Pith. sign in

REVIEW 3 major objections 4 minor 31 references

DnR-nonverbal: Cinematic Audio Source Separation Dataset Containing Non-Verbal Sounds

T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Adding non-verbal vocal sounds to the speech stem of a cinematic source separation dataset fixes the model's tendency to misroute laughter and screams into the effects stem.

desk verdict Useful new CASS dataset with a real domain-mismatch fix, but the evaluation shares a contamination risk between train and eval and needs a cleaner external check. read the letter →

arxiv 2506.02499 v2 pith:7X5AI4CJ submitted 2025-06-03 cs.SD eess.AS

Takuya Hasumi , Yusuke Fujita This is my paper
classification cs.SDeess.AS
keywords cinematicaudiosourceseparationnon-verbalsoundsspeechstemDnR-nonverbalFSD50KFreeSoundBandItdatasetcuration
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 claims that the reason cinematic audio source separation models push laughter, screams, and other emotionally heightened voices into the effects stem is that existing training datasets only contain reading-style speech. To fix this, the authors build DnR-nonverbal, a dataset that places non-verbal human vocalizations in the speech stem, using clips from FSD50K and FreeSound filtered by rules and a large language model. Training a BandIt model on DnR-v2 plus DnR-nonverbal raises speech SDR from 5.62 to 9.30 dB and effect SDR from 2.54 to 5.23 dB on a synthetic evaluation set. In a listener A/B test on real movie clips, 76.9% preferred the model retrained with the new dataset, so the fix appears to transfer beyond synthetic mixtures. If this holds, the paper's contribution is a dataset that removes a domain mismatch that made CASS models unreliable for expressive dialogue.

What carries the argument

The load-bearing object is the DnR-nonverbal dataset itself: a 60-second-track extension of DnR-v2 in which the speech stem is mixed from reading-style LibriSpeech clips interleaved with non-verbal vocal clips drawn from FSD50K (by AudioSet-ontology tags) and newly crawled from FreeSound, with rule-based and GPT-4o-based filtering to remove non-human sounds, and a mix algorithm (zero-truncated Poisson clip counts, skew-Gaussian silences, LUFS-based loudness sampling) that mirrors DnR-v2's mixing for the music and effects stems. The dataset is what changes the model's behavior; BandIt is used only as a fixed probe to measure the effect.

What would settle it

Manually audit the 471 FreeSound-derived clips (and a sample of the FSD50K clips) in the final DnR-nonverbal training set: if a substantial fraction contain non-human sounds, or if retraining on a hand-verified clean subset fails to reproduce the speech-SDR gain (9.30 dB vs 5.62 dB), the claim that clean non-verbal vocal labels drive the improvement would be falsified.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central claim is that non-verbal vocal sounds (laughter, screaming, whispering, crying, sighing, shouting) belong in the speech stem of a cinematic audio source separation dataset, and that including them there retrains the model to treat expressive voice as speech rather than as an effect. The paper shows that a BandIt model trained only on the conventional DnR-v2 dataset scores 5.62 dB on the speech stem when non-verbal sounds are counted as speech, but 6.52 dB when it is allowed to route those same sounds to the effects stem, which is direct evidence that the model is systematically misallocating them. Adding DnR-nonverbal closes that gap: the same model reaches 9.30 dB on speech while holding non-verbal sounds in the speech stem, and the subjective test on real movie audio indicates the retrained model extracts actors' voices more naturally and consistently. The intended consequence is a practical CASS tool that can handle acted-out dialogue in filmmaking and post-production.

Load-bearing premise

The curation pipeline reliably isolates non-verbal human vocal clips, excluding music, animal sounds, and other noise, so that the speech stem teaches the model the right association and the reported gains are not artifacts of contaminated labels.

Editorial extensions

If this is right

  • Models trained with DnR-nonverbal separate expressive voice (screams, laughter, whispers) into the speech stem instead of dumping it into the effects stem, closing the main failure mode of CASS models on acted-out dialogue.
  • Because the effects stem is no longer contaminated by voice, effect SDR also improves (from 2.54 to 5.23 dB), and the music stem improves slightly as a side effect.
  • The fix transfers to real movie audio: 76.9% of listeners preferred the retrained model's speech extraction as more natural and consistent.
  • The dataset pipeline (tag-based collection plus LLM filtering) can be extended to other non-verbal vocal categories, and the resulting datasets could support query-based source separation and audio captioning, as the paper notes in conclusion.

