MERaLiON/sea_audiobench_datasets_PQA

SEA-SpeechBench — Paralinguistics (AGE, ER, GR)

This dataset holds the three paralinguistic tasks of SEA-SpeechBench, a large-scale multitask benchmark for speech understanding across Southeast Asia. It contains 25,563 evaluation examples across nine languages, drawn from fourteen source corpora, each pairing an audio recording with an instruction and a reference answer.

Given the recording and the instruction, a model must identify a property of the speaker or the delivery rather than the words spoken:

• AGE — age recognition (age_* configs)
• ER — emotion recognition (er_* configs)
• GR — gender recognition (gr_* configs)

Quick start

Requires datasets>=4.0, which decodes audio through torchcodec and needs a system FFmpeg (versions 4–7):

pip install "datasets[audio]"

from datasets import load_dataset

ds = load_dataset(
    "MERaLiON/sea_audiobench_datasets_Paralinguistics",
    "er_esd_en_30",
    split="train",
)

row = ds[0]
print(row["instruction"])
print(row["answer"])
print(row["language"], row["audio_length"])

# `context` is a torchcodec AudioDecoder, not a dict
samples = row["context"].get_all_samples()
waveform = samples.data          # torch.Tensor, shape (num_channels, num_samples)
sr = samples.sample_rate         # 16000
print(waveform.shape, sr)

If you need a numpy array — most feature extractors take one — squeeze the channel dimension (all audio here is mono):

audio = samples.data.squeeze(0).numpy()

Subsets

All data lives in a single train split and is intended for evaluation only. Config names follow {task}_{source}[_{language}]_30, where the task prefix is age, er, or gr.

AGE — age recognition

Config	Source	Lang	Examples	Audio (hr)
age_cv21_en_30	Common Voice 21	en	1,000	1.6
age_cv21_ta_30	Common Voice 21	ta	1,000	1.5
age_cv21_th_30	Common Voice 21	th	775	1.0
age_cv21_vi_30	Common Voice 21	vi	833	0.8
age_cv21_zh_30	Common Voice 21	zh	1,000	1.6
Subtotal			4,608	~6.6

ER — emotion recognition

Config	Source	Lang	Examples	Audio (hr)
er_emota_ta_30	EmoTa	ta	936	0.7
er_esd_en_30	ESD	en	1,000	0.8
er_esd_zh_30	ESD	zh	1,000	0.9
er_indowave_id_30	IndoWaveSentiment	id	300	0.3
er_m3ed_30	M3ED	zh	1,000	0.4
er_tec_ta_30	TEC	ta	165	0.7
er_thai_ser_th_30	THAI SER	th	955	1.6
Subtotal			5,356	~5.3

GR — gender recognition

Config	Source	Lang	Examples	Audio (hr)
gr_cv21_en_30	Common Voice 21	en	1,000	1.6
gr_cv21_id_30	Common Voice 21	id	1,000	1.1
gr_cv21_ta_30	Common Voice 21	ta	1,000	1.5
gr_cv21_th_30	Common Voice 21	th	747	0.9
gr_cv21_vi_30	Common Voice 21	vi	765	0.8
gr_cv21_zh_30	Common Voice 21	zh	1,000	1.7
gr_emota_ta_30	EmoTa	ta	936	0.7
gr_fleurs_en_30	FLEURS	en	647	1.8
gr_fleurs_km_30	FLEURS	km	765	3.1
gr_indowave_id_30	IndoWaveSentiment	id	300	0.3
gr_m3ed_30	M3ED	zh	1,000	0.4
gr_openslr_ta_30	OpenSLR	ta	1,000	1.7
gr_sfdusc_30	ASR-SFDuSC	tl	1,000	1.2
gr_sg_streets_utterance_30	SG Streets	en	492	0.7
gr_smaldusc_30	ASR-SMalDuSC	ms	1,000	2.1
gr_thai_elderly_th_30	Thai Elderly Speech	th	992	1.4
gr_thai_ser_th_30	THAI SER	th	955	1.6
gr_vietnam_celeb_30	Vietnam-Celeb	vi	1,000	2.1
Subtotal			15,599	~24.9

Across all three tasks: 25,563 examples, ~36.8 hours.

