pith. sign in

arxiv: 2607.01733 · v1 · pith:CAMJIUCJnew · submitted 2026-07-02 · 💻 cs.CL · eess.AS

Rethinking Speech-LLM Integration for ASR: Effective Joint Speech-Text Training by Interleaving

Ruchao Fan , Yiming Wang , Rui Zhao , Liliang Ren , Keqi Deng , Xiaoyang Chen , Ali Zare , Bo Ren
show 5 more authors
Yuxuan Hu Junkun Chen Yan Huang Yelong Shen Jinyu Li
This is my paper

Pith reviewed 2026-07-03 15:15 UTC · model grok-4.3

classification 💻 cs.CL eess.AS
keywords speech recognitionlarge language modelsjoint traininginterleaved pretrainingentity recognitionmodality gapautomatic speech recognitiondomain adaptation
0
0 comments X

The pith

Interleaving speech and text in pretraining improves entity accuracy for large-scale automatic speech recognition.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper claims that standard joint speech-text training fails to fully use textual knowledge from language models once large amounts of supervised speech data are available. It introduces Joint Speech-Text Interleaved Pretraining, which builds word-level and segment-level interleaved sequences from aligned speech-text pairs for models that accept continuous speech inputs. On 38k hours of ASR data this produces steady gains in entity accuracy over both speech-only and non-interleaved joint baselines. The same approach matches the performance of synthetic pairs when real domain transcription text is used instead, and zero-shot question-answering tests indicate the interleaving narrows the input gap between speech and text while retaining the language model's generation behavior.

Core claim

Joint Speech-Text Interleaved Pretraining (JSTIP) constructs word-level and segment-level interleaved speech-text sequences within aligned pairs for speech-LLM architectures that accept continuous inputs. Experiments on 38k hours of ASR data show consistent entity accuracy improvement compared to ASR-only and joint speech-text training baselines. JSTIP achieves on-par entity recognition performance using domain transcription text compared to synthetic speech-text pairs. The zero-shot speech question answering behaviors further suggest that interleaving reduces the speech-text modality gap and preserves the LLM generative prior, which is likely the reason for the entity improvements on the AS

What carries the argument

Joint Speech-Text Interleaved Pretraining (JSTIP), which builds word-level and segment-level interleaved sequences from aligned speech-text pairs to train continuous-input speech-LLM models for ASR.

If this is right

  • Consistent entity accuracy gains appear on 38k hours of ASR data relative to ASR-only and standard joint baselines.
  • Domain adaptation reaches equivalent entity recognition when real transcription text replaces synthetic speech-text pairs.
  • Medical entity recognition becomes competitive with open-source ASR and Speech-LLM systems.
  • Zero-shot speech question answering indicates reduced modality gap and retained generative prior.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same interleaving pattern may reduce the need to synthesize speech for additional text data during domain adaptation.
  • Preservation of the generative prior could support stronger zero-shot or few-shot performance on other speech understanding tasks.
  • Interleaving might extend to other modality pairs where one side already carries extensive pretraining.

Load-bearing premise

The accuracy gains come from the interleaving mechanism itself rather than from differences in total compute, data ordering, or other training details.

What would settle it

A controlled run that matches total training steps, data volume, and ordering exactly between an interleaved version and a non-interleaved joint-training version, then checks whether the entity accuracy gap on the same test sets disappears.

Figures

Figures reproduced from arXiv: 2607.01733 by Ali Zare, Bo Ren, Jinyu Li, Junkun Chen, Keqi Deng, Liliang Ren, Ruchao Fan, Rui Zhao, Xiaoyang Chen, Yan Huang, Yelong Shen, Yiming Wang, Yuxuan Hu.

Figure 1
Figure 1. Figure 1: Frameworks for conventional Speech-LLM integration (left) and joint speech-text interleaved pretraining (right). [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Segment length distribution of the interleaved sequence [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
read the original abstract

Speech-LLM integration has shown promising results by leveraging extensive textual pretraining, yet its specific benefits for automatic speech recognition (ASR) remain unclear. We observe that as supervised ASR training data increases, the contribution of LLM priors becomes less evident, and simple speech-text joint training under-utilizes textual knowledge. We therefore propose Joint Speech-Text Interleaved Pretraining (JSTIP), an ASR-oriented pretraining strategy that constructs word-level and segment-level interleaved speech-text sequences within aligned pairs for speech-LLM architectures that accept continuous inputs. Experiments on 38k hours of ASR data show consistent entity accuracy improvement compared to ASR-only and joint speech-text training baselines. JSTIP achieves on-par entity recognition performance using domain transcription text compared to synthetic speech-text pairs, simplifying domain adaptation. Benefiting from textual pretraining and domain text data, JSTIP is competitive with open-source ASR and Speech-LLM systems in medical entity recognition. The zero-shot speech question answering behaviors further suggest that interleaving reduces the speech-text modality gap and preserves the LLM generative prior, which is likely the reason for the entity improvements on the ASR task.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 1 minor

