---
title: "ChildTalk: A Multi-Dialect Chinese Child Speech Corpus with Full-Length Child–Caregiver — Zhou 2026"
description: "Recognizing children's speech is hard for ASR systems. 25.39% CER. 29 references."
authors: ["Jiaming Zhou", "Yujie Guo", "Shiwan Zhao", "Yao Lu", "Jianye Wang", "Haoqin Sun", "Hui Wang", "Yong Qin"]
year: 2026
venue: "Findings of the Association for Computational Linguistics: ACL 2026"
source: "https://aclanthology.org/2026.findings-acl.251/"
doi: "10.18653/v1/2026.findings-acl.251"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/zhou-2026-childtalk-multi-dialect-chinese-child/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/zhou-2026-childtalk-multi-dialect-chinese-child/)
[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/5/) / ChildTalk: A Multi-Dialect Chinese Child Speech Corpus with Full-Length Child–Caregiver Conversations for Speech Recognition
# ChildTalk: A Multi-Dialect Chinese Child Speech Corpus with Full-Length Child–Caregiver Conversations for Speech Recognition
**Authors:** Jiaming Zhou, Yujie Guo, Shiwan Zhao, [Yao Lu](https://md.research.elamur.ai/free/en/authors/yao-lu/), Jianye Wang, Haoqin Sun, [Hui Wang](https://md.research.elamur.ai/free/en/authors/hui-wang/) and Yong Qin · **Year:** 2026 · **Venue:** Findings of the Association for Computational Linguistics: ACL 2026 · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)
**Links:** [Publisher's page](https://aclanthology.org/2026.findings-acl.251/) · [PDF](https://aclanthology.org/2026.findings-acl.251.pdf) · [DOI 10.18653/v1/2026.findings-acl.251](https://doi.org/10.18653/v1/2026.findings-acl.251) · [OpenAlex](https://openalex.org/W7166830745)
**Citations:** 0 citations (OpenAlex, 2026-10-14)
## Summary
Recognizing children's speech is hard for ASR systems. Child voices differ acoustically from the adult data most models are trained on, and Mandarin child corpora are scarce. Existing ones are often closed, built from read speech, cut into isolated utterances, or limited to standard Mandarin. The authors address this with ChildTalk, a public corpus of natural child–caregiver conversations. It includes 498 children aged 2–8, recorded with 500 familiar adults across mainland China. Each session is released as a complete dialogue with character-level transcriptions by professional annotators, so turn-taking and discourse context are kept. The corpus covers standard Mandarin, eight Mandarin dialect subgroups, Southern Min and Jin. The authors benchmark four system types: end-to-end models trained from scratch, fine-tuned pre-trained ASR models, zero-shot omni-modal LLMs and commercial transcription APIs. Off-the-shelf systems perform poorly on the data. Fine-tuning on ChildTalk helps on both in-domain and cross-domain tests. Feeding a few previous dialogue turns as text prompts further lowers error rates. The corpus is valuable as a benchmark for studying Chinese children's speech recognition, how well models handle dialects, and modeling at the level of whole conversations.
## Key points
- **Scale:** ChildTalk contains 112.5 hours and 88,943 utterances from 500 dyadic sessions. Speaker-independent splits are Train 79.06 h, Dev 14.46 h and Test 18.98 h. Mandarin and dialect speakers do not overlap.
- **Transcription:** Transcripts use a rich set of special tokens, such as ``, ``, ``, ``, `` and `#` for fillers. Disfluencies and repetitions are kept.
- **Zero-shot results:** On the test set, Qwen2.5-Omni scores 25.39% CER and Qwen3-Omni 28.56%. GPT-4o-transcribe reaches 47.44% and GPT-4o-mini-transcribe 48.65%.
- **Dialogue context:** For Whisper-Large-V3, CER falls from 43.25% with no context to 31.87% with three preceding turns. Most of the gain comes from fewer substitutions and deletions. A similar trend holds for Qwen-3-ASR.