Reading between the lines

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

  • The same domain-mismatch argument likely applies to other 'non-speech' vocal events in sound separation, e.g., in music source separation where beatboxing or spoken ad-libs may be routed to the wrong stem; DnR-nonverbal provides a template for auditing that.
  • The paper's own admitted failure mode — the retrained model sometimes mistakes an animal's voice for screaming — suggests that purely acoustic training has limits; a vision-modality context (as the authors suggest) or a stricter human-voice classifier would be a natural next test.
  • One unstated consequence is practical: if a CASS model can reliably keep expressive voice in the speech stem, dubbing, ADR, and subtitle workflows could be automated for dialogue that is not read calmly, which is most movie dialogue.
  • The 76.9% preference on real movie clips comes from a single small A/B test (13 raters, 20 clips); scaling that evaluation with more raters and clips, or measuring downstream tasks, would tell whether the subjective gain is robust.
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

3 major / 4 minor

Summary. The paper proposes DnR-nonverbal, a new CASS dataset that extends the DnR-v2 speech stem with non-verbal human vocal sounds (laughter, screaming, whispering, crying, sighing, shouting). Clips are collected from FSD50K and new FreeSound crawls, filtered by rule-based tag checks and an LLM (GPT-4o) that inspects tags and descriptions rather than audio, then mixed into 60-second speech stems together with reading-style speech. The authors train BandIt on DnR-v2 and on DnR-v2 + DnR-nonverbal, report substantially higher speech and effect SDR on a synthetic evaluation set, and run a small A/B test on real movie clips showing 76.9% preference for the retrained model. The central claim is that placing non-verbal sounds in the speech stem fixes the misallocation of these sounds to the effect stem.

Significance. If the central claim holds, DnR-nonverbal is a useful, focused contribution to cinematic audio source separation: it targets a real domain gap that existing datasets (DnR-v2, DnR-v3, and speech-music datasets) ignore, and it provides a public dataset with a clear baseline comparison. The paper deserves credit for releasing the dataset on Zenodo, for including both objective and subjective evaluations on real movie audio, and for transparently acknowledging limitations in Sections 3.3 and 4.4. However, the main empirical conclusion rests on the assumption that the tag- and LLM-filtered clips are actually human non-verbal vocalizations, and the paper itself admits this assumption is imperfect; this makes the clean-label evaluation a load-bearing point that needs strengthening.

major comments (3)
  1. [§3.3, §4.4, Table 2] The central claim is causal: adding non-verbal human vocal sounds to the speech stem teaches the model to extract these sounds as speech rather than effect. The paper admits in §3.3 that 'Despite LLM-based filtering, some clips still contained non-human sounds', and in §4.4 that the model can mistake an animal's voice for screaming. The evaluation set (395 FSD50K clips) is selected by the same tag-based procedure (Human voice descendant tags minus Singing) and is not acoustically verified. Since both training and evaluation pass through the same contaminated pipeline, part of the measured improvement in Table 2 (speech SDR 5.62 to 9.30 dB, effect SDR 2.54 to 5.23 dB) may reflect the model learning a mapping that assigns non-human sounds to the speech stem, rather than genuinely isolating human laughter, screams, and sighs. I request a clean-label validation: manually verify the evaluation clips (or a representative subset), remove or re-label non-human sounds, and recompute Table 2 on the verified subset. The paper should also state how many eval clips were excluded and whether the SDR gains remain after this correction.
  2. [§4.3, Eq. (3)] The SDR metric in Eq. (3) is a simplified energy ratio without the usual signal-to-distortion decomposition or scale-invariant projection of BSSEval-style SDR. The paper reports single-run numbers with no confidence intervals, error bars, or significance tests across the 100 evaluation mixtures. Since the reported gains are large, the conclusions would likely survive a statistical test, but the paper should either report standard SDR with projection or provide significance tests (e.g., paired tests across mixtures, or variance across multiple training runs) to support the claim that the improvement is not due to a few outlier clips.
  3. [§3.3] The LLM-based filtering for FreeSound clips relies on tags and descriptions only; the prompt itself states the LLM is 'determin[ing] the availability by guessing the given tags and description', not by listening to the audio. This means the LLM cannot detect acoustic contamination such as animal sounds, music, or noise that are not reflected in the metadata. Since the paper acknowledges residual contamination, the evaluation set should not be presumed clean without audio-level verification, and the manuscript should explicitly state this limitation in the evaluation section rather than only in the filtering section.
minor comments (4)
  1. [Figure 1 caption] The caption contains a typo: 'Comparision' should be 'Comparison'.
  2. [§4.4, Table 3] The A/B test uses 20 clips and 13 raters; the paper reports only the aggregate preference percentages without per-clip breakdown or inter-rater agreement. Adding a confidence interval or a per-clip best-worst count would strengthen the external support.
  3. [Bibliography] Reference [15] misspells the first author's name as 'Warcharasupat' (should be 'Watcharasupat'), and reference [17] reads 'Proceedings of Proceedings of Interspeech'; these should be corrected.
  4. [§4.2] The sentence 'The model may treat neither reading-speech nor music content as effects' is unclear and should be rephrased; the intended meaning appears to be that the model assigns an unseen sound class to the effect stem by default.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: dataset intervention is validated by held-out and external A/B evaluation.

