Pith. sign in

REVIEW 3 major objections 5 minor 73 references

MERaLiON-GR: Speech Gender Recognition Model for English and SEA Languages

T0 review · 3 major / 5 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 A LoRA-tuned conformer encoder with an ECAPA-TDNN head reports higher gender-recognition accuracy than Vox-Profile on 12 of 15 public English and Southeast Asian test sets.

desk verdict Credible 12/15 public-benchmark wins, but the abstract overclaims and the in-the-wild label pipeline needs fuller reporting; worth a serious referee. read the letter →

arxiv 2608.04433 v1 pith:TP7FAKJN submitted 2026-08-05 cs.CL cs.AI

Qiongqiong Wang , Ai Ti Aw , Nancy F. Chen , Ying Lay Chiu , Yang Ding , Yingxu He , Ridong Jiang , Zhuohan Liu
show 13 more authors
Yanfeng Lu Yi Ma Muhammad Huzaifah Nabilah Binte Md Johan Nattadaporn Lertcheva Pham Minh Duc Sailor Hardik Bhupendra Siti Umairah Binte Mohammad Salleh Shuo Sun Tarun Kumar Vangani Jeremy H. M. Wong Jinyang Wu Longyin Zhang
This is my paper · ORCID
classification cs.CLcs.AI
keywords speechgenderrecognitionparalinguisticanalysisLow-RankAdaptation(LoRA)ECAPA-TDNNconformerencoderSoutheastAsianlanguagescross-lingualgeneralizationaudiolargelanguagemodels
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

MERaLiON-GR is a speech gender-recognition system that starts from a large pre-trained conformer speech encoder, keeps the encoder frozen, and adapts it with low-rank adaptation (LoRA) plus a compact ECAPA-TDNN classifier. The paper's central claim is that this recipe reports higher accuracy than the previous state-of-the-art dedicated gender model, Vox-Profile, on 12 of 15 public English and Southeast Asian test sets, and equal accuracy on one, while a general-purpose audio large language model (Audio-LLM) falls further behind. On two-second in-the-wild Singapore speech in four languages, the model reports gains of up to 4.32 percentage points over Vox-Profile. If these results hold, they support the argument that a task-specific, parameter-efficient model is the right tool for fine-grained paralinguistic attributes and that such attributes can generalize across languages.

What carries the argument

The load-bearing mechanism is the coupling of a frozen Conformer speech encoder (a transformer variant with convolutional modules) with low-rank adaptation and a compact ECAPA-TDNN head. LoRA lets the task update only low-rank factors in the attention projections, which keeps adaptation cheap and limits catastrophic forgetting; a learned soft attention over all 25 layer outputs combines phonetic and speaker-level cues; ECAPA-TDNN, a time-delay neural network with emphasized channel attention and attention pooling, turns the frame-level representation into a fixed 256-dimensional utterance embedding; and a linear head with RMSNorm and GELU maps that embedding to the female/male logits. This combination is what lets a generic multilingual speech encoder become a specialized, cross-lingually stable gender classifier without full fine-tuning.

What would settle it

Re-annotate the SG-ECMT segments with multiple independent native-speaking annotators who never see any automatic prediction, then compare MERaLiON-GR's two-second segment accuracy with Vox-Profile on the human-only labels; if the gap shrinks or disappears, the claimed in-the-wild advantage is an artifact of label alignment.

Watch

Extended reading notes

Core claim

The paper's discovery claim is that gender recognition from speech is better served by a dedicated encoder fine-tuned with low-rank adapters than by either the current standalone state of the art or a large audio LLM. The model keeps MERaLiON-SpeechEncoder-2 frozen, inserts LoRA into its attention projections, aggregates all 25 hidden states with learned layer weights, and feeds the result through an ECAPA-TDNN head with attention pooling and a two-class linear classifier. Reported accuracy reaches 100% on English FLEURS, Tamil OpenSLR, and Thai Elderly, and the largest gains over Vox-Profile are on Tamil EmoTa (+4.71 percentage points), Vietnamese Common Voice (+3.14), and Indonesian IndoWave (+3.00), while the clear losses are on Malay SMALDUSC and Thai SER. The paper further claims that feeding the predicted gender into an Audio-LLM lifts gender-question answering accuracy substantially, by up to 60 percentage points on Vietnamese Common Voice.

