Pith. sign in

REVIEW 4 major objections 6 minor 24 references

DialectS2S: End-to-End Speech Dialogue Modeling for Low-Resource Chinese Dialects

T0 review · 4 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read DialectS2S claims that re-synthesizing speech supervision from the model's own predicted text removes the semantic mismatch that degrades dialect speech, letting an end-to-end dialogue model outperform open baselines across three Chinese…

desk verdict Useful new framework for end-to-end dialect speech dialogue with a genuinely novel self-aligned supervision loop, but the headline intelligibility claim is not established by the reported CER comparisons. read the letter →

arxiv 2608.08067 v1 pith:IJYLSXZL submitted 2026-08-08 cs.CL cs.AI

Yi Shu , Tianyu Peng , Yingzhuo Deng , Wen Yang , Jun Lin , Changming Xie , Xinyu Yu , Jiajun Zhang This is my paper · ORCID
classification cs.CLcs.AI
keywords end-to-endspeechdialogueChinesedialectslow-resourceself-alignedsupervisionspeech-to-speechmodelingdialectdatasynthesisintelligibility
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 tries to establish that end-to-end speech dialogue can be extended to low-resource Chinese dialects without the usual drop in speech quality, and that the key obstacle is a mismatch between the model's evolving semantic representations and the fixed speech targets used in training. As a model adapts to a dialect, its internal semantic space shifts; if the speech supervision stays frozen, the speech-generation module has to recover semantics it should not need to recover, and output becomes unstable and unnatural. The authors' fix is to use the adapted model's own text predictions to synthesize new speech supervision, so the targets and the hidden representations move together. They report that the resulting system, built from a data-synthesis pipeline and two-stage post-training, outperforms open baselines in dialect consistency, response quality, and intelligibility for Sichuanese, Cantonese, and the Tianjin dialect, while keeping Mandarin and English performance. A sympathetic reader would care because it offers a practical, open recipe for a capability that currently barely exists in end-to-end systems.

What carries the argument

The load-bearing mechanism is self-aligned speech supervision, where the Thinker (the module that reads speech and produces hidden semantic states $H$ and text tokens) generates text predictions $\hat{T} = \mathrm{Thinker}(X)$ after supervised fine-tuning, and a dialect TTS synthesizes $\hat{U} = \mathrm{TTS}(\hat{T})$; the Talker (the module that maps hidden states to speech tokens) is then trained with the loss $\mathcal{L}_{\mathrm{AlignTalker}} = -\sum_j \log P(\hat{u}_j \mid H, \hat{U}_{<j}; \theta_{\mathrm{Talker}})$. This makes the speech target track the evolving semantic space instead of remaining frozen, so the Talker models prosody and pronunciation rather than performing implicit semantic recovery. A second supporting mechanism is the data pipeline that rewrites existing Mandarin dialogue text into dialectal expressions with an LLM, synthesizes query and response speech from seed utterances, and filters samples with a naturalness scorer.

What would settle it

Score DialectS2S and each open baseline with the same automatic speech recognizer on their generated speech, using each model's own text as reference and also human transcripts, and compare all systems under identical conditions; if DialectS2S does not beat the baselines on matched CER or on human listening ratings, the claimed intelligibility advantage is refuted.

Watch

Extended reading notes

Core claim

The central discovery is that supervised fine-tuning changes the hidden representations of a Thinker-Talker speech dialogue model, but conventional speech supervision stays anchored to the old semantic space; the speech module then does double duty, recovering meaning while generating acoustics, and dialect speech degrades. The paper's solution is self-aligned speech supervision: after a first stage of mixed-data supervised fine-tuning, the fine-tuned Thinker generates text predictions for each input, a dialect TTS turns those predictions into speech, and that generated speech replaces the original target in the Talker's training loss. Aligning the target's semantic content with the model's current predictions reduces the mismatch, letting the Talker focus on pronunciation, prosody, and naturalness. The authors show that this second stage improves character error rate relative both to stopping after two epochs and to adding a third epoch of ordinary fine-tuning, with matched optimization steps, and that the full model beats several open speech dialogue baselines on language matching, judged response quality, and intelligibility across three Chinese dialects.

Load-bearing premise

The load-bearing premise is that the speech-recognition error rate used to measure intelligibility reflects what human listeners actually hear, because the model is compared with an external benchmark average instead of with the same baselines on the same generated speech.

