Pith. sign in

REVIEW 4 major objections 6 minor 39 references

Interleaving periodic time markers with continuous audio tokens lets an audio LLM ground answers in time across up to 120 minutes of input.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-14 12:06 UTC pith:ZREWQPRX

load-bearing objection Practical open systems paper: sparse inter-timings plus duration-mixture training make hour-scale timestamped audio QA work, with clean ablations and a real release; synthetic-transcript supervision is the main caveat, not a collapse. the 4 major comments →

arxiv 2607.10387 v1 pith:ZREWQPRX submitted 2026-07-11 eess.AS cs.CL

GigaChat Audio: Time-aware Large Audio Language Model

Aleksandr Kutsakov , Mariia Sadovina , Georgii Gospodinov , Alexandr Maximenko , Oleg Kutuzov , Pavel Bogomolov , Fyodor Minkin This is my paper
classification eess.AS cs.CL
keywords large audio language modelsaudio question answeringlong-form speech processingtemporal groundinginter-timing markerstime-aware audio LLMsynthetic supervision
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Long audio such as meetings and podcasts is only useful for interactive question-answering if the system can say not only what happened but when. This paper shows that ordinary audio language models fail at that task once recordings stretch beyond a few minutes: they invent unparseable timestamps or lose the timeline entirely. The authors fix the problem by inserting explicit time markers at regular intervals into the stream of continuous audio tokens, then training the model on large-scale synthetic question-answer pairs that force it to use those markers. The resulting model keeps high temporal accuracy from short clips out to two-hour inputs, and it can also produce time-anchored fragment descriptions and multi-part summaries. Ablations establish that both the markers themselves and a mixture of training lengths are necessary; without either ingredient, long-form grounding collapses.

Core claim

Periodic temporal anchors interleaved with continuous audio tokens are essential for stable long-form grounding. With markers every 60 seconds the model reaches 53.8 mIoU on 20–40-minute temporal grounding (rising to 65.2 with 7-second markers), while removing the anchors drops the same metric to 14.2 and sharply degrades timed descriptions and summaries. Training on a mixture of short and long recordings is likewise required for length generalization up to 120 minutes; single-duration regimes do not transfer.

What carries the argument

Inter-timings: periodic time markers (plain-text hh:mm:ss or dedicated tokens) inserted every N seconds into the continuous audio-token stream, acting as temporal anchors that the model can read and emit.

Load-bearing premise

Synthetic question-answer pairs generated by a text-only language model from automatically aligned transcripts, after slicing and a consistency verifier, are accurate enough for an audio model to learn genuine temporal localization rather than just matching transcript patterns.

What would settle it

Train an identical architecture without any inter-timing markers on the same synthetic data and measure long-form (20–40 min and 120 min) temporal-grounding mIoU; if it matches or exceeds the marker-equipped model, the claim that anchors are essential is false.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper introduces GigaChat Audio, a 10B-A1.8B MoE audio-conditioned LLM that accepts up to 120 minutes of input and produces answers, fragment descriptions, and summaries with explicit timestamps. The core design interleaves continuous audio embeddings (160 ms frames from an encoder–subsampler–projector stack) with periodic inter-timing markers (plain-text hh:mm:ss or special tokens). Temporal supervision is obtained at scale via a cascaded synthetic pipeline: WhisperX-aligned transcripts, ~10-minute slice-wise (Q,A) generation with a text-only LLM, global verification against the full transcript, and multi-sample aggregation for evaluation. The authors report competitive short-clip results and substantially stronger long-form temporal grounding than open and proprietary baselines, with ablations on marker frequency, format, special tokens, and duration-mixture training (Tables 1–4, Figure 3). Model weights and a 10k+ hour temporal dataset are released.