Load-bearing premise

The Singapore in-the-wild test set was labelled by an automatic model first and then corrected by native speakers, and the paper reports neither the human correction rate nor annotator agreement, so the segment-level gains could partly reflect agreement with that automatic labeler rather than true gender recognition.

Editorial extensions

If this is right

  • A dedicated gender-recognition model is likely to remain more accurate than a general-purpose audio LLM on this task, so deployment for paralinguistic metadata should use task-specific heads.
  • The same LoRA-plus-ECAPA recipe can be applied to other paralinguistic attributes such as age, emotion, or dialect, using the same frozen backbone.
  • The model can serve as a pseudo-labeling tool for two-second speech segments, which is useful for building larger annotated paralinguistic datasets.
  • Injecting predicted gender as explicit metadata can substantially improve audio-LLM answers to gender-related questions, suggesting a conditioning strategy for speech-language models.
  • Singapore-centric training data transfers to multiple Southeast Asian languages, supporting low-resource language applications without new task-specific training.

Reading between the lines

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

  • The paper does not test whether its SG-ECMT segment-level advantage survives fully independent human labels; because the labels were seeded by an automatic WavLM-based model, a re-annotation study with no access to those predictions would separate true gains from label alignment.
  • A dataset-level count such as 12 of 15 treats each benchmark equally; a pooled per-sample significance test could show whether the aggregate advantage is driven by the large Common Voice partitions.
  • The same frozen backbone plus LoRA and a task head is a general recipe: age, emotion, dialect, and other paralinguistic attributes could likely be added at low marginal cost, which the paper does not claim to demonstrate.
  • Extending the evaluation to Filipino, Burmese, and Lao would test the cross-lingual claim beyond the languages represented in the training corpus.
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 / 5 minor

Summary. The paper presents MERaLiON-GR, a speech gender recognition model that combines a LoRA-adapted MERaLiON-SpeechEncoder-2 backbone with an ECAPA-TDNN head and attention pooling. The authors evaluate the model on fifteen public benchmark test sets spanning English and eight Southeast Asian languages, on a smaller set of public benchmarks for Audio-LLM metadata injection, and on four internal Singapore in-the-wild segment-level sets (SG-ECMT). The central claim is that MERaLiON-GR consistently outperforms the state-of-the-art Vox-Profile model and a general-purpose Audio-LLM in both full-utterance and segment-level evaluation. The training recipe, including LoRA rank, scaling, dropout, class-balanced loss, checkpoint averaging, and hyperparameters, is reported in detail.

Significance. If the central claim holds, the paper provides a practical, parameter-efficient recipe for multilingual gender recognition that improves on public state-of-the-art results across several SEA languages, and it demonstrates that explicit paralinguistic metadata can substantially improve Audio-LLM performance on gender-related questions. The strengths of the paper are its broad public benchmark coverage, the use of an external baseline (Vox-Profile) rather than a purely self-referential comparison, the release of the model and demo, and the clear description of training configuration. However, the abstract-level claim of 'consistently surpasses' is stronger than the reported data, and the internal SG-ECMT evaluation depends on labels that were seeded by an automatic model before human correction, with no reported agreement or correction statistics. These issues make the paper's headline claim defensible on the public benchmarks but not yet fully verified as stated.