Editorial extensions

If this is right

  • A pretrained end-to-end speech dialogue model can be adapted to a new low-resource dialect with only a few thousand synthesized speech pairs, without sacrificing its original Mandarin and English ability.
  • Adding more ordinary supervised fine-tuning does not fix speech degradation and can make it worse; the self-aligned second stage gives the intelligibility gain at matched optimization steps.
  • Because the whole framework, including checkpoints, datasets, and fine-tuning code, is released, other teams can reproduce the results and port the pipeline to additional dialects.
  • Dialect speakers can hold spoken conversations in their own variety: the model matches the input dialect in over 90% of test cases for Sichuanese, Cantonese, and Tianjin, whereas leading open baselines mostly fall back to Mandarin.

Reading between the lines

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

  • Inference: The same self-alignment idea should transfer to any speech-to-speech distribution shift, such as accented English, child speech, code-switching, or a new speaker's voice, because the underlying problem is frozen supervision lagging behind drifting hidden representations, not dialect-specific.
  • Inference: The reported intelligibility numbers should be read cautiously until a matched comparison exists; the paper's CER table compares the model's own generated speech against a benchmark average, so it does not yet show that the model is more intelligible than the baselines on the same utterances.
  • Inference: A direct mechanistic test is available: measure the distance between the final-layer Thinker hidden states and the speech token targets before and after self-aligned training; the mechanism predicts this distance shrinks even when ordinary extra fine-tuning does not.
  • Inference: The pipeline's ceiling for dialect authenticity is set by the LLM's dialect rewriting and the TTS's dialect synthesis; if those components produce a shallow or standardized version of a dialect, self-aligned supervision will faithfully preserve that version rather than recover the true dialect.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes DialectS2S, an end-to-end speech dialogue model for low-resource Chinese dialects built on OpenS2S. It introduces a dialect speech dialogue synthesis pipeline (LLM-based text rewriting, CosyVoice2 TTS, UTMOS filtering) and a two-stage post-training procedure: mixed-data supervised fine-tuning followed by self-aligned speech supervision, in which the fine-tuned Thinker generates text that is synthesized via a dialect TTS and used as the Talker training target. The model is evaluated on a 250-sample multilingual/dialect benchmark for language matching accuracy, response quality (via an LLM judge), and speech intelligibility (via CER), compared with six open-source baselines, with additional ablations varying SFT epochs and the self-aligned stage.

Significance. If the results were validated, the contributions would be practically valuable: the authors open-source checkpoints, datasets, and code; the synthesis pipeline offers a scalable way to create dialect speech dialogue data; and the self-aligned supervision idea is an interesting approach to adapting speech supervision to evolving hidden representations. The paper also provides a useful comparison of several open-source speech dialogue models on dialect inputs. However, the central intelligibility claim is currently undermined by the evaluation protocol, which does not compare against baselines under the same conditions and uses the model's own text as the CER reference, making the claimed 'substantial improvements in speech intelligibility' unsupported. The underlying idea remains plausible, and the gaps could be addressed with additional experiments.

