FreyaTTS: A Compact Tokenizer-Free Flow-Matching Transformer for Turkish-First Speech Synthesis

Ahmet Erdem Pamuk; Ahmet Tunga Bayrak; Mustafa Yavuz; \"Omer Yent\"ur; Yavuz Alp Sencer \"Ozt\"urk

REVIEW 3 major objections 4 minor 13 references

FreyaTTS: A Compact Tokenizer-Free Flow-Matching Transformer for Turkish-First Speech Synthesis

T0 review · 3 major / 4 minor · reviewed 2026-08-02 · deepseek-v4-flash

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

Pith's one-line read A compact 183M-parameter, tokenizer-free, non-autoregressive model can beat larger open Turkish TTS systems on accuracy, the paper argues.

desk verdict A solid, release-friendly Turkish TTS system whose headline WER beats are plausible but rest on a single unchosen reference clip; the abstract over-claims on naturalness. read the letter →

arxiv 2607.09530 v2 pith:SR7NBRRY submitted 2026-07-10 cs.CL

Ahmet Erdem Pamuk , Ömer Yentür , Ahmet Tunga Bayrak , Yavuz Alp Sencer Öztürk , Mustafa Yavuz This is my paper · ORCID

classification cs.CL

keywords tokenizer-freeTTSTurkishspeechsynthesisnon-autoregressiveflowmatchingdiffusiontransformerdurationpredictionvoicelockedgedeployment

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

FreyaTTS is a Turkish-first text-to-speech system built to show that a small model can be both accurate and deployable. The central empirical claim is that a 183.2M-parameter transformer, trained from scratch and measuring errors by how often an automatic transcriber mishears words and characters, reaches 8.0% word error and 3.0% character error on a 495-sentence conversational benchmark—lower than two larger open systems in the comparison, at roughly 40–55% of their parameter counts—while ranking second on naturalness. The design removes the traditional frontends: no phonemizer, no grapheme-to-phoneme rules, and no discrete speech tokenizer; text enters as a 92-symbol Turkish character vocabulary, numbers are expanded to their spoken form, and a duration head predicts the whole utterance length so all audio frames can be denoised in parallel. A two-stage post-training recipe, a single-speaker voice lock followed by short-utterance coverage, turns a speaker-agnostic pretrained prior into a stable single-voice production model. If the claim holds, mid-resource languages need not wait for billion-parameter multilingual systems to get accurate, on-device synthesis.

What carries the argument

The load-bearing machinery is a conditional flow-matching Diffusion Transformer—a transformer that predicts, for every noisy latent frame, the velocity carrying it toward clean data—operating over a frozen continuous latent codec. Flow matching here means training the network to predict, for each noisy latent frame, the straight-line velocity that would carry it to the clean latent; a 32-step Euler solver integrates this velocity field at inference. Text is embedded as 92-symbol character features, refined by a stack of convolutional blocks, and used twice: a small duration head predicts the total number of 40 ms latent frames, and cross-attention layers let each audio frame query the charac

What would settle it

Run the released model on a held-out set of long (twenty-plus word) Turkish utterances that deliberately mix digit strings, dates, and isolated acronyms, and transcribe the outputs with a second independent recognizer, comparing band-matched word error against the reported 8.0%. If error climbs well above that figure or above the compact phonemizer-driven baselines on these inputs, the single-global-duration alignment premise is the point of failure; if it stays flat, the claim generalizes beyond short-to-medium conversational sentences.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is a system-level result: a from-scratch, single-voice model with 183.2M trainable parameters, working in the frozen continuous latent space of a codec that encodes at 16 kHz and reconstructs at 48 kHz, and conditioned on raw Turkish characters through cross-attention, synthesizes conversational speech with 8.0% word error and 3.0% character error on the new Freya-TR-Eval benchmark, with all outputs downsampled to the same narrow band before scoring. That is lower error than the two larger open voice-cloning systems in the comparison, and its naturalness score is second only to the largest competitor. The paper reports a real-time factor of 0.11 on a c

Load-bearing premise

The load-bearing premise is that one globally predicted utterance length, plus attention between audio frames and character positions, is enough to align text to audio without any explicit monotonic or token-wise alignment step; if Turkish's agglutinative forms or digit-dense inputs produce lengths this predictor cannot anticipate, words get skipped and error rises.