major comments (3)
  1. [Abstract, §1, §6.1, Table 1] The abstract and §7 state that MERaLiON-GR 'consistently surpasses' Vox-Profile, but Table 1 reports two losses (Malay SMALDUSC, −4.40 pp; Thai THAI-SER, −2.09 pp) and one tie (Khmer FLEURS, 99.74 both). The introduction's more careful wording ('outperforming Vox-Profile on 12 of 15 public benchmarks') should be adopted throughout, and the abstract and summary should be revised accordingly. In addition, no confidence intervals or significance tests are provided; several margins are small (e.g., IEMOCAP 98.90 vs. 97.31, Tamil Common Voice 94.00 vs. 92.30), so the reader cannot distinguish genuine improvement from evaluation noise. Please report confidence intervals, bootstrap estimates, or pairwise significance tests at least for the public benchmark comparisons.
  2. [§4.1.1, §6.2, Table 3] The SG-ECMT in-the-wild labels were automatically estimated with a WavLM-ECAPA model and then 'reviewed and corrected by native-speaking annotators,' but the paper reports no fraction of label changes, no inter-annotator agreement, and no explicit correction criteria. If the annotators were anchored by the automatic labels, the segment-level improvements in Table 3 (e.g., +4.32 pp on Malay and +4.16 pp on Tamil) could reflect alignment with the seed model's biases rather than true recognition quality. Please report the proportion of samples whose labels were changed by human review, agreement statistics (e.g., Cohen's kappa or percentage agreement on a subset), and a description of the correction protocol, or provide an alternative evaluation with independently produced human labels.
  3. [§6.1, Table 1] The paper reports 100.00% accuracy on English FLEURS, Tamil OpenSLR, and Thai Elderly. Perfect accuracy on public datasets that are likely to contain at least some label noise or utterance-level gender ambiguity is surprising and deserves explicit support. Please describe any duplicate-removal or leakage checks performed between the pretraining/fine-tuning corpora and these test sets, report the number of test utterances and any samples excluded due to label ambiguity, and state whether the reported 100.00% figures are stable under minor perturbations of the evaluation protocol.
minor comments (5)
  1. [§2.2] The sentence 'All BatchNorm layers are replaced with GroupNorm layers that are in the original ECAPA-TDNN structure' is unclear; please rephrase to indicate whether GroupNorm were already part of the original ECAPA-TDNN implementation used, or whether they were introduced as a modification in this work.
  2. [§5] The fixed prompt for MERaLiON-v2 is a reasonable choice, but a single prompt may underestimate the Audio-LLM's capability; a brief sensitivity check with one or two alternative phrasings would strengthen the claim that the Audio-LLM is a fair baseline.
  3. [§3.2] The text says training ran for 15 epochs with early stopping based on development loss, but also that final parameters were obtained by averaging the four lowest-loss checkpoints; please clarify whether the averaged checkpoints came from the same 15-epoch run and how early stopping interacted with checkpoint selection.
  4. [§6.1, Table 2] The GR-AudioLLM results in Table 2 are evaluated on a subset of the public benchmarks (8 of 15); the text should explicitly state that this subset was chosen because gender-related metadata injection is only applicable to those datasets, or explain the selection criterion.
  5. [Throughout] There are a few typographical and formatting issues, including inconsistent spacing in Equation (2) and the use of 'V ox-Profile' with a line break in several places; a careful proofread would improve readability.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: central claim is benchmarked against independent public sets and an external baseline; self-citations are provenance, not load-bearing.

full rationale

The derivation chain is self-contained with respect to the central claim. MERaLiON-GR is trained on VoxCeleb1 and IMDA partitions, then evaluated on 15 public benchmarks against Vox-Profile and an Audio-LLM. No evaluation-set label is used to fit or select the model, so the reported accuracies are external measurements rather than outputs of the model's own construction. The SG-ECMT in-the-wild set uses labels seeded by an external WavLM-ECAPA model and then corrected by native annotators; the unreported correction rate is a data-quality limitation, not circularity, because the model is not trained or early-stopped on those labels. Self-citations to MERaLiON-SpeechEncoder-2, MERaLiON-v2, and earlier paralinguistic-metadata work are provenance and contextual support, not load-bearing reductions of the scored claim. Table 2 intentionally injects MERaLiON-GR's own predictions as metadata, which is the tested intervention rather than a derivation of the result. The abstract's 'consistently surpasses' is broader than the intro's disclosed 12-of-15 with one tie and two losses, but that is an overstatement, not circularity. No equation or fitted parameter is reused as a predicted quantity, and no uniqueness or ansatz is imported from self-citations to force the outcome.

