Pith. sign in

REVIEW 4 major objections 6 minor 29 references

M3-DuplexBench: A Multi-Turn, Multilingual, Multidomain Benchmark for Full-Duplex Spoken Dialogue Models

T0 review · 4 major / 6 minor · reviewed 2026-08-03 · deepseek-v4-flash

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

Pith's one-line read M3-DuplexBench evaluates full-duplex spoken dialogue models under three context conditions and shows that supplying a coherent dialogue history improves both answer accuracy and turn-taking timing.

desk verdict Useful benchmark with a real comparability fix, but the Full-context timing results may partly measure the model copying the reference timing schedule rather than intrinsic multi-turn ability. read the letter →

arxiv 2607.29125 v1 pith:J6PGU4IK submitted 2026-07-31 cs.CL

Ryo Fukuda , Atsushi Ando , Hiroki Kanagawa , Takatomo Kano , Marc Delcroix , Naohiro Tawara , Yuya Chiba This is my paper
classification cs.CL
keywords full-duplexspokendialoguebenchmarkturn-takingmulti-turnevaluationteacher-forcedconditioningmultilingualtask-orientedchat
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

This paper introduces M3-DuplexBench, a benchmark for full-duplex spoken dialogue systems that can listen while speaking. Its central claim is that multi-turn evaluation becomes fair and informative when every model is forced to condition on the same complete dialogue history, including the system's own previous speech. Under that protocol, the paper shows that dialogue history improves answer accuracy in multi-turn question answering, that models adapt their response timing to the rhythm of the context, that turn-taking is easier in task-oriented QA than in casual chat, and that Japanese full-duplex models' main gap is language understanding rather than timing.

What carries the argument

The key mechanism is teacher-forced full-context conditioning: before the target event, the model's state is driven by both user and system speech from a pre-recorded dialogue, so all models see the identical coherent history. This is applied to four event types (turn shift, pause, backchannel, barge-in) with timing and content metrics, and compared against single-turn (None) and user-history-only (User) conditions.

What would settle it

Run a free-running simulation where the model itself produces the system speech for the same dialogue, then evaluate at the same event; if timing and content scores diverge systematically from the teacher-forced condition (especially in chat), the full-context protocol is measuring an artifact of forced conditioning rather than natural multi-turn ability.

Watch

Extended reading notes

Core claim

M3-DuplexBench evaluates models under three context conditions — no history, user history only, and teacher-forced full history — and the paper's main discovery is that the full-context condition is the most reliable static setting, because it removes the context mismatch that plagues user-only conditioning. Using this protocol on Moshi-based models, the paper finds that dialogue history helps English models answer context-dependent questions, that models move their turn-taking latency closer to human reference timing when given full context, and that the Japanese-language models evaluated lag primarily in content generation, not in turn-taking behavior.

Load-bearing premise

The load-bearing premise is that teacher-forced full-context conditioning produces valid, natural model behavior at the evaluation window, even though the model's state is driven by pre-recorded system speech it did not generate itself.

Editorial extensions

If this is right

  • If the full-context protocol is accepted, future FDSDS benchmarks can compare models on identical multi-turn histories rather than divergent simulated conversations.
  • Dialogue history should be treated as a genuine input feature in FDSDS evaluation, since it measurably changes both timing and content.
  • The observed cross-domain gap suggests that turn-taking evaluations should be domain-specific; chat and task-oriented QA place different demands on when to speak.
  • The cross-lingual finding points Japanese full-duplex development toward content generation and understanding rather than timing subroutines.
  • The paper's method for synthesizing task-oriented spoken dialogues from text QA corpora can be reused to build benchmark data for other languages.