- **Access:** The corpus is available at github.com/NKU-HLT/ChildTalk for non-commercial research under a data use agreement. The dialect and age distributions are skewed: Mandarin, Southwestern Mandarin and 7–8-year-olds are overrepresented.
*Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.*
Fields: [Natural language processing](https://md.research.elamur.ai/free/en/shelves/nlp/)
Topics: [🎙️ Speech & spoken language](https://md.research.elamur.ai/free/en/topics/speech/)
## Institutions
- [Nankai University](https://md.research.elamur.ai/free/en/universities/nankai-university/) — University, Tianjin, China · 2 works · 624 citations
## Cite
Cite as: Jiaming Zhou, Yujie Guo, Shiwan Zhao, Yao Lu, Jianye Wang, Haoqin Sun, Hui Wang and Yong Qin. 2026. ChildTalk: A Multi-Dialect Chinese Child Speech Corpus with Full-Length Child–Caregiver Conversations for Speech Recognition. Findings of the Association for Computational Linguistics: ACL 2026.
The paper itself, not this page. BibTeX from the library's record of the work.
```bibtex
@inproceedings{zhou2026childtalk,
title = {{ChildTalk: A Multi-Dialect Chinese Child Speech Corpus with Full-Length Child–Caregiver Conversations for Speech Recognition}},
author = {Jiaming Zhou, Yujie Guo, Shiwan Zhao, Yao Lu, Jianye Wang, Haoqin Sun, Hui Wang and Yong Qin},
booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
year = {2026},
doi = {10.18653/v1/2026.findings-acl.251},
url = {https://doi.org/10.18653/v1/2026.findings-acl.251}
}
```
- [← previous Child-directed speech facilitates production, not comprehension, in BabyLMs](https://md.research.elamur.ai/free/en/papers/bunzeck-2026-child-directed-speech-facilitates-production/)
- [next → Chinese Coreference Corpus from Verb-Driven Events and Its Application for LLM Evaluation](https://md.research.elamur.ai/free/en/papers/wang-2026-chinese-coreference-corpus/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/5/)
## References
29 references, 8 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.
1. Keyu An, Qian Chen, Chong Deng, Zhihao Du, Changfeng Gao, Zhifu Gao, Yue Gu, Ting He, Hangrui Hu, Kai Hu, Shengpeng Ji, Yabin Li, Zerui Li, Heng Lu, Xiang Lv, Bin Ma, Ziyang Ma, Chongjia Ni, Changhe Song, and 13 others. 2024. Funaudiollm: Voice understanding and generation foundation models for natu… — [arXiv:2407.04051](https://arxiv.org/abs/2407.04051) [find](https://scholar.google.com/scholar?q=Keyu+An%2C+Qian+Chen%2C+Chong+Deng%2C+Zhihao+Du%2C+Changfeng+Gao%2C+Zhifu+Gao%2C+Yue+Gu%2C+Ting+He%2C+Hangrui+Hu%2C+Kai+Hu%2C+Shengpeng+Ji%2C+Yabin+Li%2C+Zerui+Li%2C+Heng+Lu%2C+Xiang+Lv%2C+Bin+Ma%2C+Ziyang+Ma%2C+Chongjia+Ni%2C+Changhe+S)
2. Alexei Baevski, Yuhao Zhou, Abdelrahman Mohamed, and Michael Auli. 2020. wav2vec 2.0: A framework for self-supervised learning of speech representations. Advances in neural information processing systems, 33:12449–12460. [our card](https://md.research.elamur.ai/free/en/papers/baevski-2020-wav2vec-2-0-framework-self/)