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

This is an empirical engineering paper. The central claim rests on architecture and training hyperparameters chosen by hand, a set of label-assumption axioms, and an unexamined assumption about pretraining overlap, rather than on a closed-form derivation or a new theoretical entity.

free parameters (5)
  • LoRA rank and scaling factor = 16
    Chosen by hand in Section 2.1; no ablation or sensitivity analysis is reported.
  • LoRA dropout rate = 0.05
    Chosen by hand in Section 2.1; standard value, no tuning reported.
  • Label smoothing epsilon = 0.1
    Chosen by hand in Section 3.2; affects the training loss but no ablation is shown.
  • Optimizer hyperparameters = LRs 5e-5 (LoRA) and 6e-4 (head) with weight decays 5e-4 and 8e-5
    Set by hand in Section 3.2; no search procedure or sensitivity analysis is reported.
  • Class-balance weights w_i = computed from effective number of samples
    Fitted to training-set class frequencies via Cui et al. 2019; used in Eq. (2) and affects training.
assumptions (5)
  • domain assumption MERaLiON-SpeechEncoder-2 representations are suitable for gender classification across the target languages.
    The entire pipeline assumes the frozen pretrained encoder plus LoRA adapters carries enough speaker and gender information; introduced in Section 2.1.
  • domain assumption Public benchmark gender labels are reliable and comparable across datasets.
    Accuracy is computed against these labels; no label-noise analysis is provided in Section 6.1.
  • domain assumption Human-corrected WavLM-ECAPA pseudo-labels for SG-ECMT are an adequate ground truth.
    Section 4.1.1 does not report correction rate or inter-annotator agreement; the segment-level claims in Table 3 depend on this.
  • ad hoc to paper No significant overlap exists between the pretraining corpus of MERaLiON-SpeechEncoder-2 and the public test sets.
    The paper does not check contamination; if overlap exists, reported gains could be inflated.
  • ad hoc to paper The fixed prompt used for the Audio-LLM baseline is a fair elicitation of its gender-recognition ability.
    Prompt engineering can substantially change LLM accuracy; no prompt search or alternative prompts are reported in Section 5.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MERaLiON-GR: Speech Gender Recognition Model for English and SEA Languages." pith.science (2026). https://pith.science/paper/TP7FAKJN

@misc{pith2026260804433,
  author       = {Pith},
  title        = {Pith review of: MERaLiON-GR: Speech Gender Recognition Model for English and SEA Languages},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TP7FAKJN}},
  note         = {Machine review of arXiv:2608.04433}
}
read the original abstract

We present MERaLiON-GR, a speech gender recognition system that performs binary classification (female / male) on English and Southeast Asian (SEA) languages. The model finetunes MERaLiON-SpeechEncoder-2, a large conformer based transformer pre-trained on a broad speech corpus, and applies parameter efficient fine-tuning via Low-Rank Adaptation (LoRA) to adapt the encoder to the gender recognition task, and appends a multi-scale ECAPA-TDNN down stream network with attention pooling and a lightweight linear classifier. Extensive evaluations across multilingual Singaporean and Southeast Asian languages (English, Chinese, Malay, Tamil, Thai, Vietnamese, Indonesian, and Khmer) show that MERaLiON-GR consistently surpasses the state-of-the-art gender recognition model Vox-Profile and a large Audio-LLM, in both full-utterance and segment level evaluation modes. The results underscore the value of dedicated speech models in achieving accurate paralinguistic understanding and strong cross-lingual generalization.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

73 extracted references · 63 canonical work pages

  1. [1]

    Findings of EMNLP 2025 , year=

    Benchmarking Contextual and Paralinguistic Reasoning in Speech-LLMs: A Case Study with In-the-Wild Data , author=. Findings of EMNLP 2025 , year=

    work page 2025

  2. [2]

    Contextual Paralinguistic Data Creation for Multi-Modal Speech-LLM: Data Condensation and Spoken

    Wang, Qiongqiong and Sailor, Hardik B and Liu, Tianchi and Aw, Ai Ti , booktitle=. Contextual Paralinguistic Data Creation for Multi-Modal Speech-LLM: Data Condensation and Spoken

    work page

  3. [3]