major comments (4)
  1. [§5.4, Table 3] The claim that DialectS2S 'significantly outperforms existing open-source baselines' in speech intelligibility is not supported by the reported data. Table 3 compares DialectS2S's CER (8.69% dialect average) to the FRASR2-AED benchmark average (11.67%) on natural speech, not to the CER of any baseline model on the same generated speech. None of the baseline models are run through the same ASR pipeline on the same test queries, and the reference for DialectS2S is its own synchronously generated text rather than a human transcript. This is not a like-for-like comparison and cannot establish a relative intelligibility advantage over baselines.
  2. [§5.4, Table 5, and §3.3 Eqs. (2)-(4)] The CER evaluation uses the model's own generated text as the reference for its own speech. Because the self-aligned supervision training synthesizes speech from the model's own text predictions (Eq. 3) and trains the Talker to produce speech matching that text (Eq. 4), the CER reductions in Table 5 may simply reflect improved internal text-speech consistency, which is exactly what the training optimizes, rather than improved absolute intelligibility. Human listening tests, or an external ASR evaluated against fixed human reference transcripts, are needed to establish intelligibility gains; the current metric conflates self-consistency with intelligibility.
  3. [§3.3, Eqs. (2)-(4)] The self-aligned supervision is circular in that the training target is derived from the model's own predictions. If the Thinker produces erroneous text, the TTS synthesizes speech for that erroneous text and the Talker is trained to match it, potentially reinforcing the model's own errors. The claimed mechanism—reducing semantic mismatch between hidden states and speech targets—is plausible, but the reported CER results do not separate the effect of aligning with the model's own semantics from the effect of making the model more self-consistent. The authors should report CER against human reference transcripts, or compare with a condition where supervision is generated from ground-truth text, to support the mechanism claim.
  4. [§5.1-§5.3] The language-matching and response-quality evaluations rely on a single 250-sample benchmark, automatic language identification (FireRedLID), and an LLM judge (Qwen3-Plus) without reported statistical significance, confidence intervals, or human validation. The near-zero baseline scores for dialect conditions (e.g., 0-2% for Tianjin) are expected because those models are not designed for dialects, but the claim of 'significant' improvements over baselines is not supported without significance tests or a more thorough analysis of the judge's reliability.
minor comments (6)
  1. [§3.2, Appendix A] The text states that 'Further seed-set details are provided in Appendix A,' but Appendix A does not include details on the 100 dialect seed utterances or the UTMOS filtering threshold; these would be needed to reproduce the data pipeline.
  2. [Appendix A, Table 6] The cascade comparison is based on only 50 Sichuanese samples, which is a very small sample size; the latency and quality numbers should be interpreted with caution, and the paper should state whether these differences are stable across repeated runs.
  3. [Figure 2] The PCA visualization of 'final-layer Thinker hidden states' does not specify which layer is used, how token-level representations are aggregated, or which samples are projected; the reader cannot verify the claim of hidden-state drift without these details.
  4. [§6] The paper states that sft-3ep and DialectS2S are trained with comparable numbers of optimization steps, but no step counts are given; reporting them would strengthen the fairness of the ablation comparison.
  5. [§5.4] The sentence 'Previous experiments show that existing baseline models mainly generate Mandarin Chinese rather than dialect speech under dialect interaction settings' is vague; it should reference the specific tables (e.g., Table 1) or provide the evidence explicitly.
  6. [General] The paper claims to 'fully open-source' the framework but does not provide a repository link or URL; the authors should include an accessible link in the final version.

Circularity Check

1 steps flagged · score 6.0 of 10

Speech-intelligibility prediction is self-referential: CER is computed against the model's own text, which is also the source of the self-aligned training targets; the cross-benchmark comparison does not involve baselines, so the headline intelligibility gain is partly forced by construction.

  1. self definitional [Section 3.3, Eqs. (2)-(4); Section 5.4, Table 3]
    "Aligned speech supervision is then synthesized from the predicted text using a dialect TTS model. ... The text responses generated by the model are used as references, while FRASR2-AED transcribes the generated speech responses for CER computation."

    The self-aligned supervision U is built from the model's own text prediction T = Thinker(X) via U = TTS(T), and the Talker loss (Eq. 4) optimizes the model to produce that U given hidden states H. The intelligibility metric in Section 5.4 then computes CER against 'the text responses generated by the model' — i.e., the same T used to create the supervision. Thus the measured CER is a self-consistency score between the model's text and its own speech, which is precisely the quantity the self-aligned objective trains to minimize. The improvement over sft-2ep/sft-3ep is therefore at least in part a consequence of the training objective and evaluation protocol, not an independent measure of human-perceived dialect intelligibility.

full rationale

DialectS2S contains substantial non-circular engineering: the dialect data synthesis pipeline (LLM rewriting plus CosyVoice2 synthesis with UTMOS filtering), the mixed-data SFT objective, and the external evaluations for language matching (FireRedLID) and response quality (Qwen3-Plus text scoring) are all independent of the paper's novelty claim. The self-cited base model OpenS2S [17] is an open-source artifact and is used as a starting point, not as an unverifiable uniqueness argument. The circularity is localized to the speech-intelligibility claim. In Eq. (3), the 'aligned' speech supervision is TTS of the model's own predicted text; in Section 5.4, CER is computed against that same self-generated text. Consequently, the intelligibility gain reported in Tables 3 and 5 partly reduces to how well the model matches its own text, the exact objective of Eq. (4). The comparison to the FRASR2-AED benchmark average is not like-for-like (natural vs. synthetic speech, different references, no baselines under the same protocol), so the headline claim of 'substantial improvements in speech intelligibility' over baselines is unsupported as stated. This is a partial circularity affecting one of the three headline evaluation axes, not the entire framework; hence score 6 rather than higher.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The method relies on several unverified modeling and data assumptions but introduces no new physical or ontological entities. The key free parameters are the hand-selected seed utterances and the unspecified UTMOS filtering threshold; the training objectives in Eqs. (1) and (4) are standard cross-entropy losses with no fitted constants beyond ordinary hyperparameters.