Reading between the lines

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

  • Because the full-context condition forces the model to follow a pre-recorded system trajectory, it may overestimate a model's ability to maintain coherent multi-turn state when the model would have chosen different wording; a natural extension is to check agreement between teacher-forced and free-running continuations on the same history.
  • The finding that User context often degrades timing suggests that earlier multi-turn benchmarks that used user-only conditioning may have understated models' true multi-turn timing ability; this could be tested by re-running those benchmarks with full-context conditioning.
  • The synthetic task-domain dialogues are built by sampling timing statistics from chat corpora; an untested assumption is whether chat timing transfers to QA, so the paper's cross-domain conclusions could be validated with human-negotiated QA timing samples.
  • The benchmark currently only supports Moshi-style architectures because of the parallel-stream conditioning requirement; if the protocol were adapted to cascaded models, it might reveal whether the cross-lingual content gap is specific to end-to-end systems.
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 M3-DuplexBench, a multi-turn, multilingual (English/Japanese), multidomain (chat and task-oriented QA) benchmark for full-duplex spoken dialogue systems. It defines four event types (turn shift, pause, backchannel, barge-in) and evaluates models under three context conditions: None, user-side-only, and teacher-forced full-context. Experiments with four Moshi-based models report timing and content metrics and claim that (1) dialogue history helps multi-turn QA, (2) models adjust response timing to dialogue context, (3) smooth turn-taking is easier in task-oriented than chat dialogues, and (4) Japanese models lag mainly in content rather than timing. The paper proposes a representative value averaging the None and Full conditions and compares models across languages and domains.

Significance. If the protocol is valid, M3-DuplexBench would be a useful, controlled multi-turn evaluation framework for FDSDSs, extending coverage to Japanese and to task-oriented QA with multi-turn context. The idea of comparing context conditions is thoughtful and the event extraction follows prior work. However, the current manuscript does not release code/data, and several load-bearing aspects of the evaluation—particularly the timing information implicitly provided by teacher-forced full-context conditioning and the unexplained timing statistics of the synthetic task dialogues—need to be addressed before the stated findings can be relied upon. The content evaluations also lack reliability evidence. With these concerns resolved, the benchmark could be a solid contribution to the community.

major comments (4)
  1. [Section IV-A / V-B.1] In the Full condition, the model conditions on the reference system speech stream, so it observes the exact onset/offset times of all previous system turns. Moshi-style models can use these inter-turn intervals to predict the current SHIFT timing. The large latency reductions under Full (e.g., Moshi T1: 2.453s -> T3: 0.430s) and the fact that Full latencies approach the dataset means in Table II are thus compatible with timing imitation rather than context-based adjustment. The paper's teacher-forced limitation statement does not address this channel. Please add a control that jitters/time-shifts the system turns in the conditioning stream; if latency follows the shifted schedule, the timing findings are artifacts.
  2. [Section III-B.2 / Table II / V-B.3] The synthetic task-oriented dialogues are built by sampling SHIFT/PAUSE/BC from the chat reference corpora, yet Table II reports average SHIFT durations of 0.45s (En) and 0.35s (Ja) for task dialogues vs 0.88s/0.72s for chat. This large, unexplained discrepancy means the cross-domain timing comparison is not controlled: task SHIFT events are intrinsically shorter, so Full-condition models will naturally show lower latency and higher TOR. Please explain the discrepancy, and either match SHIFT distributions across domains or validate the synthetic timing against natural task-oriented speech.
  3. [Section IV-B.2] All content metrics (relevance, consistency, QA accuracy) rely on a single GPT-5 nano judge, with no inter-judge agreement, no human validation, and no statement of whether the judge inputs are ASR transcripts or audio. Whisper errors on Japanese TTS output could differentially lower content scores, threatening the cross-lingual and context-effect claims (e.g., T8 vs T12). Report judge reliability on a sample, use multiple judges, and specify the judge input.
  4. [Overall / Data Availability] The manuscript does not state whether the benchmark data, event extraction code, synthetic dialogue generation pipeline, or evaluation scripts will be released. As a benchmark contribution, this absence prevents reproduction and adoption. An availability statement with a public release is required for the contribution to be usable.
minor comments (6)
  1. [Section III-A] The notation e=(ℓ,s,t) is not fully defined for SHIFT events with overlaps (t<s). Clarify how s and t relate to the speaker-change boundary and how the evaluation window W=[s,s+Δ] is anchored.
  2. [Section V-B.1 / Tables IV-V] The 'None&Full' representative value is an unweighted average of two qualitatively different conditions. Provide a rationale or report the two conditions separately in summary tables.
  3. [Figures 2-3] The text in the case-study figures is very small; consider providing the example transcripts in a readable table or enlarged font.
  4. [References / Table I] Reference [9] appears as both MTR-DuplexBench and MTRDuplexBench; standardize the spelling. Table I uses 'MTR' without defining it in the caption.
  5. [Section V-A] It would be helpful to state the model sizes or parameter counts for the four evaluated models to aid comparison.
  6. [Table II] Clarify what 'dur' represents (mean SHIFT duration? standard deviation?) and how it is computed from the generated dialogues.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: benchmark is externally grounded and model results are empirical, not derived from fitted inputs.