3. Linda Bell, Johan Boye, Joakim Gustafson, Mattias Heldner, Anders Lindström, and Mats Wirén. 2005. The swedish nice corpus–spoken dialogues between children and embodied characters in a computer game scenario. In Interspeech 2005-Eurospeech, 9th European Conference on Speech Communication and Techno… [find](https://scholar.google.com/scholar?q=Linda+Bell%2C+Johan+Boye%2C+Joakim+Gustafson%2C+Mattias+Heldner%2C+Anders+Lindstr%C3%B6m%2C+and+Mats+Wir%C3%A9n%2E+2005%2E+The+swedish+nice+corpus%E2%80%93spoken+dialogues+between+children+and+embodied+characters+in+a+computer+game+)
4. Nancy F Chen, Rong Tong, Darren Wee, Pei Xuan Lee, Bin Ma, and Haizhou Li. 2016. Singakids-mandarin: Speech corpus of singaporean children speaking mandarin chinese. In Interspeech, pages 1545–1549. [find](https://scholar.google.com/scholar?q=Nancy+F+Chen%2C+Rong+Tong%2C+Darren+Wee%2C+Pei+Xuan+Lee%2C+Bin+Ma%2C+and+Haizhou+Li%2E+2016%2E+Singakids%2Dmandarin%3A+Speech+corpus+of+singaporean+children+speaking+mandarin+chinese%2E+In+Interspeech%2C+pages+1545%E2%80%931549%2E)
5. Catia Cucchiarini, Joris Driesen, H Van Hamme, and EP Sanders. 2008. Recording speech of children, non-natives and elderly people for hlt applications: the jasmin-cgn corpus. [find](https://scholar.google.com/scholar?q=Catia+Cucchiarini%2C+Joris+Driesen%2C+H+Van+Hamme%2C+and+EP+Sanders%2E+2008%2E+Recording+speech+of+children%2C+non%2Dnatives+and+elderly+people+for+hlt+applications%3A+the+jasmin%2Dcgn+corpus%2E)
6. Katherine Demuth and Annie Tremblay. 2008. Prosodically-conditioned variability in children’s production of french determiners. Journal of child language, 35(1):99–127. [find](https://scholar.google.com/scholar?q=Katherine+Demuth+and+Annie+Tremblay%2E+2008%2E+Prosodically%2Dconditioned+variability+in+children%E2%80%99s+production+of+french+determiners%2E+Journal+of+child+language%2C+35%281%29%3A99%E2%80%93127%2E)
7. Linhao Dong, Shuang Xu, and Bo Xu. 2018. Speech-transformer: a no-recurrence sequence-to-sequence model for speech recognition. In 2018 IEEE international conference on acoustics, speech and signal processing (ICASSP), pages 5884–5888. IEEE. [find](https://scholar.google.com/scholar?q=Linhao+Dong%2C+Shuang+Xu%2C+and+Bo+Xu%2E+2018%2E+Speech%2Dtransformer%3A+a+no%2Drecurrence+sequence%2Dto%2Dsequence+model+for+speech+recognition%2E+In+2018+IEEE+international+conference+on+acoustics%2C+speech+and+signal+pr)
8. Ruchao Fan, Natarajan Balaji Shankar, and Abeer Alwan. 2024. Benchmarking children’s asr with supervised and self-supervised speech foundation models. In Interspeech 2024, pages 5173–5177. [find](https://scholar.google.com/scholar?q=Ruchao+Fan%2C+Natarajan+Balaji+Shankar%2C+and+Abeer+Alwan%2E+2024%2E+Benchmarking+children%E2%80%99s+asr+with+supervised+and+self%2Dsupervised+speech+foundation+models%2E+In+Interspeech+2024%2C+pages+5173%E2%80%935177%2E)
9. Jun Gao, Aijun Li, and Ziyu Xiong. 2012. Mandarin multimedia child speech corpus: Cass_child. In 2012 International Conference on Speech Database and Assessments, pages 7–12. IEEE. [find](https://scholar.google.com/scholar?q=Jun+Gao%2C+Aijun+Li%2C+and+Ziyu+Xiong%2E+2012%2E+Mandarin+multimedia+child+speech+corpus%3A+Cass%5Fchild%2E+In+2012+International+Conference+on+Speech+Database+and+Assessments%2C+pages+7%E2%80%9312%2E+IEEE%2E)