full rationale

The paper makes a causal-empirical claim: training a CASS model with non-verbal vocal clips placed in the speech stem changes how those sounds are assigned at test time. This is tested directly by comparing the DnR-v2 baseline against DnR-v2 + DnR-nonverbal on a held-out evaluation set and on external movie clips through an A/B test. No fitted parameter is renamed as a prediction; Eq. (2) merely defines a decomposition of the speech stem and is not used to derive the reported SDR improvements. There are no load-bearing self-citations: references to DnR-v2, BandIt, and DnR-v3 are to external work by non-overlapping authors, and no uniqueness or ansatz result is imported from prior work of the present authors. The admitted contamination issues — LLM-filtered FreeSound clips that still contain non-human sounds and the model mistaking an animal's voice for screaming — are label-quality and external-validity concerns, not circularity. The training and evaluation sets do share the same tag-based construction pipeline, which could modestly inflate the reported gains relative to a fully independent benchmark, but this is a correctness risk rather than a reduction of the central claim to its own inputs. The empirical comparison remains meaningful under the paper's stated assumptions.

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

The dataset construction relies on several domain assumptions about labeling, curation, and evaluation. No new scientific entities are postulated.

free parameters (5)
  • lambda_r (expected number of reading-style clips) = 6
    Chosen for the zero-truncated Poisson distribution in Algorithm 1; affects the composition of speech stems.
  • lambda_n (expected number of non-verbal clips) = 5
    Chosen for the zero-truncated Poisson distribution in Algorithm 1; sets the density of non-verbal sounds.
  • alpha (skewness of silence duration distribution) = 5
    Skew Gaussian parameter in Algorithm 1; controls the distribution of silence gaps between clips.
  • sigma (scale of silence duration distribution) = 2
    Skew Gaussian parameter in Algorithm 1; affects typical silence duration in seconds.
  • Loudness values (As, Am, Af, Ab) = -17, -21, -21, -29 LUFS
    Target loudness for speech, music, foreground effect, and background effect stems; borrowed from DnR-v2.
assumptions (5)
  • domain assumption Non-verbal vocal sounds belong in the speech stem for CASS.
    Section 3.1 justifies this labeling; the paper argues it matches movie audio, but this is a definitional choice that determines the training target.
  • domain assumption FSD50K and FreeSound tags accurately label non-verbal human vocal sounds.
    Section 3.2 relies on AudioSet ontology tags; tag noise is acknowledged by the filtering steps.
  • ad hoc to paper The LLM (GPT-4o) filtering decisions are acceptably reliable for selecting valid clips.
    Section 3.3 uses an LLM prompt to filter FreeSound clips; the paper notes some non-human clips remain, so this assumption is imperfect.
  • domain assumption SDR as defined in Eq. (3) is a meaningful measure of separation quality.
    The simplified energy-ratio SDR lacks the projection used in standard BSS-Eval; the paper does not compare with standard metrics.
  • domain assumption BandIt is representative of current CASS models.
    Section 4.1 uses BandIt as the only model; results may not transfer to other architectures.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DnR-nonverbal: Cinematic Audio Source Separation Dataset Containing Non-Verbal Sounds." pith.science (2026). https://pith.science/paper/7X5AI4CJ

@misc{pith2026250602499,
  author       = {Pith},
  title        = {Pith review of: DnR-nonverbal: Cinematic Audio Source Separation Dataset Containing Non-Verbal Sounds},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7X5AI4CJ}},
  note         = {Machine review of arXiv:2506.02499}
}
read the original abstract