Significance. Temporal grounding for long recordings is a genuine bottleneck for current audio LLMs; the paper isolates a simple, practical mechanism (periodic inter-timings) and shows, via controlled ablations, that removing anchors collapses 20–40 min mIoU from 53.8 to 14.2 and that duration-mixture training is required for length generalization up to 120 min. The open release of weights and a large temporal dataset is a concrete contribution that can accelerate follow-on work. The engineering choices (marker frequency vs. token cost, plain-text vs. special tokens, silence and language filtering) are carefully quantified and useful to practitioners. If the claimed audio-native localization holds under stronger controls, the work would be a solid reference system for time-aware long-form audio understanding.

major comments (4)
  1. Section 3.5 / Figure 2 and Table 1: The central claim that inter-timings enable genuine long-form audio grounding rests almost entirely on synthetic (Q,A) pairs generated by a text-only LLM from WhisperX transcripts of the same recordings used at training time. Slice-wise generation plus a global verifier only enforce transcript-internal consistency; they do not force the model to use the continuous audio stream rather than acoustic co-occurrence patterns that align with transcript phrases and the interleaved markers. AMI questions are themselves auto-generated from transcripts, and the long-form TGr / description / summary sets are synthetic. Please add at least one of: (i) a human-annotated long-form temporal-QA subset, (ii) an audio-ablation or transcript-only control that quantifies residual performance when audio is corrupted or replaced, or (iii) a clear analysis showing that local
  2. Section 3.6 and Table 1 (Descriptions, Summ columns): Fragment descriptions and timed summaries are scored solely by an LLM-as-a-judge protocol whose correlation with human judgments is not reported for this domain. The paper itself notes that reference-based metrics correlate imperfectly with humans, yet provides no human calibration, inter-annotator agreement, or judge-prompt sensitivity study. Because these scores are used to claim superiority on open-ended timed generation, a modest human validation study (or release of judge prompts + agreement numbers) is needed before the open-ended results can be treated as load-bearing.
  3. Tables 1–4 and Figure 3: All headline mIoU / MAE / judge scores are point estimates with no error bars, multi-seed variance, or statistical tests. Given that SFT mixtures, synthetic generation temperature, and verifier thresholds are free parameters, single-run differences (e.g., 53.8 vs. 14.2 mIoU, or 60 s vs. 7 s anchors) cannot be assessed for robustness. Please report at least multi-seed standard deviations or bootstrap intervals on the key ablations that support the necessity of inter-timings and duration mixtures.
  4. Section 4.1 / Table 1: Long-form comparison against strong systems is incomplete. TimeAudio is reported only on short clips (authors note degradation beyond ~2 min); Qwen3-Omni and Gemini 3 Flash are included, but there is no systematic comparison against long-context ASR+LLM pipelines that already emit word-level timestamps (e.g., WhisperX + LLM over the transcript). Such a pipeline is a natural and strong baseline for timestamped QA and would clarify how much of the gain is attributable to end-to-end audio conditioning versus access to aligned time markers.
minor comments (6)
  1. Figure 1 caption and Section 3.1: Clarify whether inter-timings are inserted only in the input stream, only in the target, or both, and whether the final end-of-audio marker is always present at inference.
  2. Table 1 header: “Tm(↑) AES(↑) Rd(↓)” for summarization is dense; expand once in the caption or define AES explicitly as (Acc + 1 − Err + Style)/3 next to the table.
  3. Section 3.4: The silence-ratio and language-ID thresholds (p(English)>0.7) are free parameters that affect the 14k-hour pool; a short sensitivity note or appendix table would help reproducibility.
  4. Section 4.3 / Table 2: “extra tokens” require much higher TG data ratios to match plain-text; the discussion could briefly note whether this is an initialization or capacity issue.
  5. Typos / polish: “V oxtral”, “VibeV oice”, “front-loading bias” hyphenation is inconsistent; “inter-timings” is introduced without a formal definition before Figure 1. A short glossary or first-use definition would help.
  6. References: Several 2025–2026 arXiv preprints are cited; ensure final versions or DOIs are updated at camera-ready if available.