Editorial extensions

If this is right

At 183M parameters, one second of audio synthesized in about 0.11 seconds on a consumer GPU and in about real time on a laptop CPU, the system opens an operating point—accurate conversational TTS on edge hardware—that multi-billion-parameter backbones cannot reach.
Because the pipeline learns pronunciation end-to-end from a 92-symbol character vocabulary, Turkish agglutinative morphology, vowel harmony, acronyms, and in-context numbers are absorbed into the weights rather than dictated by hand-built frontend rules, removing a class of rule-based failure modes.
Non-autoregressive parallel denoising over a predicted duration sidesteps the left-to-right error accumulation of autoregressive decoders, which the paper argues is the reason word and character error stay low on conversational text.
The two-stage post-training recipe—single-speaker voice lock, then short-utterance coverage—converts a speaker-agnostic prior into a stable single-voice product, with cross-generation pitch variation collapsing from 74.9 Hz to 5.0 Hz as measured evidence.
The released benchmark, model weights, and evaluation harness make the headline word/character error numbers reproducible without redistributing third-party audio, so the comparison against larger systems can be checked independently.

Reading between the lines

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

Editorial extension: the paper's own Section 5 limitation—a single global length prediction cannot prevent word-skip and long-horizon drift—identifies duration prediction, not acoustics, as the mechanism most likely to bound the approach; replacing the global duration head with a monotonic or token-wise aligner is the natural next step, and the paper says such an aligner is already prototyped.
Editorial extension: the same recipe—frozen codec, compact non-autoregressive flow-matching transformer, and voice-lock fine-tuning—should transfer to other mid-resource agglutinative languages or digit-heavy domains, since it removes the phonemizer and grapheme-to-phoneme frontend; duration prediction quality would be the main transfer risk.
Editorial extension: because word error is measured through a single automatic recognizer, a second recognizer or human transcription on the released benchmark would resolve whether the 8.0% advantage over larger systems is robust or partly an artifact of recognizer choice; the paper's bootstrap and band-matching checks reduce but do not eliminate this concern.

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.

Desk Editor's Note

read the letter

Worth a look if you care about mid-resource TTS or edge deployment. The system is not a conceptual breakthrough — it’s a clean synthesis of known pieces: frozen AudioVAE2 latents, a character-level DiT with flow matching and cross-attention, a duration head, and a two-stage post-training recipe. But the combination is new for Turkish, the release is unusually complete (weights, code, seeded benchmark, eval harness), and the evaluation is more honest than most.

The credibility comes from the calibration anchors. Freya-TR-Eval transcribes human recordings at 9.7% WER under the same band-matched pipeline, and the frozen codec alone costs 2.6% WER. Against those, FreyaTTS at 8.0% is plausible. Bootstrap CIs are reported, and they’re not hiding a pathological variance. They also plainly state that Piper and MMS-TTS get lower WER, and that the MOS column is a ranking of point estimates with overlapping CIs — which is the right way to read it. The voice-lock measurement (F0 std collapsing from 74.9 to 5.0 Hz) is concrete and well quantified.

Soft spots, in ascending order. First, the abstract says 'rule-free' while the system uses a digit-expansion frontend and a small inference wrapper (duration floor, voicing retry). Minor, but the label is doing work it shouldn’t. Second, the same abstract claims 'highest naturalness (MOS)' when Table 2 shows XTTS-v2 with a higher point estimate and overlapping intervals. That’s misleading at the abstract level, even though the paper body corrects it. Third and most substantive: the comparison against the zero-shot cloners XTTS-v2 and F5-TTS uses one fixed shared reference clip, and the paper concedes cloner scores can be sensitive to that choice — but never tests it. If that clip is unusually bad for those models, the 11.1% and 24.3% WERs overstate the gap, and the headline claim is weaker than reported. This isn’t fatal to the paper — 8.0% WER is still a good number on a 495-sentence benchmark, and the efficiency results stand on their own — but a few reference clips and a sensible selection criterion would have made the central claim much more robust. It's an addressable experiment, not a fundamental flaw.

The single-global-duration alignment concern is real for long inputs but the paper already flags word-skip and drift as failure modes, and the measured WER shows the mechanism works on this distribution. Not a major issue.

This paper deserves a serious referee. The system is reproducible, the limitations are stated, and the failures are the kind that revision and a sensitivity analysis can fix. I’d send it to review and ask for the abstract to match the body and for reference-clip sensitivity on the zero-shot baselines. The engineering is solid.