We propose a new dataset for cinematic audio source separation (CASS) that handles non-verbal sounds. Existing CASS datasets only contain reading-style sounds as a speech stem. These datasets differ from actual movie audio, which is more likely to include acted-out voices. Consequently, models trained on conventional datasets tend to have issues where emotionally heightened voices, such as laughter and screams, are more easily separated as an effect, not speech. To address this problem, we build a new dataset, DnR-nonverbal. The proposed dataset includes non-verbal sounds like laughter and screams in the speech stem. From the experiments, we reveal the issue of non-verbal sound extraction by the current CASS model and show that our dataset can effectively address the issue in the synthetic and actual movie audio. Our dataset is available at https://zenodo.org/records/15470640.

Figures

Figures reproduced from arXiv: 2506.02499 by the authors.

Figure 1
Figure 1. Comparision of conventional CASS datasets and pro￾posed DnR-nonverbal. Unlike conventional datasets such as DnR-v2, our dataset covers non-verbal sounds often observed in movie audio. Divide and Remaster v2 (DnR-v2) [9] uses LibriSpeech [12]. The DnR-v3 extends DnR-v2 by collecting multiple ASR cor￾pora to support multi-lingual speeches. The domain mismatch between the synthetic dataset and realistic cinematic audio… view at source ↗
Figure 2
Figure 2. Filtering procedure to extract non-verbal clips from FSD50K dataset and FreeSound. Our definition of including non-verbal sounds as a speech stem is justified in representing vocal content within movie au￾dio. In movie audio, non-verbal sounds are uttered alongside linguistic speech and recorded on the same channel. Splitting them into different stems or ignoring them is unnatural. Rather, treating them as a single … view at source ↗
Figure 3
Figure 3. Property of DnR-nonverbal dataset [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

31 extracted references · 26 canonical work pages

  1. [1]

    Typically, this task defines speech, music, and effects as the exclusive target stems

    Introduction Cinematic audio source separation (CASS) [1] aims to decom- pose the movie audio into sources. Typically, this task defines speech, music, and effects as the exclusive target stems. The CASS helps restore old movies and analyze movie content by demixing the audio. The technique may also be applicable to detect copyrighted music from audio in ...

    arXiv 2025

  2. [2]

    CASS and conventional datasets 2.1. CASS formulation The mixing process of CASS is defined as follows: y=x s +x m +x e,(1) wherex s,x m, andx e denote the monaural waveforms of speech, music, and effect respectively, andyis the mixture of stems.x e can be defined as the mixture of foregroundx f and backgroundx b effects. The CASS task is estimatingx s,x m...

    work page

  3. [3]

    Motivation In the actual movie audio, we can decomposex s as follows: xs =x v +x n,(2) wherex v andx n correspond to the waveforms of verbal and non-verbal sounds, respectively

    DnR-nonverbal 3.1. Motivation In the actual movie audio, we can decomposex s as follows: xs =x v +x n,(2) wherex v andx n correspond to the waveforms of verbal and non-verbal sounds, respectively. As described in Sec. 2.2, the conventional dataset contains only reading-style speech as ver- bal sounds (i.e.,x v ≈x r, wherex r is a reading-style speech) and...

    work page 1968

  4. [4]

    Settings To evaluate the effectiveness of the proposed dataset, we con- ducted CASS experiments

    Experiments 4.1. Settings To evaluate the effectiveness of the proposed dataset, we con- ducted CASS experiments. As a CASS model, we used Ban- dIt [10] with long short-term memory [24] backbone. The model is trained for 100 epochs by the sum of a frequency- domain mean-absolute-error (MAE) loss and a time-domain MAE loss [11] using the Adam optimizer [25...

    work page

  5. [5]

    To address this issue, we built a new dataset containing non-verbal sounds namedDnR- nonverbal

    Conclusion In this paper, we highlighted the underlying issue of the con- ventional CASS dataset: non-verbal sounds are excluded in any stems, which led the trained CASS model to treat ex- pressive voice as an effect stem. To address this issue, we built a new dataset containing non-verbal sounds namedDnR- nonverbal. Our dataset contains non-verbal sounds...

    work page

  6. [6]