free parameters (2)
  • Dialect seed utterances = 100 per dialect
    Section 3.2 selects 100 seed utterances per dialect from KeSpeech, WenetSpeech-Yue, and WenetSpeech-Chuan; the count and speaker mix are hand-chosen and directly condition the synthesized training voices.
  • UTMOS filtering threshold = Not specified
    Section 3.2 filters synthesized samples by UTMOS score but never reports the threshold, making the training data composition a hidden free parameter.
assumptions (4)
  • domain assumption Hidden representations H produced by the Thinker encode enough semantic content for the Talker to synthesize speech.
    Section 3.3 frames all speech dialogue models as Thinker-Talker; the self-aligned stage and loss LAlign_Talker in Eq. (4) presuppose that H is a sufficient conditioning signal for speech generation.
  • domain assumption The evolution of hidden distributions during dialect SFT is the cause of degraded speech quality, and aligning supervision with the model's own text predictions mitigates this.
    Section 3.3 and Figure 2 assert a hidden-state shift after SFT and propose self-aligned supervision without a controlled experiment isolating this mechanism.
  • domain assumption LLM-rewritten dialect texts and CosyVoice2-synthesized dialect speech faithfully represent the target dialects.
    Section 3.2 builds the training corpus on these outputs with no human validation or comparison against native dialect speech.
  • domain assumption UTMOS, FireRedLID, and FRASR2-AED provide valid automatic proxies for naturalness, dialect identity, and intelligibility.
    Sections 3.2 and 5 use these off-the-shelf tools as ground truth for filtering and evaluation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DialectS2S: End-to-End Speech Dialogue Modeling for Low-Resource Chinese Dialects." pith.science (2026). https://pith.science/paper/IJYLSXZL

@misc{pith2026260808067,
  author       = {Pith},
  title        = {Pith review of: DialectS2S: End-to-End Speech Dialogue Modeling for Low-Resource Chinese Dialects},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IJYLSXZL}},
  note         = {Machine review of arXiv:2608.08067}
}
read the original abstract

Current end-to-end speech dialogue models are primarily optimized for mainstream languages and remain limited in low-resource dialect scenarios due to the scarcity of dialect speech data. Moreover, during dialect adaptation, the semantic representation space of speech dialogue models continuously evolves, while conventional speech supervision remains unchanged, leading to semantic inconsistency between hidden representations and speech targets and degrading speech stability and naturalness. To address these issues, we propose DialectS2S, an end-to-end speech dialogue model for Chinese dialects. We first develop a scalable dialect speech dialogue synthesis pipeline for efficient data construction. We further introduce a two-stage post-training strategy with self-aligned speech supervision, which aligns the semantic content of speech supervision with the evolved semantic representations of the model to improve dialect speech generation quality. Experimental results show that DialectS2S consistently outperforms existing baselines across multiple Chinese dialects in speech dialogue, achieving substantial improvements in dialect consistency, response quality, and speech intelligibility. Our work provides an efficient and scalable solution for end-to-end speech dialogue modeling in low-resource dialect scenarios. To facilitate future research and practical applications, we fully open-source the DialectS2S framework, including model checkpoints, training datasets, and fine-tuning code.

Figures

Figures reproduced from arXiv: 2608.08067 by the authors.

Figure 1
Figure 1. The proposed unified Thinker-Talker architecture [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Visualization of the final-layer Thinker hidden states before and after su￾pervised fine-tuning. As illustrated in [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Self-aligned supervision aligns speech targets with semantic representations produced by the Thinker module, enabling the Talker module to focus on prosody and pronunciation modeling. By aligning speech supervision with the model’s semantic predictions, the proposed strategy reduces semantic mismatch between hidden representations [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

24 extracted references · 6 canonical work pages

  1. [1]

    arXiv preprint arXiv:2509.12508 (2025)

    An, K., Chen, Y., Chen, Z., Deng, C., Du, Z., Gao, C., Gao, Z., Gong, B., Li, X., Li, Y., et al.: Fun-asr technical report. arXiv preprint arXiv:2509.12508 (2025)

    arXiv 2025

  2. [2]

    arXiv preprint arXiv:2509.22727 (2025)