10. Marta Garrote and A Moreno Sandoval. 2008. Chiede, a spontaneous child language corpus of spanish. In Proceedings of the 3rd International LABLITA Workshop in Corpus Linguistics. [find](https://scholar.google.com/scholar?q=Marta+Garrote+and+A+Moreno+Sandoval%2E+2008%2E+Chiede%2C+a+spontaneous+child+language+corpus+of+spanish%2E+In+Proceedings+of+the+3rd+International+LABLITA+Workshop+in+Corpus+Linguistics%2E)
11. Matteo Gerosa, Diego Giuliani, Shrikanth Narayanan, and Alexandros Potamianos. 2009. A review of asr technologies for children’s speech. In Proceedings of the 2nd Workshop on Child, Computer and Interaction, pages 1–8. [find](https://scholar.google.com/scholar?q=Matteo+Gerosa%2C+Diego+Giuliani%2C+Shrikanth+Narayanan%2C+and+Alexandros+Potamianos%2E+2009%2E+A+review+of+asr+technologies+for+children%E2%80%99s+speech%2E+In+Proceedings+of+the+2nd+Workshop+on+Child%2C+Computer+and+Inter)
12. Anmol Gulati, James Qin, Chung-Cheng Chiu, Niki Parmar, Yu Zhang, Jiahui Yu, Wei Han, Shibo Wang, Zhengdong Zhang, Yonghui Wu, and 1 others. 2020. Conformer: Convolution-augmented transformer for speech recognition. arXiv preprint arXiv:2005.08100. a young children’s speech corpus. In Interspeech, p… [our card](https://md.research.elamur.ai/free/en/papers/gulati-2020-conformer-convolution-augmented-transformer-speech/)
13. James Kennedy, Séverin Lemaignan, Caroline Montassier, Pauline Lavalade, Bahar Irfan, Fotios Papadopoulos, Emmanuel Senft, and Tony Belpaeme. 2017. Child speech recognition in human-robot interaction: evaluations and recommendations. In Proceedings of the 2017 ACM/IEEE international conference on hu… [find](https://scholar.google.com/scholar?q=James+Kennedy%2C+S%C3%A9verin+Lemaignan%2C+Caroline+Montassier%2C+Pauline+Lavalade%2C+Bahar+Irfan%2C+Fotios+Papadopoulos%2C+Emmanuel+Senft%2C+and+Tony+Belpaeme%2E+2017%2E+Child+speech+recognition+in+human%2Drobot+interaction%3A)
14. Sungbok Lee, Alexandros Potamianos, and Shrikanth Narayanan. 1997. Analysis of children’s speech: Duration, pitch and formants. In Fifth European Conference on Speech Communication and Technology. [find](https://scholar.google.com/scholar?q=Sungbok+Lee%2C+Alexandros+Potamianos%2C+and+Shrikanth+Narayanan%2E+1997%2E+Analysis+of+children%E2%80%99s+speech%3A+Duration%2C+pitch+and+formants%2E+In+Fifth+European+Conference+on+Speech+Communication+and+Technology%2E)
15. W. S. Leung. 2018. A corpus-based study of the acquisition of post-verbal particles by cantonese-speaking children aged 2;5-5;8. Thesis, The University of Hong Kong, Pokfulam, Hong Kong SAR. [find](https://scholar.google.com/scholar?q=W%2E+S%2E+Leung%2E+2018%2E+A+corpus%2Dbased+study+of+the+acquisition+of+post%2Dverbal+particles+by+cantonese%2Dspeaking+children+aged+2%3B5%2D5%3B8%2E+Thesis%2C+The+University+of+Hong+Kong%2C+Pokfulam%2C+Hong+Kong+SAR%2E)