Circularity Check

0 steps flagged

No circular derivation; empirical ablations and external benchmarks stand independently of inputs.

full rationale

This is an empirical systems paper whose central claims (necessity of periodic inter-timings for long-form mIoU, necessity of duration-mixture training for length generalization) are established by controlled ablations (Table 1 w/ vs w/o inter-timings; Figure 3 duration-bucket matrix) and comparisons against external short- and long-form benchmarks (AudioGrounding, AMI, DAQA, Gemini, Qwen3-Omni). The synthetic supervision pipeline (Section 3.5, Figure 2) generates training targets from WhisperX transcripts via an external LLM + verifier, but the reported metrics are measured on held-out audio-level splits and external sets; nothing reduces mIoU or MAE to a fitted constant or definitional identity. Self-citations ([19] base MoE checkpoint, [22] HuBERT-style encoder pre-training) supply fixed components that are not the results being claimed. No equation, uniqueness theorem, or ansatz is imported to force the temporal-anchor or mixture conclusions. The paper therefore contains no circular steps under the defined patterns.

Axiom & Free-Parameter Ledger

5 free parameters · 4 axioms · 2 invented entities

The central claims rest on standard ML engineering choices (token interleaving, SFT mixture ratios, ASR-derived timestamps) and on the transferability of synthetic text supervision to continuous audio, not on new physical entities or free constants fitted to the target metric. Free parameters are ordinary hyperparameters that the paper itself ablates; the main domain assumptions concern the fidelity of WhisperX and of LLM-generated labels.

free parameters (5)
  • inter-timing frequency = 60 s (main); 7-240 s range
    Default 60 s (ablated from 7 s to 240 s); directly trades mIoU against token overhead and is chosen for the main reported model.
  • temporal-grounding data ratio in SFT mixture = 2.4-50 %
    Ablated from 2.4 % to 50 %; required for special timing tokens to match plain-text performance.
  • encoder and decoder learning rates = lr_enc=1e-4, lr_dec=5e-6
    Fixed during audio SFT; not derived from first principles.
  • silence-ratio filter and language-ID threshold = p(English)>0.7 plus silence thresholds
    Used to curate YODAS2 down to 14 k hours; values affect the training distribution.
  • audio frame rate and encoder chunk size = 160 ms frames, 8 s chunks
    Architectural choices that set the density of continuous tokens relative to time markers.
axioms (4)
  • domain assumption WhisperX word-level timestamps are accurate enough to serve both as synthetic supervision sources and as reference intervals for evaluation.
    Invoked in Section 3.4; every generated label and several benchmarks rest on this alignment.
  • domain assumption Text-only LLM generation over timestamped transcripts, followed by a global verifier, yields supervision that transfers to an audio-conditioned model.
    Core of the cascaded pipeline in Section 3.5 and Figure 2.
  • domain assumption LLM-as-a-judge scores on factual accuracy, timing structure, and style correlate adequately with human judgments for timed descriptions and summaries.
    Section 3.6 evaluation protocol; no paper-specific human correlation is reported.
  • ad hoc to paper Continuous audio embeddings can be interleaved with discrete textual or special timing tokens without further architectural changes beyond supervised fine-tuning.
    Design premise of Section 3.1 and Figure 1.
invented entities (2)
  • inter-timings (periodic time markers interleaved with continuous audio tokens) independent evidence
    purpose: Supply explicit absolute-time anchors so the language model can emit parseable timestamps for long inputs.
    Central design element; ablation removing them collapses long-form mIoU. Similar separators appear in video LLMs, so independent evidence exists outside this paper.
  • cascaded synthetic pipeline with transcript slicing and global verifier no independent evidence
    purpose: Generate scalable temporal QA and timed-summary supervision for hour-scale audio without manual annotation.
    Introduced in Figure 2 / Section 3.5; quality is assessed only by the pipeline's own verifier and multi-sample filter, not by external human audit.

pith-pipeline@v1.1.0-grok45 · 14071 in / 3194 out tokens · 58019 ms · 2026-07-14T12:06:18.624846+00:00 · methodology

0 comments
read the original abstract

Temporal grounding in long recordings remains challenging for audio-conditioned LLMs. We present a time-aware audio LLM that answers questions with explicit timestamps over up to 120 minutes of input. Our approach interleaves periodic time markers with continuous audio tokens using large-scale synthetic supervision from a cascaded pipeline. Our model achieves strong temporal-grounding accuracy on short and long benchmarks and supports time-anchored fragment descriptions and summaries. Extensive ablations examine how time representation, marker frequency, tokenization, and duration-mixture design affect accuracy and computational cost. We release model weights and datasets to support further research on time-aware audio understanding, available at https://huggingface.co/ai-sage/GigaChat3.1-Audio-10B-A1.8B.