Referee Report

3 major / 4 minor

Summary. FreyaTTS is a 183.2M-parameter non-autoregressive conditional flow-matching DiT that synthesizes 48 kHz Turkish speech from a 92-symbol character vocabulary in the frozen 25 Hz latent space of AudioVAE2, without a phonemizer or discrete speech tokenizer. A duration head predicts total latent length and cross-attention over ConvNeXt-refined character features conditions a 32-step Euler ODE. A two-stage post-training recipe (single-speaker voice lock, then short-utterance coverage) converts a multi-speaker pretrained prior into a single-voice production model. The paper's central empirical claim (§4.2, Table 2) is that this 183M model achieves lower band-matched Whisper WER/CER (8.0%/3.0%) than the larger open XTTS-v2 (11.1%/3.9%) and F5-TTS (24.3%/10.9%) on the released Freya-TR-Eval benchmark, ranks second in MOS, and runs with RTF 0.11 on an RTX 4090 (≈0.14 on H100). The paper releases weights, training/inference code, the benchmark with seeded build scripts, and the evaluation harness.

Significance. The contribution is potentially significant for mid-resource and edge TTS. The paper ships unusually complete artifacts: a released benchmark with seeded construction, a reusable evaluation protocol, bootstrap CIs, calibration anchors (human recordings at 9.7% WER; AudioVAE-only floor at 2.6% WER), an ablation of the inference wrapper, and an ECAPA speaker-consistency check. The NAR design and efficiency measurements are plausible and the limitations in §5 are candid. If the headline comparison is made robust, this would be a valuable reference point for compact Turkish-first synthesis. The main threat is not circularity but the external validity of the comparison protocol: a self-authored benchmark and a single reference clip for zero-shot baselines.

major comments (3)

[§4.1, Table 2] The comparison against the two larger open systems rests on one fixed shared reference clip for XTTS-v2 and F5-TTS, and the paper itself notes that 'cloner scores can be sensitive to this choice.' No reference sweep, selection criterion, or acoustic match is reported. Because the claim that FreyaTTS has lower WER/CER than both larger systems is the paper's central empirical result, the authors should provide a reference-clip robustness analysis (e.g., 5–10 reference clips spanning channel, noise, and style), report the per-reference WER/CER distributions, and state how the reported clip was selected. Without this, the 11.1% and 24.3% baseline WERs may be unrepresentatively high.
[§4.2] The bootstrap CI is computed only for FreyaTTS. The central comparison to XTTS-v2 (3.1 WER points) and F5-TTS (16.3 points) is reported as a point-estimate difference with no uncertainty or paired test. Please report paired sentence-level bootstrap CIs for ΔWER and ΔCER (FreyaTTS minus each baseline), and ideally a per-sentence error table. This is needed to distinguish a robust ordering from run-to-run or reference-choice noise.
[§3.3, §3.6, §5] The architecture relies on a single global duration prediction plus cross-attention as the only text-to-audio pathway, and §5 explicitly concedes residual word-skip and long-horizon drift that 'a single global length prediction cannot prevent,' plus the need for spoken-form digit expansion. The current benchmark contains 3–13-word sentences and no reported digit-density or long-form breakdown, so the 'highly reliable conversational synthesis' claim is demonstrated only on a narrow distribution. A stress-test subset (long utterances, digit-dense text, isolated tokens) with per-condition WER/CER would bound these failure modes and is directly relevant to the paper's stated use case.

minor comments (4)

[§4.2, Abstract] 'Highest naturalness (MOS) among the compact systems' is a point estimate; the 95% CIs of the top four systems overlap (Table 2). The abstract should say 'highest MOS point estimate' or 'highest mean MOS'.
[§3.4] The pretraining corpus is described only as 'large-scale, high-quality internal' with no hours, speaker count, or recording/session details. Please include these metadata for reproducibility.
[§3.1, §3.5] Calling the system 'end-to-end' while digit strings are expanded by a text frontend is slightly overstated; clarify that digit expansion is a text-normalization step and that the acoustic path remains tokenizer-free.
[§4.4, Table 3] It would help to state explicitly whether the RTF numbers include AudioVAE2 decoding for all systems (the text says so, but the table caption should repeat it) and to define the word-range buckets used for 'short', 'medium', and 'long'.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the headline WER/CER is an externally anchored, protocol-matched measurement, and no fitted parameter or self-citation is repackaged as a prediction.