    Incorporating Contextual Paralinguistic Understanding in Large Speech-Language Models , author=. Proc. IEEE Automatic Speech Recognition and Understanding Workshop (ASRU) , year=

    work page

  4. [4]

    arXiv preprint arXiv:2312.11805 , year=

    Gemini: A Family of Highly Capable Multimodal Models , author=. arXiv preprint arXiv:2312.11805 , year=

    arXiv

  5. [5]

    Proceedings of the 40th International Conference on Machine Learning (ICML) , year =

    Radford, Alec and Kim, Jong Wook and Xu, Tao and Brockman, Greg and McLeavey, Christine and Sutskever, Ilya , title =. Proceedings of the 40th International Conference on Machine Learning (ICML) , year =

    work page

  6. [6]

    He, Yingxu and Liu, Zhuohan and Sun, Shuo and Wang, Bin and Zhang, Wenyu and Zou, Xunlong and Chen, Nancy F and Aw, Ai Ti , journal=

    work page

  7. [7]

    Ziyang Ma and Mingjie Chen and Hezhao Zhang and Zhisheng Zheng and Wenxi Chen and Xiquan Li and Jiaxin Ye and Xie Chen and Thomas Hain , year =

    work page

  8. [8]

    Lotfian and C

    R. Lotfian and C. Busso , title =. IEEE Transactions on Affective Computing , volume =. 2019 , pages =

    work page 2019

Show all 73 references
  1. [9]

    , journal=

    Josh Achiama and Steven Adler and Sandhini Agarwal and Lama Ahmad and Ilge Akkaya and Florencia Leoni Aleman and Diogo Almeida and Janko Altenschmidt and Sam Altman and Shyamal Anadkat and et al. , journal=

  2. [10]

    Psychological review , year=

    What’s Basic About Basic Emotions? , author=. Psychological review , year=

  3. [11]

    Academic Press , year=

    Theories of Emotion , author=. Academic Press , year=

  4. [12]

    An argument for basic emotions , Volume =

    Paul Ekman , Journal =. An argument for basic emotions , Volume =

  5. [13]

    1977 , author =

    Evidence for a three-factor theory of emotions , journal =. 1977 , author =

    1977

  6. [14]

    Findings of the Association for Computational Linguistics (ACL) , year=

    emotion2vec: Self-Supervised Pre-Training for Speech Emotion Representation , author=. Findings of the Association for Computational Linguistics (ACL) , year=

  7. [15]

    Chang and Sungbok Lee and Shrikanth S

    Carlos Busso and Murtaza Bulut and Chi-Chun Lee and Ebrahim Kazemzadeh and Emily Mower Provost and Samuel Kim and Jeannette N. Chang and Sungbok Lee and Shrikanth S. Narayanan , journal=. 2008 , volume=

    2008

  8. [16]

    Soujanya Poria and Devamanyu Hazarika and Navonil Majumder and Gautam Naik and Erik Cambria and Rada Mihalcea , booktitle=

  9. [17]

    and King, Irwin

    Cui, Wenqian and Yu, Dianzhi and Jiao, Xiaoqi and Meng, Ziqiao and Zhang, Guangyan and Wang, Qichao and Guo, Steven Y. and King, Irwin. Recent Advances in Speech Language Models: A Survey. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics ...

    2025

  10. [18]

    Brecht Desplanques and Jenthe Thienpondt and Kris Demuynck , year =. Proc. Interspeech , pages =

  11. [19]

    ICLR , year=

    Mixup: Beyond empirical risk minimization , author=. ICLR , year=

  12. [20]

    Snyder, David and Chen, Guoguo and Povey, Daniel , journal=

  13. [21]

    2014 , publisher=

    Cao, Huai and Cooper, David G and Keutmann, Michael K and Gur, Ruben C and Nenkova, Ani and Verma, Raghuveer , journal=. 2014 , publisher=

    2014

  14. [22]

    Huang, Zhenpeng and Yang, Hao and Yan, Ming and Yang, Qing , booktitle=

  15. [23]

    Emotional Speech Dataset (

    Zhou, Yixuan and Zhou, Lei , booktitle=. Emotional Speech Dataset (. 2018 , pages=

    2018

  16. [24]