Figures

Figures reproduced from arXiv: 2607.10387 by Aleksandr Kutsakov, Alexandr Maximenko, Fyodor Minkin, Georgii Gospodinov, Mariia Sadovina, Oleg Kutuzov, Pavel Bogomolov.

Figure 1
Figure 1. Figure 1: Architecture of a time-aware audio LLM. Token streams are interleaved: text tokens, continuous audio tokens, timing tokens (either text-form hh:mm:ss or special ones). We present GigaChat Audio, a time-aware audio￾conditioned LLM that supports up to 120 minutes of input and produces time-anchored answers, fragment descriptions, and summaries with explicit timestamps. Our approach interleaves continuous aud… view at source ↗
Figure 2
Figure 2. Figure 2: Synthetic supervision pipeline for temporal grounding. Audio is transcribed and aligned to obtain a timestamped transcript. For QA-style data, we sample a ∼10-minute transcript fragment and generate (Q, A) pairs; a verifier checks each pair against the full timestamped transcript to enforce global consistency. We split train/eval at the audio level (all questions from the same recording are in the same spl… view at source ↗
Figure 3
Figure 3. Figure 3: Length extrapolation up to 120 min for temporal grounding in mIoU (↑). Rows are training duration windows, columns are eval audio lengths. Shading is column-normalized, indicating relative quality within each eval duration. longer than two minutes, frequently producing UNK and non￾parseable timestamps. We therefore report its results only when the benchmark duration matches its original setup [PITH_FULL_I… view at source ↗

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

39 extracted references · 12 linked inside Pith

  1. [1]

    Introduction Long recordings such as meetings, podcasts, lectures, and call- center logs are increasingly consumed through interactive inter- faces: users ask questions, request summaries, and navigate to specific evidence. In these settings,temporal groundingis not a cosmetic feature but a verifiability primitive: a system should answer not onlywhathappe...

  2. [2]

    Earlier open AudioLLMs [5, 6] demonstrate broad au- dio understanding and dialogue capabilities

    Related Work Modern multimodal systems can follow instructions from speech and audio inputs, including GPT-4o [1], Gemini [2], and open-weight omni models such as Qwen3-Omni [4] and V ox- tral [3]. Earlier open AudioLLMs [5, 6] demonstrate broad au- dio understanding and dialogue capabilities. While these mod- arXiv:2607.10387v1 [eess.AS] 11 Jul 2026 Figu...

    Pith/arXiv arXiv 2026

  3. [3]

    audio tokens

    Method 3.1. Model Our model is atime-awareaudio LLM built by attaching an au- dio front-end to a 10B-A1.8B MoE text checkpoint with a 256k- token context [19] and fine-tuning it on audio SFT data with ex- plicit time signals (Figure 1). Training is performed using data parallelism only, without tensor or sequence parallelism. We use anencoder→subsampler→p...

  4. [4]

    We then obtain word-level timestamps and time-aligned transcripts us- ing WhisperX; the same alignment is also used to estimate the silence ratio [9]

    and filter to English with a SpeechBrain V oxLingua107 ECAPA-TDNN language-ID model, keeping only samples with p(English)>0.7(16k hours after filtering) [27, 28]. We then obtain word-level timestamps and time-aligned transcripts us- ing WhisperX; the same alignment is also used to estimate the silence ratio [9]. Finally, we filter recordings by silence-ra...

  5. [5]

    when was this phrase spoken?

    Experiments 4.1. Main comparison across tasks and datasets Table 1 summarizes performance across four benchmarks: (i) AUDIOGROUNDING[10]. Short-clip dataset (980 samples, 7–10 s). Questions refer to acoustic events, and the model predicts a temporal interval. We report mIoU. (ii) AMI MEETINGCORPUS[32]. Meeting recordings (15– 50 min) with word-level times...