full rationale

The paper's central empirical claim — that the 183.2M FreyaTTS attains 8.0% WER / 3.0% CER on Freya-TR-Eval, lower than XTTS-v2 and F5-TTS — is supported by a measured evaluation, not by a fitted quantity presented as a prediction. The benchmark texts are drawn from Common Voice 17-tr, CoVoST2-tr, and generated conversational lines; the paper states that 'FreyaTTS is trained only on internal corpora (sections 3.4 and 3.5) disjoint from the benchmark sources, so every item probes generalization.' The WER/CER protocol is applied identically to all systems, uses an unmodified Whisper-large-v3 transcriber, and is anchored by two external calibration points: real FLEURS-tr recordings transcribe at 9.7% WER under the same pipeline, and re-encoding real target-speaker audio through the frozen AudioVAE costs 2.6% WER. The duration head, flow-matching objective, and voice-lock fine-tuning are model-training components, not predictions that reduce by construction to the evaluation target. The 'content similarity' metric (1 − normalized character edit distance between input text and transcript) is a direct self-consistency measurement used only for the voice-lock analysis, not the headline comparison. There are no author self-citations invoked as load-bearing external facts: AudioVAE2/VoxCPM2 is an external, Apache-2.0 frozen codec, and the other cited components (flow matching, adaLN-zero, RoPE, Whisper, ECAPA-TDNN) are standard external methods. The acknowledged sensitivity of zero-shot cloner scores to the single shared reference clip in §4.1 is a measurement-validity concern, not circularity: the reference clip is not fitted to the paper's claim, and the paper explicitly discloses it. The §5 limitations — residual word-skip, long-horizon drift, narrowband fidelity ceiling — are honest boundary statements and do not make the derivation circular. No step in the derivation is equivalent by construction to its input, so the appropriate finding is no significant circularity.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The ledger has no invented physical entities; the model's parameters are learned, not free. The listed free parameters are minor hyperparameters and inference heuristics. The main load-bearing assumptions are the sufficiency of the frozen codec's latent space and the unreleased internal corpus.

free parameters (3)

lambda_dur = 0.1 pretraining, 0.2 SFTv2
Hand-chosen weight for the duration-prediction loss term in eq. (1); it is a hyperparameter, not derived from data.
Euler steps = 32
Fixed number of ODE solver steps at inference; no guidance, no sway sampling; chosen empirically.
inference wrapper heuristics (duration floor, voicing retry thresholds) = not specified
Inference wrapper heuristics in §3.6 are ad hoc and not quantified, which weakens replication slightly.

assumptions (5)

standard math Conditional flow matching with a linear path and independent Gaussian prior is a valid generative training objective.
Used in eq. (1), following Lipman et al. (2023) and Tong et al. (2024); standard.
domain assumption The frozen AudioVAE2 latent space preserves enough phonetic/acoustic detail for text-to-speech.
The paper measures a 2.6% WER floor when re-encoding real audio (§4.2), but it is still an external codec whose latent distribution is not under the authors' control.
domain assumption The internal Turkish pretraining corpus is large, high-quality, and disjoint from the benchmark sources.
Stated in §3.4/§4.1 but the corpus is not released, so this premise cannot be independently verified.
domain assumption A single predicted utterance duration plus cross-attention over character features suffices to align text and audio frames without explicit monotonic alignment.
Core to the NAR design; §5 acknowledges residual word-skip and long-horizon drift, i.e., this assumption holds imperfectly.
domain assumption Character-level input with digit frontend expansion is sufficient for Turkish text normalization.
The paper relies on a 92-symbol character vocabulary plus digit-string expansion to cover Turkish orthography; stated in §1 and §3.3.

how reviews work

0 comments

Cite this review

Pith. "Pith review of FreyaTTS: A Compact Tokenizer-Free Flow-Matching Transformer for Turkish-First Speech Synthesis." pith.science (2026). https://pith.science/paper/SR7NBRRY