    IEEE Transactions on Affective Computing , volume=

    Busso, Carlos and. IEEE Transactions on Affective Computing , volume=. 2017 , publisher=

    2017

  17. [25]

    GitHub repository , howpublished =

    Chaoqun Liu and Mahani Aljunied and Guizhen Chen and Hou Pong Chan and Weiwen Xu and Yu Rong and Wenxuan Zhang , title =. GitHub repository , howpublished =. 2025 , publisher =

    2025

  18. [26]

    Chen, Jingyi and Guo, Zhimeng and Chun, Jiyun and Wang, Pichao and Perrault, Andrew and Elsner, Micha , journal=

  19. [27]

    and Edin, Joakim and Igel, Christian and Kirchhoff, Katrin and Li, Shang-Wen and Livescu, Karen and Maaløe, Lars and Sainath, Tara N

    Mohamed, Abdelrahman and Lee, Hung-yi and Borgholt, Lasse and Havtorn, Jakob D. and Edin, Joakim and Igel, Christian and Kirchhoff, Katrin and Li, Shang-Wen and Livescu, Karen and Maaløe, Lars and Sainath, Tara N. and Watanabe, Shinji , journal=. Self-Supervised Speech Represe...

  20. [28]

    doi:10.21437/Interspeech.2025-1283 , issn =

    Drishya Uniyal and Vinayak Abrol , year =. doi:10.21437/Interspeech.2025-1283 , issn =

    2025 doi

  21. [29]

    2025 , author =

    Speech self-supervised representations benchmarking: A case for larger probing heads , journal =. 2025 , author =

    2025

  22. [30]

    and Richerson, Elizabeth and Robinson, Donita and Busso, Carlos , booktitle=

    Naini, Abinay Reddy and Kohler, Mary A. and Richerson, Elizabeth and Robinson, Donita and Busso, Carlos , booktitle=. Generalization of Self-Supervised Learning-Based Representations for Cross-Domain Speech Emotion Recognition , year=

  23. [31]

    Kaplan and Tamer Nadeem , year =

    Mohamed Osman and Daniel Z. Kaplan and Tamer Nadeem , year =. doi:10.21437/Interspeech.2024-2440 , issn =

    2024 doi

  24. [32]

    Chen Wang and Minpeng Liao and Zhongqiang Huang and Jinliang Lu and Junhong Wu and Yuchen Liu and Chengqing Zong and Jiajun Zhang , journal=

  25. [33]

    Chen Wang and Minpeng Liao and Zhongqiang Huang and Junhong Wu and Chengqing Zong and Jiajun Zhang , journal=

  26. [34]

    arXiv preprint arXiv:2306.12925 , year=

    Rubenstein, Paul K and Asawaroengchai, Chulayuth and Nguyen, Duc Dung and Bapna, Ankur and Borsos, Zal. arXiv preprint arXiv:2306.12925 , year=

    arXiv

  27. [35]

    Qian Chen and Yunfei Chu and Zhifu Gao and Zerui Li and Kai Hu and Xiaohuan Zhou and Jin Xu and Ziyang Ma and Wen Wang and Siqi Zheng and others , journal=

  28. [36]

    arXiv preprint arXiv:2402.12786 , year=

    Advancing large language models to capture varied speaking styles and respond properly in spoken conversations , author=. arXiv preprint arXiv:2402.12786 , year=

    arXiv

  29. [37]

    Interspeech , year=

    Speech self-supervised representation benchmarking: Are we doing it right? , author=. Interspeech , year=

  30. [38]

    arXiv preprint arXiv:2410.01162 , year=

    Frozen Large Language Models Can Perceive Paralinguistic Aspects of Speech , author=. arXiv preprint arXiv:2410.01162 , year=

    arXiv

  31. [39]

    The Thirty-eighth Annual Conference on Neural Information Processing Systems (NeurIPS) , year =

    Paralinguistics-Aware Speech-Empowered Large Language Models for Natural Conversation , author=. The Thirty-eighth Annual Conference on Neural Information Processing Systems (NeurIPS) , year =

  32. [40]