  6. [6]

    Our results show that temporal ground- ing in long recordings remains a major bottleneck for exist- ing multimodal models, which often degrade sharply beyond a few minutes of audio

    Conclusion We presented a time-aware Audio LLM that supports up to 120 minutes of input and produces answers and summaries explic- itly anchored in time. Our results show that temporal ground- ing in long recordings remains a major bottleneck for exist- ing multimodal models, which often degrade sharply beyond a few minutes of audio. We demonstrate that p...

  7. [7]

    All outputs were carefully reviewed, edited, and validated by the authors, who take full responsibility for the fi- nal content

    Generative AI Use Disclosure Generative AI tools were used for limited language editing (grammar, clarity, and stylistic refinement) during manuscript preparation. All outputs were carefully reviewed, edited, and validated by the authors, who take full responsibility for the fi- nal content. As described in Section 3.5 and Section 3.6, large language mode...

  8. [8]

    Gpt-4o system card,

    OpenAI, “Gpt-4o system card,” 2024. [Online]. Available: https://arxiv.org/abs/2410.21276

    Pith/arXiv arXiv 2024

  9. [9]

    Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities,

    Gemini Team, Google, “Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities,” 2025. [Online]. Available: https://arxiv.org/abs/2507.06261

    Pith/arXiv arXiv 2025

  10. [10]

    V oxtral,

    A. H. Liuet al., “V oxtral,” 2025. [Online]. Available: https://arxiv.org/abs/2507.13264

    Pith/arXiv arXiv 2025

  11. [11]

    Qwen3-omni technical report,

    J. Xuet al., “Qwen3-omni technical report,” 2025. [Online]. Available: https://arxiv.org/abs/2509.17765

    Pith/arXiv arXiv 2025

  12. [12]

    Qwen2-audio technical report,

    Y . Chu, J. Xu, Q. Yang, H. Wei, X. Wei, Z. Guo, Y . Leng, Y . Lv, J. He, J. Lin, C. Zhou, and J. Zhou, “Qwen2-audio technical report,” 2024. [Online]. Available: https://arxiv.org/abs/2407.10759

    Pith/arXiv arXiv 2024

  13. [13]

    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,” 2024. [Online]. Available: https://arxiv.org/abs/2310.13289

    Pith/arXiv arXiv 2024

  14. [14]

    VIBEVOICE-ASR technical report,

    Z. Penget al., “VIBEVOICE-ASR technical report,” 2026. [Online]. Available: https://arxiv.org/abs/2601.18184

    arXiv 2026

  15. [15]

    Moss transcribe diarize technical report,

    MOSI.AI, “Moss transcribe diarize technical report,” 2026. [Online]. Available: https://arxiv.org/abs/2601.01554

    Pith/arXiv arXiv 2026

  16. [16]

    WhisperX: Time- Accurate Speech Transcription of Long-Form Audio,

    M. Bain, J. Huh, T. Han, and A. Zisserman, “WhisperX: Time- Accurate Speech Transcription of Long-Form Audio,” inInter- speech 2023, 2023, pp. 4489–4493

    2023

  17. [17]

    Text-to-audio grounding: Building correspondence between captions and sound events,

    X. Xu, H. Dinkel, M. Wu, and K. Yu, “Text-to-audio grounding: Building correspondence between captions and sound events,” in ICASSP 2021 - 2021 IEEE International Conference on Acous- tics, Speech and Signal Processing (ICASSP), 2021, pp. 606–610

    2021

  18. [18]

    Multi-domain audio question answering toward acoustic content reasoning in the dcase 2025 challenge,

    C.-H. H. Yang, S. Ghosh, Q. Wang, J. Kim, H. Hong, S. Kumar, G. Zhong, Z. Kong, S. Sakshi, V . Lokegaonkar, O. Nieto, R. Duraiswami, D. Manocha, G. Kim, J. Du, R. Valle, and B. Catanzaro, “Multi-domain audio question answering toward acoustic content reasoning in the dcase 2025 challenge,” 2025. [Online]. Available: https://arxiv.org/abs/2505.07365

    arXiv 2025

  19. [19]