@misc{pith2026260709530,
  author       = {Pith},
  title        = {Pith review of: FreyaTTS: A Compact Tokenizer-Free Flow-Matching Transformer for Turkish-First Speech Synthesis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SR7NBRRY}},
  note         = {Machine review of arXiv:2607.09530}
}

read the original abstract

We introduce Freya-TTS, a compact, tokenizer-free, Turkish-first text-to-speech model designed for highly reliable and efficient conversational synthesis. Freya-TTS is a 183.2M-parameter non-autoregressive conditional flow-matching Diffusion Transformer (DiT) that operates in the continuous latent space of the frozen AudioVAE2 (16 kHz encode, 48 kHz decode), allowing the model to focus its capacity on text-to-latent mapping while inheriting high-quality 48 kHz reconstruction. We advance the framework along three key dimensions: (1) rule-free end-to-end modeling from a 92-symbol Turkish character vocabulary without a phonemizer, grapheme-to-phoneme frontend, or discrete speech tokenizer, with digit strings expanded to their spoken form at the text frontend; (2) non-autoregressive parallel denoising, which predicts the entire latent sequence simultaneously over a predicted duration; and (3) a production-oriented two-stage post-training recipe consisting of single-speaker voice locking and short-utterance coverage, improving speaker consistency and robustness on short inputs. On the Freya-TR-Eval benchmark, Freya-TTS achieves a band-matched word error rate (WER) of 8.0% and character error rate (CER) of 3.0%, lower error than both larger open systems in its field, XTTS-v2 and F5-TTS, at 40-55% of their parameter count, together with the highest naturalness (MOS) among the compact systems. The model achieves a real-time factor of 0.11 on a consumer GPU (RTX 4090; ~0.14 mean on an H100) and synthesizes in real time on a laptop CPU, making it well suited for resource-constrained edge deployment. We release the model weights, training and inference code, and evaluation benchmark under the Apache-2.0 license.

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

13 extracted references · 12 linked inside Pith

[1]

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

Philip Anastassiou, Jiawei Chen, Jitong Chen, Yuanzhe Chen, Zhuo Chen, Ziyi Chen, Jian Cong, Lelai Deng, Chuang Ding, Lu Gao, et al. Seed-tts: A family of high-quality versatile speech generation models.arXiv preprint arXiv:2406.02430,

arXiv
[3]

High fidelity neural audio compression.arXiv preprint arXiv:2210.13438,

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

arXiv
[6]

Cfg-zero*: Improved classifier-free guidance for flow matching models.arXiv preprint arXiv:2503.18886,

Weichen Fan, Amber Yijia Zheng, Dejia Zhu, Yao Ma, Nikola Liu, Zhangyang Wang, and Dilin Liu. Cfg-zero*: Improved classifier-free guidance for flow matching models.arXiv preprint arXiv:2503.18886,

arXiv
[7]

Qwen3-tts technical report.arXiv preprint arXiv:2601.15621,

Hangrui Hu, Xinfa Zhu, Ting He, Dake Guo, Bin Zhang, Xiong Wang, Zhifang Guo, Ziyue Jiang, Hongkun Hao, Zishan Guo, et al. Qwen3-tts technical report.arXiv preprint arXiv:2601.15621,

arXiv
[8]

Minicpm4: Ultra-efficient llms on end devices.arXiv preprint arXiv:2506.07900,

MiniCPM Team, Chaojun Xiao, Yuxuan Li, Xu Han, Yuzhuo Bai, Jie Cai, Haotian Chen, Wentong Chen, Xin Cong, Ganqu Cui, et al. Minicpm4: Ultra-efficient llms on end devices.arXiv preprint arXiv:2506.07900,

arXiv
[11]

Voxcpm2 technical report.arXiv preprint arXiv:2606.06928,

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. Voxcpm2 technical report.arXiv preprint arXiv:2606.06928,

arXiv
[13]

Omnivoice: Towards omnilingual zero-shot text-to-speech with diffusion language models.arXiv preprint arXiv:2604.00688, 2026

Han Zhu, Lingxuan Ye, Wei Kang, Zengwei Yao, Liyong Guo, Fangjun Kuang, Zhifeng Han, Weiji Zhuang, Long Lin, and Daniel Povey. Omnivoice: Towards omnilingual zero-shot text-to-speech with diffusion language models.arXiv preprint arXiv:2604.00688, 2026

arXiv 2026
[2020]

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

Zhihao Du, Qian Chen, Shiliang Zhang, Kai Hu, Heng Lu, Yexin Yang, Hangrui Hu, Siqi Zheng, Yue Gu, Ziyang Ma, et al. Cosyvoice: A scalable multilingual zero-shot text-to-speech synthesizer based on supervised semantic tokens.arXiv preprint arXiv:2407.05407,

arXiv

Show all 13 references

[2021]