Summary. The manuscript presents Joint Speech-Text Interleaved Pretraining (JSTIP), a strategy for speech-LLM architectures that interleaves word-level and segment-level speech-text sequences within aligned pairs. Experiments on 38k hours of ASR data are reported to show consistent entity accuracy improvements over ASR-only and joint speech-text baselines. JSTIP is said to achieve on-par entity recognition with domain transcription text compared to synthetic pairs, be competitive in medical entity recognition, and zero-shot QA behaviors suggest reduced modality gap and preserved LLM prior.

Significance. Should the gains prove attributable specifically to the interleaving approach after appropriate controls, the method could provide an effective way to leverage extensive textual pretraining for ASR without relying on synthetic data, thereby simplifying domain adaptation. The indication that interleaving helps maintain the LLM's generative capabilities while bridging modalities may have implications for multi-modal speech applications.

major comments (1)
  1. [Experimental section] The attribution of entity accuracy gains to the interleaving mechanism (abstract) lacks support from a controlled experiment that fixes total tokens, data ordering, and sequence length while varying only the interleaving construction. Current baselines do not isolate this factor, raising the possibility that gains arise from other unstated training details.
minor comments (1)
  1. The abstract would be strengthened by including at least one quantitative result (e.g., absolute accuracy numbers or relative improvement) to convey the scale of the reported gains.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address the major comment below and will revise the paper accordingly to strengthen the experimental claims.

read point-by-point responses

  1. Referee: [Experimental section] The attribution of entity accuracy gains to the interleaving mechanism (abstract) lacks support from a controlled experiment that fixes total tokens, data ordering, and sequence length while varying only the interleaving construction. Current baselines do not isolate this factor, raising the possibility that gains arise from other unstated training details.

    Authors: We agree that a fully isolated ablation fixing total tokens, data ordering, and sequence length while varying solely the interleaving construction would provide stronger evidence. Our joint speech-text baseline uses identical ASR training data (38k hours), the same total token volume, and comparable training hyperparameters and shuffling procedures, with the primary difference being sequence construction (concatenated pairs versus word- and segment-level interleaving within aligned pairs). However, interleaving inherently affects per-sequence token distribution and effective length, so sequence length was not strictly fixed across conditions. We will revise the experimental section to explicitly document token counts, ordering, and length statistics for all conditions and add a clarifying discussion of these controls. If compute resources permit, we will also include a limited additional ablation that enforces stricter matching on sequence length. revision: partial

Circularity Check

0 steps flagged

No circularity; empirical comparisons independent of inputs

full rationale

The paper is an empirical study that proposes the JSTIP training strategy and reports measured entity accuracy improvements on 38k hours of ASR data against baselines. No equations, derivations, or parameter-fitting steps are described that reduce any reported result to a self-referential definition or fitted input. Central claims rest on experimental outcomes rather than theoretical reductions, self-citation chains, or renamed known results. The work is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the empirical observation that interleaving improves entity accuracy; the abstract supplies no free parameters, no new axioms beyond standard supervised learning assumptions, and no invented entities.

axioms (1)
  • domain assumption Aligned speech-text pairs exist and can be segmented at word and segment levels without introducing alignment errors that would dominate the training signal.
    Invoked when the method constructs interleaved sequences from aligned pairs; if alignment quality is poor the interleaving benefit disappears.

pith-pipeline@v0.9.1-grok · 5773 in / 1410 out tokens · 19815 ms · 2026-07-03T15:15:39.304077+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

45 extracted references · 28 canonical work pages · 13 internal anchors

  1. [1]

    GPT-4 Technical Report

    J. Achiam, S. Adler, S. Agarwal, L. Ahmad, I. Akkaya, F. L. Aleman, D. Almeida, J. Altenschmidt, S. Altman, S. Anadkatet al., “Gpt-4 technical report,”arXiv preprint arXiv:2303.08774, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  2. [2]

    Llama 2: Open Foundation and Fine-Tuned Chat Models

    H. Touvron, L. Martin, K. Stone, P. Albert, A. Almahairi, Y . Babaei, N. Bashlykov, S. Batra, P. Bhargava, S. Bhosaleet al., “Llama 2: Open foundation and fine-tuned chat models,”arXiv preprint arXiv:2307.09288, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  3. [3]