    Listening between the frames: Bridging temporal gaps in large audio-language models,

    H. Wang, Y . Li, S. Ma, H. Liu, and X. Wang, “Listening between the frames: Bridging temporal gaps in large audio-language models,” 2025. [Online]. Available: https://arxiv.org/abs/2511. 11039

    2025

  20. [20]

    Language-based audio moment retrieval,

    H. Munakata, T. Nishimura, S. Nakada, and T. Komatsu, “Language-based audio moment retrieval,” inICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2025, pp. 1–5

    2025

  21. [21]

    CASTELLA: Long audio dataset with captions and temporal boundaries,

    H. Munakata, T. Imamura, T. Nishimura, and T. Komatsu, “CASTELLA: Long audio dataset with captions and temporal boundaries,” 2026. [Online]. Available: https://arxiv.org/abs/ 2511.15131

    arXiv 2026

  22. [22]

    Localizing moments in video with temporal language,

    L. A. Hendricks, O. Wang, E. Shechtman, J. Sivic, T. Darrell, and B. Russell, “Localizing moments in video with temporal language,” inProceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, E. Riloff, D. Chiang, J. Hockenmaier, and J. Tsujii, Eds. Brussels, Belgium: Association for Computational Linguistics, Oct.-Nov. 2018,...

    2018

  23. [23]

    Timemarker: A versatile video-llm for long and short video understanding with superior temporal localization ability,

    S. Chen, X. Lan, Y . Yuan, Z. Jie, and L. Ma, “Timemarker: A versatile video-llm for long and short video understanding with superior temporal localization ability,” 2024. [Online]. Available: https://arxiv.org/abs/2411.18211

    Pith/arXiv arXiv 2024

  24. [24]

    Longcat- flash technical report,

    M. L. Team, Bayan, B. Li, and B. L. et al., “Longcat- flash technical report,” 2025. [Online]. Available: https: //arxiv.org/abs/2509.01322

    arXiv 2025

  25. [25]

    G-eval: NLG evaluation using gpt-4 with better human alignment,

    Y . Liu, D. Iter, Y . Xu, S. Wang, R. Xu, and C. Zhu, “G-eval: NLG evaluation using gpt-4 with better human alignment,” in Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, H. Bouamor, J. Pino, and K. Bali, Eds. Singapore: Association for Computational Linguistics, Dec. 2023, pp. 2511–2522. [Online]. Available: https:...

    2023

  26. [26]

    GigaChat3-10B-A1.8B,

    ai-sage, “GigaChat3-10B-A1.8B,” https://huggingface.co/ ai-sage/GigaChat3-10B-A1.8B, 2025, hugging Face model card; accessed 2026-02-25

    2025

  27. [27]

    FlashAttention-2: Faster attention with better parallelism and work partitioning,

    T. Dao, “FlashAttention-2: Faster attention with better parallelism and work partitioning,” inThe Twelfth International Conference on Learning Representations, 2024. [Online]. Available: https: //openreview.net/forum?id=mZn2Xyh9Ec

    2024

  28. [28]

    Hubert: Self-supervised speech rep- resentation learning by masked prediction of hidden units,

    W.-N. Hsu, B. Bolte, Y .-H. H. Tsai, K. Lakhotia, R. Salakhut- dinov, and A. Mohamed, “Hubert: Self-supervised speech rep- resentation learning by masked prediction of hidden units,” IEEE/ACM Transactions on Audio, Speech, and Language Pro- cessing, vol. 29, pp. 3451–3460, 2021

    2021

  29. [29]

    GigaAM: Efficient Self-Supervised Learner for Speech Recognition,

    A. Kutsakov, A. Maximenko, G. Gospodinov, P. Bogomolov, and F. Minkin, “GigaAM: Efficient Self-Supervised Learner for Speech Recognition,” inInterspeech 2025, 2025, pp. 1213–1217

    2025

  30. [30]