    Chen, Z., Chen, G., Wang, Y., Ding, C., Zhang, W.Q., et al.: Diamoe-tts: A unified ipa-based dialect tts framework with mixture-of-experts and parameter-efficient zero-shot adaptation. arXiv preprint arXiv:2509.22727 (2025)

    arXiv 2025

  3. [3]

    arXiv preprint arXiv:2604.27393 (2026)

    Cui, J., Xu, B., Wang, C., Yu, T., Sun, W., Xu, Y., Wang, T., He, Z., Ma, W., Cai, T., et al.: Minicpm-o 4.5: Towards real-time full-duplex omni-modal interaction. arXiv preprint arXiv:2604.27393 (2026)

    arXiv 2026

  4. [4]

    In: ICASSP 2026 - 2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

    Dai, Y., Zhang, Z., Wang, S., Li, L., Guo, Z., Zuo, T., Wang, S., Xue, H., Wang, C., Wang, Q., Xu, X., Bu, H., Li, J., Kang, J., Zhang, B., Xie, L.: Wenetspeech-chuan: A large-scale sichuanese corpus with rich annotation for dialectal speech processing. In: ICASSP 2026 - 2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP...

    arXiv 2026

  5. [5]

    arXiv preprint arXiv:2410.00037 (2024)

    Défossez, A., Mazaré, L., Orsini, M., Royer, A., Pérez, P., Jégou, H., Grave, E., Zeghidour, N.: Moshi: a speech-text foundation model for real-time dialogue. arXiv preprint arXiv:2410.00037 (2024)

    arXiv 2024

  6. [6]

    arXiv preprint arXiv:2504.18425 (2025)

    Ding, D., Ju, Z., Leng, Y., Liu, S., Liu, T., Shang, Z., Shen, K., Song, W., Tan, X., Tang, H., et al.: Kimi-audio technical report. arXiv preprint arXiv:2504.18425 (2025)

    arXiv 2025

  7. [7]

    arXiv preprint arXiv:2412.10117 (2024)

    Du, Z., Wang, Y., Chen, Q., Shi, X., Lv, X., Zhao, T., Gao, Z., Yang, Y., Gao, C., Wang, H., et al.: Cosyvoice 2: Scalable streaming speech synthesis with large language models. arXiv preprint arXiv:2412.10117 (2024)

    arXiv 2024

  8. [8]

    In: International Conference on Learning Representations

    Fang, Q., Guo, S., Zhou, Y., Ma, Z., Zhang, S., Feng, Y.: Llama-omni: Seamless speech interaction with large language models. In: International Conference on Learning Representations. vol. 2025, pp. 57607–57624 (2025)

    work page 2025

Show all 24 references
  1. [9]

    Proceedings of the AAAI Conference on Artificial Intelligence40, 31627– 31635 (Mar 2026).https://doi.org/10.1609/aaai.v40i37.40429

    Li, L., Guo, Z., Chen, H., Dai, Y., Zhang, Z., Xue, H., Zuo, T., Wang, C., Wang, S., Xu, X., Bu, H., Li, J., Kang, J., Zhang, B., Yuan, R., Zhou, Z., Xue, W., Xie, L.: Wenetspeech-yue: Alarge-scalecantonesespeech corpuswithmulti-dimensionalan- notation. Proceedings of the AAAI...

    2026 doi

  2. [10]

    arXiv preprint arXiv:1711.05101 (2017)

    Loshchilov, I., Hutter, F.: Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101 (2017)

    2017 arXiv

  3. [11]

    In: SC20: international conference for high performance computing, networking, storage and analysis

    Rajbhandari, S., Rasley, J., Ruwase, O., He, Y.: Zero: Memory optimizations to- ward training trillion parameter models. In: SC20: international conference for high performance computing, networking, storage and analysis. pp. 1–16. IEEE (2020)

    2020

  4. [12]

    In: Proceed- ings of the 26th ACM SIGKDD international conference on knowledge discovery & data mining

    Rasley, J., Rajbhandari, S., Ruwase, O., He, Y.: Deepspeed: System optimizations enable training deep learning models with over 100 billion parameters. In: Proceed- ings of the 26th ACM SIGKDD international conference on knowledge discovery & data mining. pp. 3505–3506 (2020)

    2020

  5. [13]