Data fields

Field	Type	Description
context	Audio(sampling_rate=16000)	The audio recording.
instruction	string	The instruction given to the model.
answer	string	The reference answer.
audio_length	float64	Duration of context, in seconds.
language	string	Language code of the subset.

See the paper for the instruction and answer formats, the label sets used for each task, and the evaluation protocol.

Source data

These tasks introduce no new recordings. All audio comes from existing corpora; the contribution is the utterance selection, instructions, and reference answers.

Source	Tasks	Lang	Upstream license
Common Voice 21	AGE, GR	en, id, ta, th, vi, zh	MPL 2.0
EmoTa	ER, GR	ta	EACL
ESD	ER	en, zh	MIT
FLEURS	GR	en, km	CC BY 4.0
IndoWaveSentiment	ER, GR	id	CC BY 4.0
M3ED	ER, GR	zh	CC BY-NC-ND 4.0
OpenSLR	GR	ta	CC BY-SA 4.0
ASR-SFDuSC	GR	tl	CC BY-NC-ND 4.0
SG Streets	GR	en	Not specified
ASR-SMalDuSC	GR	ms	CC BY-NC-ND 4.0
TEC	ER	ta	Not specified
Thai Elderly Speech	GR	th	CC BY-SA 4.0
THAI SER	ER, GR	th	CC BY-SA 4.0
Vietnam-Celeb	GR	vi	CC BY 4.0

Citation

If you use this dataset, please cite our benchmark and the source corpora.

This benchmark

@inproceedings{liao2026seaspeechbench,
  title     = {{SEA-SpeechBench}: A Large-Scale Multitask Benchmark for
               Speech Understanding Across Southeast Asia},
  author    = {Liao, Jingyi and Zhang, Wenyu and Liu, Zhuohan and
               He, Yingxu and Lin, Geyu and Zou, Xunlong and Sun, Shuo and
               Alsagoff, Syed Ali Redha and Aw, Ai Ti},
  booktitle = {Proceedings of the 2026 Conference on Empirical Methods in
               Natural Language Processing (EMNLP)},
  year      = {2026}
}

Source corpora

@inproceedings{commonvoice,
  title     = {Common Voice: A Massively-Multilingual Speech Corpus},
  author    = {Ardila, Rosana and Branson, Megan and Davis, Kelly and
               Kohler, Michael and Meyer, Josh and Henretty, Michael and
               Morais, Reuben and Saunders, Lindsay and Tyers, Francis and
               Weber, Gregor},
  booktitle = {Conference on Language Resources and Evaluation},
  year      = {2020}
}

@inproceedings{emota,
  title     = {EmoTa: A Tamil Emotional Speech Dataset},
  author    = {Thevakumar, Jubeerathan and Thavarasa, Luxshan and
               Sivatheepan, Thanikan and Kugarajah, Sajeev and
               Thayasivam, Uthayasanker},
  booktitle = {Proceedings of the First Workshop on Challenges in Processing
               South Asian Languages (CHiPSAL 2025)},
  pages     = {193--201},
  month     = {January},
  year      = {2025},
  address   = {Abu Dhabi, UAE},
  publisher = {International Committee on Computational Linguistics},
  url       = {https://aclanthology.org/2025.chipsal-1.19/}
}

@article{esd,
  title     = {Emotional voice conversion: Theory, databases and ESD},
  author    = {Zhou, Kun and Sisman, Berrak and Liu, Rui and Li, Haizhou},
  journal   = {Speech Communication},
  volume    = {137},
  pages     = {1--18},
  year      = {2022},
  publisher = {Elsevier}
}

@inproceedings{fleurs,
  title     = {FLEURS: Few-shot Learning Evaluation of Universal
               Representations of Speech},
  author    = {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 = {IEEE Spoken Language Technology Workshop},
  year      = {2022}
}

@misc{IndoWaveSentiment,
  title        = {IndoWaveSentiment: Indonesian Audio Dataset for Emotion
                  Classification},
  author       = {Bustamin, Anugrayani and Rizky, Andi M. and Warni, Elly and
                  Sari Areni, Intan and Indrabayu, Indrabayu},
  year         = {2024},
  howpublished = {Mendeley Data, Version 1},
  publisher    = {Universitas Hasanuddin},
  doi          = {10.17632/j9ytfdzy27.1},
  url          = {https://data.mendeley.com/datasets/j9ytfdzy27/1}
}