    DeepSeek-V3 Technical Report

    A. Liu, B. Feng, B. Xue, B. Wang, B. Wu, C. Lu, C. Zhao, C. Deng, C. Zhang, C. Ruanet al., “Deepseek-v3 technical report,”arXiv preprint arXiv:2412.19437, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  4. [4]

    Gemma 3 Technical Report

    G. Team, A. Kamath, J. Ferret, S. Pathak, N. Vieillard, R. Merhej, S. Perrin, T. Matejovicova, A. Ram ´e, M. Rivi `ereet al., “Gemma 3 technical report,”arXiv preprint arXiv:2503.19786, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  5. [5]

    Qwen3 Technical Report

    A. Yang, A. Li, B. Yang, B. Zhang, B. Hui, B. Zheng, B. Yu, C. Gao, C. Huang, C. Lvet al., “Qwen3 technical report,”arXiv preprint arXiv:2505.09388, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  6. [6]

    On decoder-only architecture for speech-to-text and large language model integration,

    J. Wu, Y . Gaur, Z. Chen, L. Zhou, Y . Zhu, T. Wang, J. Li, S. Liu, B. Ren, L. Liuet al., “On decoder-only architecture for speech-to-text and large language model integration,” inProc. ASRU, 2023

    2023

  7. [7]

    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

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  8. [8]

    SALMONN: Towards generic hearing abilities for large language models,

    C. Tang, W. Yu, G. Sun, X. Chen, T. Tan, W. Li, L. Lu, Z. MA, and C. Zhang, “SALMONN: Towards generic hearing abilities for large language models,” inProc. ICLR, 2024

    2024

  9. [9]

    Phi-4-Mini Technical Report: Compact yet Powerful Multimodal Language Models via Mixture-of-LoRAs

    A. Abouelenin, A. Ashfaq, A. Atkinson, H. Awadalla, N. Bach, J. Bao, A. Benhaim, M. Cai, V . Chaudhary, C. Chenet al., “Phi-4-mini technical report: Compact yet powerful multimodal language models via mixture- of-LoRAs,”arXiv preprint arXiv:2503.01743, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  10. [10]

    Kimi-Audio Technical Report

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

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  11. [11]

    Step-Audio 2 Technical Report

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

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  12. [12]

    Qwen3-Omni Technical Report

    J. Xu, Z. Guo, H. Hu, Y . Chu, X. Wang, J. He, Y . Wang, X. Shi, T. He, X. Zhuet al., “Qwen3-omni technical report,”arXiv preprint arXiv:2509.17765, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  13. [13]

    SLM-S2ST: A multimodal language model for direct speech- to-speech translation,

    Y . Hu, H. Wu, R. Fan, X. Wang, H. Lu, Y . Qian, and J. Li, “SLM-S2ST: A multimodal language model for direct speech- to-speech translation,” inProc. ASRU, 2025. [Online]. Available: https://arxiv.org/abs/2506.04392

    work page arXiv 2025

  14. [14]

    Fun-audio-chat technical report,

    T. F. Team, Q. Chen, L. Cheng, C. Deng, X. Li, J. Liu, C.-H. Tan, W. Wang, J. Xu, J. Yeet al., “Fun-audio-chat technical report,”arXiv preprint arXiv:2512.20156, 2025

    work page arXiv 2025

  15. [15]

    FireRedASR: Open-source industrial-grade mandarin speech recognition models from encoder- decoder to LLM integration,

    K.-T. Xu, F.-L. Xie, X. Tang, and Y . Hu, “FireRedASR: Open-source industrial-grade mandarin speech recognition models from encoder- decoder to LLM integration,”arXiv preprint arXiv:2501.14350, 2025

    work page arXiv 2025

  16. [16]

    Seed-ASR: Understanding diverse speech and contexts with LLM-based speech recognition,

    Y . Bai, J. Chen, J. Chen, W. Chen, Z. Chen, C. Ding, L. Dong, Q. Dong, Y . Du, K. Gaoet al., “Seed-ASR: Understanding diverse speech and contexts with LLM-based speech recognition,”arXiv preprint arXiv:2407.04675, 2024

    work page arXiv 2024

  17. [17]

    Fun-ASR technical report,

    K. An, Y . Chen, C. Deng, C. Gao, Z. Gao, B. Gong, X. Li, Y . Li, X. Lv, Y . Jiet al., “Fun-ASR technical report,”arXiv preprint arXiv:2509.12508, 2025

    work page arXiv 2025

  18. [18]

    Index-ASR technical report,

    Z. Song, L. Wang, W. Deng, Z. Yang, Y . Wu, and B. Xia, “Index-ASR technical report,”arXiv preprint arXiv:2601.00890, 2025

    work page arXiv 2025

  19. [19]