full rationale

M3-DuplexBench is a benchmark-construction paper, not a derivation chain. The evaluation data come from external sources (Candor, MagicData, TopiOCQA) and the timing metrics follow existing benchmarks (Full-Duplex-Bench, Full-Duplex-Bench v1.5). The models evaluated (Moshi, PersonaPlex, J-Moshi, LLM-jp-Moshi) are independent systems; none of the reported latencies, TORs, or content scores are fitted parameters or derived from the benchmark's inputs by equation. The central findings are empirical comparisons across context conditions, domains, and languages. There is no self-citation: reference [9] (MTR-DuplexBench) is by a different group, and the paper does not invoke any uniqueness theorem or load-bearing prior claim from the same authors. The paper explicitly acknowledges a teacher-forced trajectory gap in Section V-B.1, which is a validity limitation, not a circular reduction: the Full condition supplies reference system-side audio, so the model has access to reference timing intervals, but the model's response latency is measured, not set equal to those intervals. This is a potential confound for the timing finding, but it does not make the result equivalent to its input by construction. The benchmark's synthetic task dialogue pipeline samples timing statistics from reference chat corpora, but the resulting task SHIFT durations differ (Table II) and are reported as data statistics, not as predictions. No step reduces a claimed prediction to a fitted value, a renamed known result, or a self-citation chain. Therefore, no circularity is present.

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

No invented entities or free parameters fitted to the central claims. The paper relies on the domain assumptions listed: Moshi-style model support, GPT-5 nano judge validity, synthetic pipeline fidelity, timestamp reliability, and fixed event thresholds. These are not circular, but they are the main places a replication could break.

assumptions (5)
  • domain assumption Moshi-based architecture is the only architecture supporting full-context conditioning on parallel user/system streams.
    Section V-A: 'We focus on Moshi-based models because Full setting requires direct conditioning on parallel user and system speech streams.' This scopes conclusions to one architecture family; cascaded systems and other duplex architectures are excluded.
  • domain assumption LLM-as-a-judge (GPT-5 nano) scores on 0-2 scales for relevance/consistency/QA accuracy are valid without calibration.
    Section IV-B.2: content metrics rely on GPT-5 nano as judge; no inter-judge agreement or human correlation is reported. The soundness of content scores depends on the judge's reliability.
  • domain assumption Synthetic spoken dialogues from text QA + TTS + sampled timing statistics represent natural task-oriented spoken dialogue.
    Section III-B.1: utterances synthesized with CosyVoice 2 and placed using pause/shift/BC statistics sampled from Candor/MagicData. No validation of naturalness or ecological validity is provided; all task-domain conclusions assume this premise.
  • domain assumption Whisper ASR + Montreal Forced Aligner produce reliable word timestamps for model output across English and Japanese.
    Section IV-A: timing metrics derived from ASR + forced alignment on generated speech; ASR/alignment errors propagate into latency/TOR, no error analysis on timestamp accuracy.
  • domain assumption The event definitions (0.5 s silence IPU, 0.8 s BC threshold, 0.4 s overlap threshold) are appropriate across languages/domains.
    Section III-A: fixed thresholds taken from prior work; no sensitivity analysis; Japanese and task-domain conversations may have different pause/overlap distributions.

how reviews work

0 comments
Cite this review

Pith. "Pith review of M3-DuplexBench: A Multi-Turn, Multilingual, Multidomain Benchmark for Full-Duplex Spoken Dialogue Models." pith.science (2026). https://pith.science/paper/J6PGU4IK