@inproceedings{m3ed,
  title     = {M3ED: Multi-modal multi-scene multi-label emotional dialogue
               database},
  author    = {Zhao, Jinming and Zhang, Tenggan and Hu, Jingwen and
               Liu, Yuchen and Jin, Qin and Wang, Xinchao and Li, Haizhou},
  booktitle = {Annual Meeting of the Association for Computational
               Linguistics},
  year      = {2022}
}

@inproceedings{OpenSLR2,
  title     = {{Open-source Multi-speaker Speech Corpora for Building
               Gujarati, Kannada, Malayalam, Marathi, Tamil and Telugu
               Speech Synthesis Systems}},
  author    = {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 = {Proceedings of The 12th Language Resources and Evaluation
               Conference (LREC)},
  pages     = {6494--6503},
  month     = may,
  year      = {2020},
  address   = {Marseille, France},
  publisher = {European Language Resources Association (ELRA)},
  url       = {https://www.aclweb.org/anthology/2020.lrec-1.800}
}

@misc{sfdusc_ph,
  title        = {ASR-SFDuSC: A Scripted Filipino Daily-use Speech Corpus},
  author       = {{Magic Data Technology}},
  year         = {2023},
  howpublished = {MagicHub},
  url          = {https://magichub.com/datasets/filipino-scripted-speech-corpus-daily-use-sentence/}
}

@inproceedings{sg_streets,
  title     = {Enriching Rare Word Representations in Neural Language Models
               by Embedding Matrix Augmentation},
  author    = {Khassanov, Yerbolat and Zeng, Zhiping and Pham, Van Tung and
               Xu, Haihua and Chng, Eng Siong},
  booktitle = {Interspeech 2019},
  pages     = {3505--3509},
  month     = sep,
  year      = {2019},
  publisher = {ISCA},
  doi       = {10.21437/Interspeech.2019-1858}
}

@misc{SMalDuSC,
  title        = {{ASR-SMalDuSC}: A Scripted Malay Daily-use Speech Corpus},
  author       = {{Magic Data Technology}},
  howpublished = {\url{https://magichub.com/datasets/malay-scripted-speech-corpus-daily-use-sentence/}},
  year         = {2023}
}

@misc{tec,
  title        = {tamil-audio-emotion-classification},
  author       = {Thanushs25},
  year         = {2024},
  howpublished = {\url{https://huggingface.co/datasets/Thanushs25/tamil-audio-emotion-classification}}
}

@misc{thaielderly,
  title        = {Thai Elderly Speech dataset by Data Wow and VISAI},
  author       = {{VISAI AI Company Limited} and {Data Wow Company Limited}},
  year         = {2022},
  howpublished = {\url{https://github.com/VISAI-DATAWOW/Thai-Elderly-Speech-dataset/releases/tag/v1.0.0}}
}

@article{thaiser,
  title   = {THAI Speech Emotion Recognition (THAI-SER) corpus},
  author  = {Wongpithayadisai, Jilamika and Chaksangchaichot, Chompakorn and
             Sangnark, Soravitt and Prakrankamanant, Patawee and
             Gangwanpongpun, Krit and Boonpunmongkol, Siwa and
             Milindasuta, Premmarin and Na-Pombejra, Dangkamon and
             Nutanong, Sarana and Chuangsuwanich, Ekapol},
  journal = {arXiv preprint arXiv:2507.09618},
  year    = {2025}
}

@inproceedings{vietnam_celeb,
  title     = {{Vietnam-Celeb: a large-scale dataset for Vietnamese speaker
               recognition}},
  author    = {Pham, Viet Thanh and Nguyen, Xuan Thai Hoa and Hoang, Vu and
               Nguyen, Thi Thu Trang},
  booktitle = {Proc. INTERSPEECH 2023},
  pages     = {1918--1922},
  year      = {2023},
  doi       = {10.21437/Interspeech.2023-1989}
}