    Speech recognition meets large language model: Benchmarking, models, and exploration,

    Z. Ma, G. Yanget al., “Speech recognition meets large language model: Benchmarking, models, and exploration,” inProc. AAAI, 2025

    2025

  20. [20]

    Efficient scaling for LLM- based ASR,

    B. Mu, Y . Shao, K. Wei, D. Yu, and L. Xie, “Efficient scaling for LLM- based ASR,”arXiv preprint arXiv:2508.04096, 2025

    work page arXiv 2025

  21. [21]

    Qwen3-ASR Technical Report

    X. Shi, X. Wang, Z. Guo, Y . Wang, P. Zhang, X. Zhang, Z. Guo, H. Hao, Y . Xi, B. Yanget al., “Qwen3-asr technical report,”arXiv preprint arXiv:2601.21337, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  22. [22]

    Transducer-Llama: Integrating LLMs into streamable transducer-based speech recognition,

    K. Deng, J. Guoet al., “Transducer-Llama: Integrating LLMs into streamable transducer-based speech recognition,” inProc. ICASSP, 2025

    2025

  23. [23]

    Granite-speech: open-source speech-aware LLMs with strong English ASR capabilities,

    G. Saon, A. Dekel, A. Brooks, T. Nagano, A. Daniels, A. Satt, A. Mittal, B. Kingsbury, D. Haws, E. Moraiset al., “Granite-speech: Open-source speech-aware llms with strong english asr capabilities,”arXiv preprint arXiv:2505.08699, 2025

    work page arXiv 2025

  24. [24]

    Train short, infer long: Speech-llm enables zero-shot streamable joint asr and diarization on long audio,

    M. Shi, X. Xiao, R. Fan, S. Ling, and J. Li, “Train short, infer long: Speech-llm enables zero-shot streamable joint asr and diarization on long audio,”arXiv preprint arXiv:2511.16046, 2025

    work page arXiv 2025

  25. [25]

    Rlbr: Reinforcement learning with biasing rewards for contextual speech large language models,

    B. Ren, R. Fan, Y . Shen, W. Chen, and J. Li, “Rlbr: Reinforcement learning with biasing rewards for contextual speech large language models,” inProc. ICASSP, 2026. [Online]. Available: https://arxiv.org/abs/2601.13409

    work page arXiv 2026

  26. [26]

    Wav2Prompt: End-to-end speech prompt learning and task-based fine-tuning for text-based LLMs,

    K. Deng, G. Sun, and P. C. Woodland, “Wav2Prompt: End-to-end speech prompt learning and task-based fine-tuning for text-based LLMs,” in Proc. NAACL (V olume 1: Long Papers), 2025

    2025

  27. [27]

    Alignformer: Modality matching can achieve better zero-shot instruction-following speech-LLM,

    R. Fan, B. Ren, Y . Hu, R. Zhao, S. Liu, and J. Li, “Alignformer: Modality matching can achieve better zero-shot instruction-following speech-LLM,”IEEE Journal of Selected Topics in Signal Processing, 2025

    2025

  28. [28]

    Qwen2.5-Omni Technical Report

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

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  29. [29]

    V oxtral,

    A. H. Liu, A. Ehrenberg, A. Lo, C. Denoix, C. Barreau, G. Lample, J.-M. Delignon, K. R. Chandu, P. von Platen, P. R. Muddireddyet al., “V oxtral,”arXiv preprint arXiv:2507.13264, 2025

    work page arXiv 2025

  30. [30]

    SpiRit- LM: Interleaved spoken and written language model,

    T. A. Nguyen, B. Muller, B. Yu, M. R. Costa-jussa, M. Elbayad, S. Popuri, C. Ropers, P.-A. Duquenne, R. Algayres, R. Mavlyutov, I. Gat, M. Williamson, G. Synnaeve, J. Pino, B. Sagot, and E. Dupoux, “SpiRit- LM: Interleaved spoken and written language model,”Transactions of the Association for Computational Linguistics, vol. 13, pp. 30–52,

  31. [31]

    Available: https://aclanthology.org/2025.tacl-1.2/

    [Online]. Available: https://aclanthology.org/2025.tacl-1.2/

    2025

  32. [32]

    Enhancing generalization of speech large language models with multi-task behavior imitation and speech-text interleaving,

    J. Xie, X. Li, H. Wang, Y . Yu, Y . Xiang, X. Wu, and Z. Wu, “Enhancing generalization of speech large language models with multi-task behavior imitation and speech-text interleaving,” inProc. Interspeech, 2025. [Online]. Available: https://arxiv.org/abs/2505.18644

    work page arXiv 2025

  33. [33]

