text-to-speech: Octave 2.https://dev.hume.ai/docs/text-to-speech-tts/overview, 2026. Accessed: 2026-08-10

    2026

  67. [75]

    Vibevoice-realtime: Real-time streaming text-to-speech.https://github.com/microsoft/ VibeVoice, 2025

    Microsoft. Vibevoice-realtime: Real-time streaming text-to-speech.https://github.com/microsoft/ VibeVoice, 2025. Accessed: 2026-08-10

    2025

  68. [76]

    Aura-2 text-to-speech performance.https://developers.deepgram.com/changelog/2025/5/ 14, 2025

    Deepgram. Aura-2 text-to-speech performance.https://developers.deepgram.com/changelog/2025/5/ 14, 2025. Accessed: 2026-08-10

    2025

  69. [77]

    MiniMax Speech-2.8.https://www.minimax-speech.com/, 2026

    MiniMax. MiniMax Speech-2.8.https://www.minimax-speech.com/, 2026. Accessed: 2026-07-30

    2026

  70. [78]

    V oxCPM2 technical report.arXiv preprint arXiv:2606.06928, 2026

    Yixuan Zhou, Guoyang Zeng, Xin Liu, Xiang Li, Renjie Yu, Jiancheng Gui, Jiaheng Wu, Ziyang Wang, Xudong Shen, Runchuan Ye, Zhisheng Zhang, Jiuyang Zhou, Bingsong Bai, Weiyue Sun, Mengyuan Deng, Qundong Shi, Zhiyong Wu, and Zhiyuan Liu. V oxCPM2 technical report.arXiv preprint ...

    2026 arXiv

  71. [79]

    Introducing S2.1 Pro: Our most expressive TTS model yet.https://fish.audio/blog/ s2-1-pro-free-api/, 2026

    Fish Audio. Introducing S2.1 Pro: Our most expressive TTS model yet.https://fish.audio/blog/ s2-1-pro-free-api/, 2026. Accessed: 2026-07-30

    2026

  72. [80]

    Eleven v3.https://elevenlabs.io/v3, 2026

    ElevenLabs. Eleven v3.https://elevenlabs.io/v3, 2026. Accessed: 2026-07-30

    2026

  73. [81]

    NVV-SuperBench: Beyond words, beyond quality—benchmarking nonverbal vocal- izations in speech generation.arXiv preprint arXiv:2604.16211, 2026

    Liumeng Xue, Weizhen Bian, Jiahao Pan, Wenxuan Wu, Yilin Ren, Boyi Kang, Jingbin Hu, Ziyang Ma, Shuai Wang, Xinyuan Qian, et al. NVV-SuperBench: Beyond words, beyond quality—benchmarking nonverbal vocal- izations in speech generation.arXiv preprint arXiv:2604.16211, 2026. doi:...

    2026 doi arXiv

  74. [82]

    NV-Bench: Benchmark of nonverbal vocalization synthesis for expressive text-to-speech generation.arXiv preprint arXiv:2603.15352, 2026

    Qinke Ni, Huan Liao, Dekun Chen, Yuxiang Wang, and Zhizheng Wu. NV-Bench: Benchmark of nonverbal vocalization synthesis for expressive text-to-speech generation.arXiv preprint arXiv:2603.15352, 2026. doi: 10.48550/arXiv.2603.15352

    2026 doi

  75. [83]

    Emotional voice conversion: Theory, databases and ESD

    Kun Zhou, Berrak Sisman, Rui Liu, and Haizhou Li. Emotional voice conversion: Theory, databases and ESD. Speech Communication, 137:1–18, 2022. doi: 10.1016/j.specom.2021.11.006

    2022 doi

  76. [84]

    Gemini 3.1 Pro model card.https://deepmind.google/models/model-cards/ gemini-3-1-pro/, 2026

    Google DeepMind. Gemini 3.1 Pro model card.https://deepmind.google/models/model-cards/ gemini-3-1-pro/, 2026. Accessed: 2026-07-30

    2026

  77. [85]

    Ruskin Raj Manku, Yuzhi Tang, Xingjian Shi, Mu Li, and Alexander J. Smola. Emergenttts-eval: Evaluating tts models on complex prosodic, expressiveness, and linguistic challenges using model-as-a-judge. InAdvances in Neural Information Processing Systems, volume 38, 2025

    2025

  78. [86]

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

    Ziyang Ma, Zhisheng Zheng, Jiaxin Ye, Jinchao Li, Zhifu Gao, Shiliang Zhang, and Xie Chen. emotion2vec: Self-supervised pre-training for speech emotion representation. InFindings of the Association for Computational Linguistics: ACL 2024, pages 15747–15760, 2024. doi: 10.18653...

    2024 doi

  79. [88]

    Mint-bench: A com- prehensive multilingual benchmark for instruction-following text-to-speech.arXiv preprint arXiv:2604.17958, 2026

    Huakang Chen, Jingbin Hu, Liumeng Xue, Qirui Zhan, Wenhao Li, Guobin Ma, Hanke Xie, Dake Guo, Linhan Ma, Yuepeng Jiang, Bengu Wu, Pengyuan Xie, Chuan Xie, Qiang Zhang, and Lei Xie. Mint-bench: A com- prehensive multilingual benchmark for instruction-following text-to-speech.ar...

    2026 arXiv

  80. [2025]

    Open release with a unified semantic–acoustic audio tokenizer; no accompanying arXiv report

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.