16. Z. Mai, M. Shang, J. Liu, S. Yan, S. Matthews, and V. Yip. 2024. Acquiring chinese in us, hong kong and beijing: three new corpora and three verbal structures. Paper presented at the XVIth International Congress for the Study of Child Language (IASCL-2024). July 15–19. [find](https://scholar.google.com/scholar?q=Z%2E+Mai%2C+M%2E+Shang%2C+J%2E+Liu%2C+S%2E+Yan%2C+S%2E+Matthews%2C+and+V%2E+Yip%2E+2024%2E+Acquiring+chinese+in+us%2C+hong+kong+and+beijing%3A+three+new+corpora+and+three+verbal+structures%2E+Paper+presented+at+the+XVIth+Internation)
17. OpenAI, :, Aaron Hurst, Adam Lerer, Adam P. Goucher, Adam Perelman, Aditya Ramesh, and Aidan Clark et al. 2024. Gpt-4o system card. Preprint, arXiv:2410.21276. [our card](https://md.research.elamur.ai/free/en/papers/openai-2024-gpt-4o-system-card/)
18. Yifan Peng, Siddharth Dalmia, Ian Lane, and Shinji Watanabe. 2022. Branchformer: Parallel mlp-attention architectures to capture local and global context for speech recognition and understanding. In International Conference on Machine Learning, pages 17627–17643. PMLR. [find](https://scholar.google.com/scholar?q=Yifan+Peng%2C+Siddharth+Dalmia%2C+Ian+Lane%2C+and+Shinji+Watanabe%2E+2022%2E+Branchformer%3A+Parallel+mlp%2Dattention+architectures+to+capture+local+and+global+context+for+speech+recognition+and+understanding%2E+In+I)
19. Humberto Pérez-Espinosa, Juan Martínez-Miranda, Ismael Espinosa-Curiel, Josefina Rodríguez-Jacobo, Luis Villaseñor-Pineda, and Himer Avila-George. 2020. Iesc-child: an interactive emotional children’s speech corpus. Computer Speech & Language, 59:55–74. [find](https://scholar.google.com/scholar?q=Humberto+P%C3%A9rez%2DEspinosa%2C+Juan+Mart%C3%ADnez%2DMiranda%2C+Ismael+Espinosa%2DCuriel%2C+Josefina+Rodr%C3%ADguez%2DJacobo%2C+Luis+Villase%C3%B1or%2DPineda%2C+and+Himer+Avila%2DGeorge%2E+2020%2E+Iesc%2Dchild%3A+an+interactive+emotional+children%E2%80%99s)
20. Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, and Ilya Sutskever. 2023a. Robust speech recognition via large-scale weak supervision. In International conference on machine learning, pages 28492–28518. PMLR. [our card](https://md.research.elamur.ai/free/en/papers/radford-2022-robust-speech-recognition-large-scale/)
21. Wei-Ning Hsu, Benjamin Bolte, Yao-Hung Hubert Tsai, Kushal Lakhotia, Ruslan Salakhutdinov, and Abdelrahman Mohamed. 2021. Hubert: Self-supervised speech representation learning by masked prediction of hidden units. IEEE/ACM transactions on audio, speech, and language processing, 29:3451–3460. [our card](https://md.research.elamur.ai/free/en/papers/hsu-2021-hubert-self-supervised-speech-representation/)
22. Alec Radford, Jong Wook Kim, Tao Xu, Greg Brockman, Christine McLeavey, and Ilya Sutskever. 2023b. Robust speech recognition via large-scale weak supervision. In International conference on machine learning, pages 28492–28518. PMLR. [our card](https://md.research.elamur.ai/free/en/papers/radford-2022-robust-speech-recognition-large-scale/)