    Injecting text in self-supervised speech pretraining,

    Z. Chen, Y . Zhang, A. Rosenberg, B. Ramabhadran, G. Wang, and P. Moreno, “Injecting text in self-supervised speech pretraining,” inProc. ASRU, 2021

    2021

  34. [34]

    SpeechLM: Enhanced speech pre-training with unpaired textual data,

    Z. Zhang, S. Chen, L. Zhou, Y . Wu, S. Ren, S. Liu, Z. Yao, X. Gong, L. Dai, J. Liet al., “SpeechLM: Enhanced speech pre-training with unpaired textual data,”arXiv preprint arXiv:2209.15329, 2022

    work page arXiv 2022

  35. [35]

    SpeechT5: Unified-modal encoder- decoder pre-training for spoken language processing,

    J. Ao, R. Wang, L. Zhou, S. Liu, S. Ren, Y . Wu, T. Ko, Q. Li, Y . Zhang, Z. Wei, Y . Qian, J. Li, and F. Wei, “SpeechT5: Unified-modal encoder- decoder pre-training for spoken language processing,” inProc. ACL, 2022

    2022

  36. [36]

    SLAM: A unified encoder for speech and language modeling via speech-text joint pre-training,

    A. Bapna, Y .-a. Chung, N. Wu, A. Gulati, Y . Jia, J. H. Clark, M. Johnson, J. Riesa, A. Conneau, and Y . Zhang, “SLAM: A unified encoder for speech and language modeling via speech-text joint pre-training,”ArXiv, vol. abs/2110.10329, 2021

    work page arXiv 2021

  37. [37]

    JOIST: A joint speech and text streaming model for ASR,

    T. N. Sainath, R. Prabhavalkar, A. Bapna, Y . Zhang, Z. Huo, Z. Chen, B. Li, W. Wang, and T. Strohman, “JOIST: A joint speech and text streaming model for ASR,” inProc. SLT, 2023

    2023

  38. [38]

    Joint unsupervised and supervised training for multilingual ASR,

    J. Bai, B. Li, Y . Zhang, A. Bapna, N. Siddhartha, K. C. Sim, and T. N. Sainath, “Joint unsupervised and supervised training for multilingual ASR,” inProc. ICASSP, 2022

    2022

  39. [39]

    FastInject: Injecting unpaired text data into CTC-based ASR training,

    K. Deng and P. C. Woodland, “FastInject: Injecting unpaired text data into CTC-based ASR training,” inProc. ICASSP, 2024

    2024

  40. [40]

    Multitask training with text data for end-to-end speech recognition,

    P. Wang, T. N. Sainath, and R. J. Weiss, “Multitask training with text data for end-to-end speech recognition,” inProc. Interspeech, 2021

    2021

  41. [41]

    An attention-based joint acoustic and text on-device end-to-end model,

    T. N. Sainath, R. Pang, R. J. Weiss, Y . He, C.-c. Chiu, and T. Strohman, “An attention-based joint acoustic and text on-device end-to-end model,” inProc. ICASSP, 2020

    2020

  42. [42]

    Maestro: Matched speech text representations through modality matching,

    Z. Chen, Y . Zhang, A. Rosenberg, B. Ramabhadran, P. J. Moreno, A. Bapna, and H. Zen, “Maestro: Matched speech text representations through modality matching,” inProc. Interspeech, 2022

    2022

  43. [43]

    Improving joint speech-text repre- sentations without alignment,

    C. Peyser, Z. Meng, R. Prabhavalkar, A. Rosenberg, T. Sainath, M. Picheny, K. Cho, and K. Hu, “Improving joint speech-text repre- sentations without alignment,” inProc. Interspeech, 2023

    2023

  44. [44]

    Measuring Massive Multitask Language Understanding

    D. Hendrycks, C. Burns, S. Basart, A. Zou, M. Mazeika, D. Song, and J. Steinhardt, “Measuring massive multitask language understanding,” arXiv preprint arXiv:2009.03300, 2020

    work page internal anchor Pith review Pith/arXiv arXiv 2009

  45. [45]

    Ultraeval-audio: A unified framework for comprehensive evaluation of audio foundation models,

    Q. Shi, J. Zhou, B. Lin, J. Cui, G. Zeng, Y . Zhou, Z. Wang, X. Liu, Z. Luo, Y . Wanget al., “Ultraeval-audio: A unified framework for comprehensive evaluation of audio foundation models,”arXiv preprint arXiv:2601.01373, 2026

    work page arXiv 2026