    Cheng, Xize and Hu, Ruofan and Yang, Xiaoda and Lu, Jingyu and Fu, Dongjie and Wang, Zehan and Ji, Shengpeng and Huang, Rongjie and Zhang, Boyang and Jin, Tao and others , booktitle=

  33. [41]

    Asia Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) , pages=

    Empower Typed Descriptions by Large Language Models for Speech Emotion Recognition , author=. Asia Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) , pages=. 2024 , organization=

    2024

  34. [42]

    Xu, Yaoxun and Chen, Hangting and Yu, Jianwei and Huang, Qiaochu and Wu, Zhiyong and Zhang, Shi-Xiong and Li, Guangzhi and Luo, Yi and Gu, Rongzhi , booktitle=

  35. [43]

    2024 , pages=

    Lin, Wei-Cheng and Ghaffarzadegan, Shabnam and Bondi, Luca and Kumar, Abinaya and Das, Samarjit and Wu, Ho-Hsiang , booktitle=. 2024 , pages=

    2024

  36. [44]

    Beyond Silent Letters: Amplifying LLM s in Emotion Recognition with Vocal Nuances

    Wu, Zehui and Gong, Ziwei and Ai, Lin and Shi, Pengyuan and Donbekci, Kaan and Hirschberg, Julia. Beyond Silent Letters: Amplifying LLM s in Emotion Recognition with Vocal Nuances. Findings of the Association for Computational Linguistics: NAACL 2025. 2025

    2025

  37. [45]

    Interspeech , doi =

    Lu, Ke-Han and Chen, Zhehuai and Fu, Szu-Wei and Huang, He and Ginsburg, Boris and Wang, Yu-Chiang and Lee, Hung-yi , year =. Interspeech , doi =

  38. [46]

    Lu, Ke-Han and Chen, Zhehuai and Fu, Szu-Wei and Yang, Chao-Han Huck and Balam, Jagadeesh and Ginsburg, Boris and Wang, Yu-Chiang and Lee, Hung-yi , year =

  39. [47]

    arXiv preprint arXiv:2505.14648 , year=

    Vox-Profile: A Speech Foundation Model Benchmark for Characterizing Diverse Speaker and Speech Traits , author=. arXiv preprint arXiv:2505.14648 , year=

    arXiv

  40. [48]

    Baevski, Alexei and Zhou, Henry and Mohamed, Abdelrahman and Auli, Michael , booktitle=

  41. [49]

    Hsu, Wei-Ning and Bolte, Benjamin and Tsai, Yao-Hung Hubert and Lakhotia, Kushal and Salakhutdinov, Ruslan and Mohamed, Abdelrahman , journal=

  42. [50]

    Hu, Edward J and Shen, Yaliu and Wallis, Phillip and Allen-Zhu, Zeyuan and Li, Yuanzhi and Wang, Shean and Wang, Lu and Chen, Weizhu , booktitle=

  43. [51]

    A Rank Stabilization Scaling Factor for Fine-Tuning with

    Kalajdzievski, Damjan , journal=. A Rank Stabilization Scaling Factor for Fine-Tuning with

  44. [52]

    Yang, Shu-wen and Chi, Po-Han and Chuang, Yung-Sung and Lai, Cheng-I Jeff and Lakhotia, Kushal and Lin, Yist Y and Lee, Hung-yi , booktitle=

  45. [53]

    International Conference on Learning Representations (ICLR) , year=

    Long-Tail Learning via Logit Adjustment , author=. International Conference on Learning Representations (ICLR) , year=

  46. [54]

    European Conference on Computer Vision (ECCV) , pages=

    A Discriminative Feature Learning Approach for Deep Face Recognition , author=. European Conference on Computer Vision (ECCV) , pages=

  47. [55]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages=

    Class-Balanced Loss Based on Effective Number of Samples , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , pages=

  48. [56]

    Proceedings of the International Conference on Machine Learning (ICML) , year=

    Self-Supervised Learning with Random-Projection Quantizer for Speech and Audio Representation , author=. Proceedings of the International Conference on Machine Learning (ICML) , year=

  49. [57]

    Proceedings of Interspeech , pages=

    Acoustic-Prosodic and Lexical Cues to Deception and Trust: Deciphering How People Detect Lies , author=. Proceedings of Interspeech , pages=