@misc{pith2026260729125,
  author       = {Pith},
  title        = {Pith review of: M3-DuplexBench: A Multi-Turn, Multilingual, Multidomain Benchmark for Full-Duplex Spoken Dialogue Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/J6PGU4IK}},
  note         = {Machine review of arXiv:2607.29125}
}
read the original abstract

Full-duplex spoken dialogue systems (FDSDSs) can listen while speaking, enabling natural behaviors such as smooth turn-taking, backchannel handling, and user barge-in handling. However, fair comparisons in multi-turn conversations remain a challenge. In addition, existing benchmarks provide limited coverage of languages and dialogue domains. We propose M3-DuplexBench, a multi-turn, multilingual, multidomain benchmark for FDSDSs. M3-DuplexBench supports English and Japanese and covers both casual conversation and multi-turn question answering. In addition, we evaluate models under multiple dialogue context settings, including single-turn, user-only, and teacher-forced full-context settings, to analyze how dialogue history affects model behavior. Experiments with recent FDSDSs reveal model-specific turn-taking characteristics, clear performance gaps across languages and domains, and mixed effects of dialogue context.

Figures

Figures reproduced from arXiv: 2607.29125 by the authors.

Figure 1
Figure 1. Overview of the proposed M3-DuplexBench evaluation framework. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Case study showing the effect of dialogue context in task-oriented QA. [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Case studies of model responses in multi-turn QA. In these examples, [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

29 extracted references · 4 linked inside Pith

  1. [1]

    From turn-taking to synchronous dialogue: A survey of full-duplex spoken language models,

    Y . Chen and H. Yu, “From turn-taking to synchronous dialogue: A survey of full-duplex spoken language models,”arXiv preprint arXiv:arXiv:2509.14515, 2025

    arXiv 2025

  2. [2]

    Turn-taking in conversational systems and human-robot interaction: A review,

    G. Skantze, “Turn-taking in conversational systems and human-robot interaction: A review,”Computer Speech & Language, vol. 67, 2021

    2021

  3. [3]

    Generative spoken dialogue language modeling,

    T. A. Nguyen, E. Kharitonov, J. Copet, Y . Adi, W.-N. Hsu, A. Elkahky, P. Tomasello, R. Algayres, B. Sagot, A. Mohamedet al., “Generative spoken dialogue language modeling,”Transactions of the Association for Computational Linguistics, vol. 11, pp. 250–266, 2023

    2023

  4. [4]

    Efficient and Direct Duplex Modeling for Speech-to-Speech Language Model,

    K. Hu, E. Hosseini-Asl, C. Chen, E. Casanova, S. Ghosh, P. ˙Zelasko, Z. Chen, J. Li, J. Balam, and B. Ginsburg, “Efficient and Direct Duplex Modeling for Speech-to-Speech Language Model,” in26th Annual Conference of the International Speech Communication Association (INTERSPEECH 2025), 2025, pp. 2715–2719

    2025

  5. [5]

    Talking turns: Benchmarking audio foundation models on turn-taking dynamics,

    S. Arora, Z. Lu, C.-C. Chiu, R. Pang, and S. Watanabe, “Talking turns: Benchmarking audio foundation models on turn-taking dynamics,” in13th International Conference on Learning Representations (ICLR 2025), 2025

    2025

  6. [6]

    Full-duplex-bench: A benchmark to evaluate full-duplex spoken dialogue models on turn-taking capabilities,

    G.-T. Lin, J. Lian, T. Li, Q. Wang, G. Anumanchipalli, A. H. Liu, and H.- Y . Lee, “Full-duplex-bench: A benchmark to evaluate full-duplex spoken dialogue models on turn-taking capabilities,” in2025 IEEE Automatic Speech Recognition and Understanding Workshop (ASRU), 2025, pp. 1–8

    2025

  7. [7]

    Full-duplex-bench v1.5: Evaluating overlap handling for full-duplex speech models,

    G.-T. Lin, S.-Y . S. Kuan, Q. Wang, J. Lian, T. Li, S. Watanabe, and H.-y. Lee, “Full-duplex-bench v1.5: Evaluating overlap handling for full-duplex speech models,” in2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP 2026). IEEE, 2026, pp. 19 447–19 451