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

Xinsheng Wang, Mingqi Jiang, Ziyang Ma, Ziyu Zhang, Songxiang Liu, Linqin Li, Zheng Liang, Qixi Zheng, Rui Wang, Xiaoqin Feng, et al. Spark-tts: An efficient llm-based text-to-speech model with single-stream decoupled speech tokens.arXiv preprint arXiv:2503.01710,

arXiv
[2023]

Xtts: a massively multilingual zero-shot text-to-speech model.arXiv preprint arXiv:2406.04904,

Edresson Casanova, Kelly Davis, Eren Gölge, Görkem Göknar, Iulian Gulea, Logan Hart, Aya Aljafari, Joshua Meyer, Reuben Morais, Samuel Olayemi, et al. Xtts: a massively multilingual zero-shot text-to-speech model.arXiv preprint arXiv:2406.04904,

arXiv
[2024]

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

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

arXiv
[2025]

Voxcpm: Tokenizer-free tts for context-aware speech generation and true-to-life voice cloning.arXiv preprint arXiv:2509.24650,

Yixuan Zhou, Guoyang Zeng, Xin Liu, Xiang Li, Renjie Yu, Ziyang Wang, Runchuan Ye, Weiyue Sun, Jiancheng Gui, Kehan Li, et al. Voxcpm: Tokenizer-free tts for context-aware speech generation and true-to-life voice cloning.arXiv preprint arXiv:2509.24650,
[2026]

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

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,

arXiv

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.

[1] [1]

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

Philip Anastassiou, Jiawei Chen, Jitong Chen, Yuanzhe Chen, Zhuo Chen, Ziyi Chen, Jian Cong, Lelai Deng, Chuang Ding, Lu Gao, et al. Seed-tts: A family of high-quality versatile speech generation models.arXiv preprint arXiv:2406.02430,

arXiv

[2] [3]

High fidelity neural audio compression.arXiv preprint arXiv:2210.13438,

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

arXiv

[3] [6]

Cfg-zero*: Improved classifier-free guidance for flow matching models.arXiv preprint arXiv:2503.18886,

Weichen Fan, Amber Yijia Zheng, Dejia Zhu, Yao Ma, Nikola Liu, Zhangyang Wang, and Dilin Liu. Cfg-zero*: Improved classifier-free guidance for flow matching models.arXiv preprint arXiv:2503.18886,

arXiv

[4] [7]

Qwen3-tts technical report.arXiv preprint arXiv:2601.15621,

Hangrui Hu, Xinfa Zhu, Ting He, Dake Guo, Bin Zhang, Xiong Wang, Zhifang Guo, Ziyue Jiang, Hongkun Hao, Zishan Guo, et al. Qwen3-tts technical report.arXiv preprint arXiv:2601.15621,

arXiv

[5] [8]

Minicpm4: Ultra-efficient llms on end devices.arXiv preprint arXiv:2506.07900,

MiniCPM Team, Chaojun Xiao, Yuxuan Li, Xu Han, Yuzhuo Bai, Jie Cai, Haotian Chen, Wentong Chen, Xin Cong, Ganqu Cui, et al. Minicpm4: Ultra-efficient llms on end devices.arXiv preprint arXiv:2506.07900,

arXiv

[6] [11]

Voxcpm2 technical report.arXiv preprint arXiv:2606.06928,

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. Voxcpm2 technical report.arXiv preprint arXiv:2606.06928,

arXiv

[7] [13]

Omnivoice: Towards omnilingual zero-shot text-to-speech with diffusion language models.arXiv preprint arXiv:2604.00688, 2026

Han Zhu, Lingxuan Ye, Wei Kang, Zengwei Yao, Liyong Guo, Fangjun Kuang, Zhifeng Han, Weiji Zhuang, Long Lin, and Daniel Povey. Omnivoice: Towards omnilingual zero-shot text-to-speech with diffusion language models.arXiv preprint arXiv:2604.00688, 2026

arXiv 2026

[8] [2020]

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

Zhihao Du, Qian Chen, Shiliang Zhang, Kai Hu, Heng Lu, Yexin Yang, Hangrui Hu, Siqi Zheng, Yue Gu, Ziyang Ma, et al. Cosyvoice: A scalable multilingual zero-shot text-to-speech synthesizer based on supervised semantic tokens.arXiv preprint arXiv:2407.05407,

arXiv