  50. [58]

    Wu, Zhizheng and Kinnunen, Tomi and Evans, Nicholas and Yamagishi, Junichi and Hanilci, Cemal and Sahidullah, Md and Sizov, Aleksandr , booktitle=

  51. [59]

    2024 , eprint=

    MERaLiON-SpeechEncoder: Towards a Speech Foundation Model for Singapore and Beyond , author=. 2024 , eprint=

    2024

  52. [60]

    2022 , publisher=

    Chen, Sanyuan and Wang, Chengyi and Chen, Zhengyang and Wu, Yu and Liu, Shujie and Chen, Zhuo and Li, Jinyu and Kanda, Naoyuki and Yoshioka, Takuya and Xiao, Xiong and others , journal=. 2022 , publisher=

    2022

  53. [61]

    Arsha Nagrani and Joon Son Chung and Andrew Zisserman , year =. Proc. Interspeech , pages =

  54. [62]

    Chung, Joon Son and Nagrani, Arsha and Zisserman, Andrew , booktitle=

  55. [63]

    Advancing

    Wang, Bin and Zou, Xunlong and Sun, Shuo and Zhang, Wenyu and He, Yingxu and Liu, Zhuohan and Wei, Chengwei and Chen, Nancy F and Aw, AiTi , journal=. Advancing

  56. [64]

    Building the

    Jia Xin Koh and Aqilah Mislan and Kevin Khoo and Brian Ang and Wilson Ang and Charmaine Ng and Ying-Ying Tan , year =. Building the. Proc. Interspeech , pages =. doi:10.21437/Interspeech.2019-1525 , issn =

    2019 doi

  57. [65]

    2008 , publisher=

    Busso, Carlos and Bulut, Murtaza and Lee, Chi-Chun and Kazemzadeh, Abe and Mower, Emily and Kim, Samuel and Chang, Jeannette N and Lee, Sungbok and Narayanan, Shrikanth S , journal=. 2008 , publisher=

    2008

  58. [66]

    2023 , organization=

    Conneau, Alexis and Ma, Min and Khanuja, Simran and Zhang, Yu and Axelrod, Vera and Dalmia, Siddharth and Riesa, Jason and Rivera, Clara and Bapna, Ankur , booktitle=. 2023 , organization=

    2023

  59. [67]

    Proceedings of the 12th Language Resources and Evaluation Conference (LREC) , pages=

    Common Voice: A Massively-Multilingual Speech Corpus , author=. Proceedings of the 12th Language Resources and Evaluation Conference (LREC) , pages=

  60. [68]

    Open-source Multi-speaker Speech Corpora for Building

    He, Fei and Chu, Shan-Hui Cathy and Kjartansson, Oddur and Rivera, Clara and Katanova, Anna and Gutkin, Alexander and Demirsahin, Isin and Johny, Cibu and Jansche, Martin and Sarin, Supheakmungkol and Pipatsrisawat, Knot , booktitle=. Open-source Multi-speaker Speech Corpora f...

    2020

  61. [69]

    2025 , address=

    Thevakumar, Jubeerathan and Thavarasa, Luxshan and Sivatheepan, Thanikan and Kugarajah, Sajeev and Thayasivam, Uthayasanker , booktitle=. 2025 , address=

    2025

  62. [70]

    Jilamika Wongpithayadisai and Chompakorn Chaksangchaichot and Soravitt Sangnark and Patawee Prakrankamanant and Krit Gangwanpongpun and Siwa Boonpunmongkol and Premmarin Milindasuta and Dangkamon Na-Pombejra and Sarana Nutanong and Ekapol Chuangsuwanich , journal =

  63. [71]

    2023 , howpublished=

    Thai Elderly Speech Dataset , author=. 2023 , howpublished=

    2023

  64. [72]

    and Warni, Elly and Areni, Intan S

    Bustamin, Anugrayani and Rizky, Andi M. and Warni, Elly and Areni, Intan S. and Indrabayu, Indrabayu , journal=. 2024 , publisher=

    2024

  65. [73]

    2023 , howpublished=

    2023

Pith tools

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