    The sound demixing challenge 2023-cinematic demixing track,

    S. Uhlich, G. Fabbro, M. Hirano, S. Takahashi, G. Wich- ern, J. Le Roux, D. Chakraborty, S. Mohanty, K. Li, Y . Luo, J. Yu, R. Gu, R. Solovyev, A. Stempkovskiy, T. Habruseva, M. Sukhovei, and Y . Mitsufuji, “The sound demixing challenge 2023-cinematic demixing track,” inProceedings of International Society for Music Information Retrieval Conference, 2024,...

    work page 2023

  7. [7]

    Supervised speech separation based on deep learning: An overview,

    D. Wang, and J. Chen, “Supervised speech separation based on deep learning: An overview,”IEEE/ACM transactions on audio, speech, and language processing, vol. 26, no. 10, pp. 1702–1726, 2018

    work page 2018

  8. [8]

    Conv-TasNet: Surpassing ideal time– frequency magnitude masking for speech separation,

    Y . Luo, and N. Mesgarani, “Conv-TasNet: Surpassing ideal time– frequency magnitude masking for speech separation,”IEEE/ACM transactions on audio, speech, and language processing, vol. 27, no. 8, pp. 1256–1266, 2019

    work page 2019

Show all 31 references
  1. [9]

    TF-GridNet: Integrating full- and sub-band modeling for speech separation,

    Z.-Q. Wang, S. Cornell, S. Choi, Y . Lee, B.-Y . Kim, and S. Watan- abe, “TF-GridNet: Integrating full- and sub-band modeling for speech separation,”IEEE/ACM Transactions on Audio, Speech, and Language Processing, vol. 31, pp. 3221–3236, 2023

    2023

  2. [10]

    The 2018 signal separation eval- uation campaign,

    F.-R. St ¨oter, A. Liutkus, N. Ito “The 2018 signal separation eval- uation campaign,” inLatent V ariable Analysis and Signal Separa- tion, Springer International Publishing, 2018, pp. 293–305

    2018

  3. [11]

    Open- Unmix - a reference implementation for music source separation,

    F.-R. St ¨oter, S. Uhlich, A. Liutkus, and Y . Mitsufuji, “Open- Unmix - a reference implementation for music source separation,” Journal of Open Source Software, vol. 4, no. 41, p. 1667, 2019

    2019

  4. [12]

    Hybrid transformers for music source separation,

    S. Rouard, F. Massa, and A. D ´efossez, “Hybrid transformers for music source separation,” inProceedings of International Confer- ence on Acoustics, Speech and Signal Processing, 2023, pp.1–5

    2023

  5. [13]

    Universal sound separation,

    I. Kavalerov, S. Wisdom, H. Erdogan, B. Patton, K. Wilson, J. Le Roux, and J. R. Hershey, “Universal sound separation,” in 2019 IEEE Workshop on Applications of Signal Processing to Au- dio and Acoustics, 2019, pp. 175–179

    2019

  6. [14]

    The cocktail fork problem: Three-stem audio separation for real- world soundtracks,

    D. Petermann, G. Wichern, Z.-Q. Wang, and J. Le Roux, “The cocktail fork problem: Three-stem audio separation for real- world soundtracks,” inProceedings of International Conference on Acoustics, Speech and Signal Processing, 2022, pp. 526–530

    2022

  7. [15]

    A general- ized bandsplit neural network for cinematic audio source separa- tion,

    K. N. Watcharasupat, C.-W. Wu, Y . Ding, I. Orife, A. J. Hipple, P. A. Williams, S. Kramer, A. Lerch, and W. Wolcott, “A general- ized bandsplit neural network for cinematic audio source separa- tion,”IEEE Open Journal of Signal Processing, vol. 5, 2023

    2023

  8. [16]

    Music source separation with band-split RNN,

    Y . Luo and J. Yu, “Music source separation with band-split RNN,” IEEE/ACM Transactions on Audio, Speech, and Language Pro- cessing, vol. 31, pp. 1893–1901, 2023

    1901

  9. [17]

    Lib- riSpeech: an ASR corpus based on public domain audio books,

    V . Panayotov, G. Chen, D. Povey, and S. Khudanpur, “Lib- riSpeech: an ASR corpus based on public domain audio books,” inProceedings of International Conference on Acoustics, Speech, and Signal Processing, 2015, pp. 5206–5210

    2015

  10. [18]

    FMA: A dataset for music analysis,

    M. Defferrard, K. Benzi, P. Vandergheynst, and X. Bresson, “FMA: A dataset for music analysis,” inProceedings of the In- ternational Society for Music Information Retrieval Conference, pp. 316–323, 2017