    2026

  8. [8]

    Full-duplex-bench-v2: A multi-turn evaluation framework for duplex dialogue systems with an automated examiner,

    G.-T. Lin, S.-Y . S. Kuan, J. Shi, K.-W. Chang, S. Arora, S. Watanabe, and H.-y. Lee, “Full-duplex-bench-v2: A multi-turn evaluation framework for duplex dialogue systems with an automated examiner,” in64th Annual Meeting of the Association for Computational Linguistics (ACL 2026), 2026, pp. 27–36

    2026

Show all 29 references
  1. [9]

    MTR-DuplexBench: Towards a comprehensive evaluation of multi-round conversations for full-duplex speech language models,

    Z. He, W. Cui, H. Xu, X.-H. Li, L. Zhu, H. Bai, M. Shaohua, and I. King, “MTR-DuplexBench: Towards a comprehensive evaluation of multi-round conversations for full-duplex speech language models,” in Findings of the Association for Computational Linguistics (ACL 2026), M. Liaka...

    2026

  2. [10]

    Towards a japanese full-duplex spoken dialogue system,

    A. Ohashi, S. Iizuka, J. Jiang, and R. Higashinaka, “Towards a japanese full-duplex spoken dialogue system,” in26th Annual Conference of the International Speech Communication Association (INTERSPEECH 2025), 2025, pp. 1783–1787

    2025

  3. [11]

    Effects of dialogue corpora properties on fine-tuning a moshi-based spoken dialogue model,

    Y . Abe, M. Saeki, A. Ohashi, S. Takamichi, S. Fujie, T. Kobayashi, T. Ogawa, and R. Higashinaka, “Effects of dialogue corpora properties on fine-tuning a moshi-based spoken dialogue model,” in16th Interna- tional Workshop on Spoken Dialogue System Technology (IWSDS 2026), 202...

    2026

  4. [12]

    The ICASSP 2026 HumDial challenge: Benchmarking human-like spoken dialogue systems in the LLM era,

    Z. Zhao, S. Wang, G. Li, H. Xue, C. Wang, S. Wang, L. Xiao, Z. Zhang, H. Bu, X. Xuet al., “The ICASSP 2026 HumDial challenge: Benchmarking human-like spoken dialogue systems in the LLM era,” in 2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICAS...

    2026

  5. [13]

    FD-Bench: A Full-Duplex Benchmarking Pipeline Designed for Full Duplex Spoken Dialogue Systems,

    Y . Peng, Y .-W. Chao, D. Ng, Y . Ma, C. Ni, B. Ma, and E. S. Chng, “FD-Bench: A Full-Duplex Benchmarking Pipeline Designed for Full Duplex Spoken Dialogue Systems,” in26th Annual Conference of the International Speech Communication Association (INTERSPEECH 2025), 2025, pp. 176–180

    2025

  6. [14]

    SoulX-duplug: Plug-and-play streaming state prediction module for realtime full-duplex speech conversation,

    R. Yan, W. Chen, Z. Liu, Z. Ma, H. Lin, H. Wen, H. Xie, J. Wu, Y . Liang, Y . Zhao, P. Feng, J. Qian, H. Meng, Y . Dai, S. Yin, M. Tao, L. Xie, K. Yu, X. Wang, and X. Chen, “SoulX-duplug: Plug-and-play streaming state prediction module for realtime full-duplex speech conversat...

    2026

  7. [15]

    The candor corpus: Insights from a large multimodal dataset of naturalistic conversation,

    A. Reece, G. Cooney, P. Bull, C. Chung, B. Dawson, C. Fitzpatrick, T. Glazer, D. Knox, A. Liebscher, and S. Marin, “The candor corpus: Insights from a large multimodal dataset of naturalistic conversation,” Science advances, vol. 9, 2023

    2023

  8. [16]

    Japanese duplex conversation training dataset,

    Beijing Magic Data Technology Co., Ltd., “Japanese duplex conversation training dataset,” https://magichub.com/datasets/ japanese-duplex-conversation-training-dataset/, 2025

    2025

  9. [17]