23. Abe Kazemzadeh, Hong You, Markus Iseli, Barbara Jones, Xiaodong Cui, Margaret Heritage, Patti Price, Elaine Andersen, Shrikanth S Narayanan, and Abeer Alwan. 2005. Tball data collection: the making of Kodali Radha and Mohan Bansal. 2022. Audio augmentation for non-native children’s speech recognitio… [find](https://scholar.google.com/scholar?q=Abe+Kazemzadeh%2C+Hong+You%2C+Markus+Iseli%2C+Barbara+Jones%2C+Xiaodong+Cui%2C+Margaret+Heritage%2C+Patti+Price%2C+Elaine+Andersen%2C+Shrikanth+S+Narayanan%2C+and+Abeer+Alwan%2E+2005%2E+Tball+data+collection%3A+the+making+of)
24. Xian Shi, Xiong Wang, Zhifang Guo, Yongqi Wang, Pei Zhang, Xinyu Zhang, Zishan Guo, Hongkun Hao, Yu Xi, Baosong Yang, and 1 others. 2026. Qwen3-asr technical report. arXiv preprint arXiv:2601.21337. [find](https://scholar.google.com/scholar?q=Xian+Shi%2C+Xiong+Wang%2C+Zhifang+Guo%2C+Yongqi+Wang%2C+Pei+Zhang%2C+Xinyu+Zhang%2C+Zishan+Guo%2C+Hongkun+Hao%2C+Yu+Xi%2C+Baosong+Yang%2C+and+1+others%2E+2026%2E+Qwen3%2Dasr+technical+report%2E+arXiv+preprint+arXiv%3A2601%2E21337%2E)
25. Deng Xiangjun and Virginia Yip. 2017. A multimedia corpus of child mandarin: The tong corpus. Journal of Chinese Linguistics. [find](https://scholar.google.com/scholar?q=Deng+Xiangjun+and+Virginia+Yip%2E+2017%2E+A+multimedia+corpus+of+child+mandarin%3A+The+tong+corpus%2E+Journal+of+Chinese+Linguistics%2E)
26. Jin Xu, Zhifang Guo, Jinzheng He, Hangrui Hu, Ting He, Shuai Bai, Keqin Chen, Jialin Wang, Yang Fan, Kai Dang, and 1 others. 2025a. Qwen2. 5-omni technical report. arXiv preprint arXiv:2503.20215. [our card](https://md.research.elamur.ai/free/en/papers/xu-2025-qwen2-5-omni-technical/)
27. Jin Xu, Zhifang Guo, Hangrui Hu, Yunfei Chu, Xiong Wang, Jinzheng He, Yuxuan Wang, Xian Shi, Ting He, Xinfa Zhu, Yuanjun Lv, Yongqi Wang, Dake Guo, He Wang, Linhan Ma, Pei Zhang, Xinyu Zhang, Hongkun Hao, Zishan Guo, and 19 others. 2025b. Qwen3-omni technical report. arXiv preprint arXiv:2509.17765. [our card](https://md.research.elamur.ai/free/en/papers/xu-2025-qwen3-omni-technical-report/)
28. Zhuoyuan Yao, Di Wu 0061, Xiong Wang, Binbin Zhang, Fan Yu, Chao Yang, Zhendong Peng, Xiaoyu Chen, Lei Xie, and Xin Lei. 2021. Wenet: Production oriented streaming and non-streaming end-to-end speech recognition toolkit. In interspeech, volume 2021, pages 4054–4058. [find](https://scholar.google.com/scholar?q=Zhuoyuan+Yao%2C+Di+Wu+0061%2C+Xiong+Wang%2C+Binbin+Zhang%2C+Fan+Yu%2C+Chao+Yang%2C+Zhendong+Peng%2C+Xiaoyu+Chen%2C+Lei+Xie%2C+and+Xin+Lei%2E+2021%2E+Wenet%3A+Production+oriented+streaming+and+non%2Dstreaming+end%2Dto%2Dend+speech+)
29. Fan Yu, Zhuoyuan Yao, Xiong Wang, Keyu An, Lei Xie, Zhijian Ou, Bo Liu, Xiulin Li, and Guanqiong Miao. 2021. The slt 2021 children speech recognition challenge: Open datasets, rules and baselines. In 2021 IEEE Spoken Language Technology Workshop (SLT), pages 1117–1123. IEEE. [find](https://scholar.google.com/scholar?q=Fan+Yu%2C+Zhuoyuan+Yao%2C+Xiong+Wang%2C+Keyu+An%2C+Lei+Xie%2C+Zhijian+Ou%2C+Bo+Liu%2C+Xiulin+Li%2C+and+Guanqiong+Miao%2E+2021%2E+The+slt+2021+children+speech+recognition+challenge%3A+Open+datasets%2C+rules+and+baselines%2E+In+)