    Saeki, T., Xin, D., Nakata, W., Koriyama, T., Takamichi, S., Saruwatari, H.: UT- MOS:UTokyo-SaruLabSystemforVoiceMOSChallenge2022.In:Interspeech2022. pp. 4521–4525 (2022).https://doi.org/10.21437/Interspeech.2022-439

    2022 doi

  6. [14]

    arXiv preprint arXiv:2505.02707 (2025)

    Shi, Y., Shu, Y., Dong, S., Liu, G., Sesay, J., Li, J., Hu, Z.: Voila: Voice-language foundation models for real-time autonomous interaction and voice role-play. arXiv preprint arXiv:2505.02707 (2025)

    2025 arXiv

  7. [15]

    Shu et al

    Tang, Z., Wang, D., Xu, Y., Sun, J., Lei, X., Zhao, S., Wen, C., Tan, X., Xie, C., Zhou, S., et al.: Kespeech: An open source speech dataset of mandarin and its eight 12 Y. Shu et al. subdialects. In: Thirty-fifth conference on neural information processing systems datasets an...

    2021

  8. [16]

    arXiv preprint arXiv:2511.15848 (2025)

    Tian, F., Zhang, X.T., Zhang, Y., Zhang, H., Li, Y., Liu, D., Deng, Y., Wu, D., Chen, J., Zhao, L., et al.: Step-audio-r1 technical report. arXiv preprint arXiv:2511.15848 (2025)

    2025

  9. [17]

    In: Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing: System Demonstrations

    Wang, C., Peng, T., Yang, W., Bai, Y., Wang, G., Lin, J., Jia, L., Wu, L., Wang, J., Zong, C., et al.: Opens2s: Advancing fully open-source end-to-end empathetic large speech language model. In: Proceedings of the 2025 Conference on Empirical Methods in Natural Language Proces...

    2025

  10. [18]

    arXiv preprint arXiv:2507.16632 (2025)

    Wu, B., Yan, C., Hu, C., Yi, C., Feng, C., Tian, F., Shen, F., Yu, G., Zhang, H., Li, J., et al.: Step-audio 2 technical report. arXiv preprint arXiv:2507.16632 (2025)

    2025 arXiv

  11. [19]

    arXiv preprint arXiv:2503.20215 (2025)

    Xu, J., Guo, Z., He, J., Hu, H., He, T., Bai, S., Chen, K., Wang, J., Fan, Y., Dang, K., et al.: Qwen2.5-omni technical report. arXiv preprint arXiv:2503.20215 (2025)

    2025 arXiv

  12. [20]

    arXiv preprint arXiv:2509.17765 (2025)

    Xu, J., Guo, Z., Hu, H., Chu, Y., Wang, X., He, J., Wang, Y., Shi, X., He, T., Zhu, X., et al.: Qwen3-omni technical report. arXiv preprint arXiv:2509.17765 (2025)

    2025 arXiv

  13. [21]

    arXiv preprint arXiv:2603.10420 (2026)

    Xu, K., Jia, Y., Huang, K., Chen, J., Li, W., Liu, K., Xie, F.L., Tang, X., Hu, Y.: Fireredasr2s: A state-of-the-art industrial-grade all-in-one automatic speech recognition system. arXiv preprint arXiv:2603.10420 (2026)

    2026

  14. [22]

    arXiv preprint arXiv:2505.09388 (2025)

    Yang, A., Li, A., Yang, B., Zhang, B., Hui, B., Zheng, B., Yu, B., Gao, C., Huang, C., Lv, C., et al.: Qwen3 technical report. arXiv preprint arXiv:2505.09388 (2025)

    2025 arXiv

  15. [23]

    arXiv preprintarXiv:2412.02612(2024).https://doi.org/10.48550/arXiv.2412.02612

    Zeng, A., Du, Z., Liu, M., Wang, K., Jiang, S., Zhao, L., Dong, Y., Tang, J.: Glm-4-voice: Towards intelligent and human-like end-to-end spoken chatbot. arXiv preprintarXiv:2412.02612(2024).https://doi.org/10.48550/arXiv.2412.02612

    2024 doi arXiv

  16. [24]

    Zhang, Y., Du, Y., Dai, Z., Ma, X., Kou, K., Wang, B., Li, H.: Echox: Towards mitigating acoustic-semantic gap via echo training for speech-to-speech llms. arXiv preprint arXiv:2509.09174 (2025) A Additional Experimental Details Response-quality rubric.Qwen3-Plus evaluates eve...

    2025 arXiv

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.