    2017

  11. [19]

    FSD50K: an open dataset of human-labeled sound events

    E. Fonseca, X. Favory, J. Pons, F. Font, and X. Serra, “FSD50K: an open dataset of human-labeled sound events”arXiv preprint arXiv:2010.00475, 2020

    2010 arXiv

  12. [20]

    Remastering di- vide and remaster: A cinematic audio source separation dataset with multilingual support,

    K. N. Warcharasupat, C.-W. Wu, and I. Orife, “Remastering di- vide and remaster: A cinematic audio source separation dataset with multilingual support,” inProceedings of International Sym- posium on the Internet of Sounds, 2024, pp. 1–10

    2024

  13. [21]

    Hy- perbolic audio source separation,

    D. Petermann, G. Wichern, A. Subramanian, and J. Le Roux, “Hy- perbolic audio source separation,” inProceedings of International Conference on Acoustics, Speech and Signal Processing, 2023, pp. 1–5

    2023

  14. [22]

    PodcastMix: A dataset for separating music and speech in podcasts,

    N. Schmidt, J. Pons, and M. Miron, “PodcastMix: A dataset for separating music and speech in podcasts,” inProceedings of Pro- ceedings of Interspeech, 2022, pp. 231–235

    2022

  15. [23]

    Jointly recogniz- ing speech and singing voices based on multi-task audio source separation,

    Y . Bai, C. Li, H. Li, Y . Zhao, and X. Wang, “Jointly recogniz- ing speech and singing voices based on multi-task audio source separation,”arXiv preprint arXiv:2404.11275, 2024

    2024 arXiv

  16. [24]

    CSTR VCTK corpus: English multi-speaker corpus for CSTR voice cloning toolkit,

    C. Veaux, J. Yamagishi, and K. MacDonald, “CSTR VCTK corpus: English multi-speaker corpus for CSTR voice cloning toolkit,” inThe Centre for Speech Technology Research, vol. 6, p. 15, 2017

    2017

  17. [25]

    AISHELL-1: An open-source Mandarin speech corpus and a speech recognition baseline,

    H. Bu, J. Du, X. Na, B. Wu, and H. Zheng, “AISHELL-1: An open-source Mandarin speech corpus and a speech recognition baseline,” inProceedings of Conference of the Oriental Chapter of the International Coordinating Committee on Speech Databases and Speech I/O Systems and Assess...

    2017

  18. [26]

    Audio set: An ontology and human-labeled dataset for audio events,

    J. F. Gemmeke, D. P. Ellis, D. Freedman, A. Jansen, W. Lawrence, R. C. Moore, M. Plakal, and M. Ritter, “Audio set: An ontology and human-labeled dataset for audio events,” inProceedings of International Conference on Acoustics, Speech and Signal Pro- cessing, 2017, pp. 776–780

    2017

  19. [27]

    GPT-4o system card,

    A. Hurst, A. Lerer, A. P. Goucher, A. Perelman, A. Ramesh, A. Clark, A. Ostrow, A. Welihinda, A. Hayes, A. Radfordet al., “GPT-4o system card,”arXiv preprint arXiv:2410.21276, 2024

    2024 arXiv

  20. [28]

    Toward a recommendation for a European standard of peak and LKFS loudness levels,

    E. Grimm, R. Van Everdingen, and M. J. L. C. Sch ¨opping, “Toward a recommendation for a European standard of peak and LKFS loudness levels,” inSMPTE Motion Imaging Journal, vol. 119, no. 3, pp. 28–34, 2010

    2010

  21. [29]

    Long short-term memory,

    S. Hochreiter and J. Schmidhuber, “Long short-term memory,” Neural Computation, vol. 9, no. 8, pp. 1735–1780, 1997

    1997

  22. [30]

    Adam: A method for stochastic opti- mization

    D. P. Kingma and J. Ba, “Adam: A method for stochastic opti- mization.” inProceedings of International Conference on Learn- ing Representations, 2015

    2015

  23. [31]

    Why does music source separation benefit from cacophony?

    C.-B. Jeon, G. Wichern, F. G. Germain, and J. Le Roux, “Why does music source separation benefit from cacophony?” inInter- national Conference on Acoustics, Speech, and Signal Processing, 2024, pp. 873–877

    2024

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.