    Outrageously large neural networks: The sparsely-gated mixture-of-experts layer,

    N. Shazeer, A. Mirhoseini, K. Maziarz, A. Davis, Q. Le, G. Hinton, and J. Dean, “Outrageously large neural networks: The sparsely-gated mixture-of-experts layer,” 2017. [Online]. Available: https://arxiv.org/abs/1701.06538

    Pith/arXiv arXiv 2017

  31. [31]

    Clotho: an audio cap- tioning dataset,

    K. Drossos, S. Lipping, and T. Virtanen, “Clotho: an audio cap- tioning dataset,” inICASSP 2020 - 2020 IEEE International Con- ference on Acoustics, Speech and Signal Processing (ICASSP), 2020, pp. 736–740

    2020

  32. [32]

    AudioCaps: Generating captions for audios in the wild,

    C. D. Kim, B. Kim, H. Lee, and G. Kim, “AudioCaps: Generating captions for audios in the wild,” inProceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), J. Burstein, C. Doran, and T. Solorio, Eds. Minneapolis, Minnesota: Association f...

    2019

  33. [33]

    Yodas: Youtube-oriented dataset for audio and speech,

    X. Li, S. Takamichi, T. Saeki, W. Chen, S. Shiota, and S. Watan- abe, “Yodas: Youtube-oriented dataset for audio and speech,” in2023 IEEE Automatic Speech Recognition and Understanding Workshop, ASRU 2023, ser. 2023 IEEE Automatic Speech Recog- nition and Understanding Workshop, ASRU 2023. Institute of Electrical and Electronics Engineers Inc., 2023, publ...

    2023

  34. [34]

    SpeechBrain: A General-Purpose Speech Toolkit,

    T. Parcollet, M. Ravanelli, P. Plantinga, A. Rouhe, S. Cornell, L. Lugosch, C. Subakan, N. Dawalatabad, A. Heba, J. Zhong, J.-C. Chou, S.-L. Yeh, S.-W. Fu, C.-F. Liao, E. Rastorgueva, F. Grondin, W. Aris, H. Na, Y . Gao, R. de Mori, and Y . Bengio, “SpeechBrain: A General-Purpose Speech Toolkit,” Mar. 2022, preprint. [Online]. Available: https://hal.scien...

    2022

  35. [35]

    V oxlingua107: A dataset for spoken language recognition,

    J. Valk and T. Alum ¨ae, “V oxlingua107: A dataset for spoken language recognition,”2021 IEEE Spoken Language Technology Workshop (SLT), pp. 652–658, 2020. [Online]. Available: https://api.semanticscholar.org/CorpusID:227209238

    2021

  36. [36]

    gpt-oss-120b & gpt-oss-20b model card,

    OpenAI, S. Agarwal, L. Ahmad, J. Ai, S. Altman, A. Applebaum et al., “gpt-oss-120b & gpt-oss-20b model card,” 2025. [Online]. Available: https://arxiv.org/abs/2508.10925

    Pith/arXiv arXiv 2025

  37. [37]

    A comparative study of quality evaluation methods for text summarization,

    H. Nguyen, H. Chen, L. Pobbathi, and J. Ding, “A comparative study of quality evaluation methods for text summarization,” ArXiv, vol. abs/2407.00747, 2024. [Online]. Available: https: //api.semanticscholar.org/CorpusID:270869494

    Pith/arXiv arXiv 2024

  38. [38]

    Summeval: Re-evaluating summarization evaluation,

    A. R. Fabbri, W. Kryscinski, B. McCann, R. Socher, and D. R. Radev, “Summeval: Re-evaluating summarization evaluation,”Transactions of the Association for Computational Linguistics, vol. 9, pp. 391–409, 2020. [Online]. Available: https://api.semanticscholar.org/CorpusID:220768873

    2020

  39. [39]

    Unleashing the killer corpus: experiences in creating the multi-everything ami meeting corpus,

    J. Carletta, “Unleashing the killer corpus: experiences in creating the multi-everything ami meeting corpus,”Language Resources and Evaluation, vol. 41, no. 2, pp. 181–190, 2007

    2007