    TopiOCQA: Open-domain conversational question answering with topic switching,

    V . Adlakha, S. Dhuliawala, K. Suleman, H. de Vries, and S. Reddy, “TopiOCQA: Open-domain conversational question answering with topic switching,”Transactions of the Association for Computational Linguistics, vol. 10, pp. 468–483, 2022

    2022

  10. [18]

    DiaSynth: Synthetic dialogue generation framework for low resource dialogue ap- plications,

    S. K. Suresh, W. Mengjun, T. Pranav, and E. S. Chng, “DiaSynth: Synthetic dialogue generation framework for low resource dialogue ap- plications,” inFindings of the Association for Computational Linguistics (NAACL 2025), 2025, pp. 673–690

    2025

  11. [19]

    SpeechDialogueFactory: A Framework for Natural Speech Dialogue Generation,

    M. Wang, Y . Bai, Y . Wang, T.-T. Vu, E. Shareghi, and G. Haffari, “SpeechDialogueFactory: A Framework for Natural Speech Dialogue Generation,” in26th Annual Conference of the International Speech Communication Association (INTERSPEECH 2025), 2025, pp. 1758– 1762

    2025

  12. [20]

    Behavior-SD: Behaviorally aware spoken dialogue generation with large language models,

    S. Lee, K.-w. Kim, and G. Kim, “Behavior-SD: Behaviorally aware spoken dialogue generation with large language models,” inProceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technolo- gies (...

    2025

  13. [21]

    Gemma 4 31b,

    Google DeepMind, “Gemma 4 31b,” https://huggingface.co/google/ gemma-4-31B, 2026

    2026

  14. [22]

    Cosyvoice 2: Scalable streaming speech synthesis with large language models,

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

    2024 arXiv

  15. [23]

    Robust speech recognition via large-scale weak super- vision,

    A. Radford, J. W. Kim, T. Xu, G. Brockman, C. McLeavey, and I. Sutskever, “Robust speech recognition via large-scale weak super- vision,”arXiv preprint arXiv:2212.04356, 2022

    2022 arXiv

  16. [24]

    faster-whisper: Faster whisper transcription with ctrans- late2,

    SYSTRAN, “faster-whisper: Faster whisper transcription with ctrans- late2,” https://github.com/SYSTRAN/faster-whisper, 2023

    2023

  17. [25]

    Montreal Forced Aligner: Trainable Text-Speech Alignment Using Kaldi,

    M. McAuliffe, M. Socolof, S. Mihuc, M. Wagner, and M. Sonderegger, “Montreal Forced Aligner: Trainable Text-Speech Alignment Using Kaldi,” inProceedings of the 18th Annual Conference of the Inter- national Speech Communication Association (INTERSPEECH 2017), 2017, pp. 498–502

    2017

  18. [26]

    GPT-5 nano Model,

    OpenAI, “GPT-5 nano Model,” https://developers.openai.com/api/docs/ models/gpt-5-nano, 2026, accessed: 2026-06-15

    2026

  19. [27]

    Moshi: a speech-text foundation model for real-time dialogue,

    A. D ´efossez, L. Mazar ´e, M. Orsini, A. Royer, P. P ´erez, H. J ´egou, E. Grave, and N. Zeghidour, “Moshi: a speech-text foundation model for real-time dialogue,”arXiv preprint arXiv:2410.00037, 2024

    2024 arXiv

  20. [28]

    Freeze-omni: A smart and low latency speech-to-speech dialogue model with frozen LLM,

    X. Wang, Y . Li, C. Fu, Y . Shen, L. Xie, K. Li, X. Sun, and L. Ma, “Freeze-omni: A smart and low latency speech-to-speech dialogue model with frozen LLM,”arXiv preprint arXiv:2411.00774, 2024

    2024 arXiv

  21. [29]

    PersonaPlex: V oice and role control for full duplex conversational speech models,

    R. Roy, J. Raiman, S.-G. Lee, T.-D. Ene, R. Kirby, S. Kim, J. Kim, and B. Catanzaro, “PersonaPlex: V oice and role control for full duplex conversational speech models,”arXiv preprint arXiv:2602.06053, 2026

    2026

Pith tools

Reviewed August 3, 2026 · model on record in the stance chip above.