{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,3,25]],"date-time":"2026-03-25T06:23:26Z","timestamp":1774419806813,"version":"3.50.1"},"reference-count":42,"publisher":"IEEE","license":[{"start":{"date-parts":[[2025,4,6]],"date-time":"2025-04-06T00:00:00Z","timestamp":1743897600000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-029"},{"start":{"date-parts":[[2025,4,6]],"date-time":"2025-04-06T00:00:00Z","timestamp":1743897600000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-037"}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"published-print":{"date-parts":[[2025,4,6]]},"DOI":"10.1109\/icassp49660.2025.10890533","type":"proceedings-article","created":{"date-parts":[[2025,3,12]],"date-time":"2025-03-12T13:52:43Z","timestamp":1741787563000},"page":"1-5","source":"Crossref","is-referenced-by-count":1,"title":["UME: Upcycling Mixture-of-Experts for Scalable and Efficient Automatic Speech Recognition"],"prefix":"10.1109","author":[{"given":"Li","family":"Fu","sequence":"first","affiliation":[{"name":"JD AI Research,Beijing,China"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Shanyong","family":"Yu","sequence":"additional","affiliation":[{"name":"JD AI Research,Beijing,China"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Siqi","family":"Li","sequence":"additional","affiliation":[{"name":"JD AI Research,Beijing,China"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Lu","family":"Fan","sequence":"additional","affiliation":[{"name":"JD AI Research,Beijing,China"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Youzheng","family":"Wu","sequence":"additional","affiliation":[{"name":"JD AI Research,Beijing,China"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Xiaodong","family":"He","sequence":"additional","affiliation":[{"name":"JD AI Research,Beijing,China"}],"role":[{"role":"author","vocabulary":"crossref"}]}],"member":"263","reference":[{"key":"ref1","article-title":"Attention is all you need","author":"Vaswani","year":"2017","journal-title":"Conference on Neural Information Processing Systems (NeurIPS)"},{"key":"ref2","doi-asserted-by":"publisher","DOI":"10.1561\/116.00000050"},{"key":"ref3","doi-asserted-by":"publisher","DOI":"10.1109\/TASLP.2023.3328283"},{"key":"ref4","doi-asserted-by":"publisher","DOI":"10.1109\/JSTSP.2022.3182537"},{"key":"ref5","doi-asserted-by":"publisher","DOI":"10.1109\/JSTSP.2022.3188113"},{"key":"ref6","article-title":"Pushing the limits of semi-supervised learning for automatic speech recognition","author":"Zhang","year":"2020","journal-title":"Conference on Neural Information Processing Systems (NeurIPS)"},{"key":"ref7","article-title":"Robust speech recognition via large-scale weak supervision","volume-title":"International Conference on Machine Learning (ICML)","author":"Radford"},{"key":"ref8","article-title":"Google usm: Scaling automatic speech recognition beyond 100 languages","author":"Zhang","year":"2023"},{"key":"ref9","doi-asserted-by":"publisher","DOI":"10.1145\/3700439"},{"key":"ref10","article-title":"Net2net: Accelerating learning via knowledge transfer","volume-title":"International Conference on Learning Representations (ICLR)","author":"Chen"},{"key":"ref11","article-title":"Data efficient neural scaling law via model reusing","volume-title":"International Conference on Machine Learning (ICML)","author":"Wang"},{"key":"ref12","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2022.emnlp-main.804"},{"key":"ref13","article-title":"Mixtral of experts","author":"Jiang","year":"2024"},{"key":"ref14","article-title":"Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context","author":"Reid","year":"2024"},{"key":"ref15","article-title":"Sparse upcycling: Training mixture-of-experts from dense checkpoints","volume-title":"International Conference on Learning Representations (ICLR)","author":"Komatsuzaki"},{"key":"ref16","article-title":"Moe jetpack: From dense checkpoints to adaptive mixture of experts for vision tasks","author":"Zhu","year":"2024"},{"key":"ref17","article-title":"Outrageously large neural networks: The sparsely-gated mixture-of-experts layer","volume-title":"International Conference on Learning Representations (ICLR)","author":"Shazeer"},{"key":"ref18","article-title":"Switch transformers: Scaling to trillion parameter models with simple and efficient sparsity","author":"Fedus","year":"2022","journal-title":"Journal of Machine Learning Research"},{"key":"ref19","doi-asserted-by":"publisher","DOI":"10.21437\/Interspeech.2021-478"},{"key":"ref20","doi-asserted-by":"publisher","DOI":"10.1109\/ICASSP43922.2022.9747065"},{"key":"ref21","doi-asserted-by":"publisher","DOI":"10.1109\/ISCSLP57327.2022.10037818"},{"key":"ref22","doi-asserted-by":"publisher","DOI":"10.21437\/Interspeech.2023-2292"},{"key":"ref23","doi-asserted-by":"publisher","DOI":"10.1109\/ASRU57964.2023.10389798"},{"key":"ref24","doi-asserted-by":"publisher","DOI":"10.1109\/ICASSP49357.2023.10096227"},{"key":"ref25","doi-asserted-by":"publisher","DOI":"10.21437\/Interspeech.2023-480"},{"key":"ref26","article-title":"U2++ moe: Scaling 4.7 x parameters with minimal impact on rtf","author":"Song","year":"2024"},{"key":"ref27","doi-asserted-by":"publisher","DOI":"10.21437\/Interspeech.2020-3015"},{"key":"ref28","doi-asserted-by":"publisher","DOI":"10.1109\/ICASSP43922.2022.9746594"},{"key":"ref29","doi-asserted-by":"publisher","DOI":"10.1109\/ASRU51503.2021.9687910"},{"key":"ref30","doi-asserted-by":"publisher","DOI":"10.1109\/ICASSP48485.2024.10445894"},{"key":"ref31","article-title":"Lora: Low-rank adaptation of large language models","volume-title":"International Conference on Learning Representations (ICLR)","author":"Hu"},{"key":"ref32","doi-asserted-by":"publisher","DOI":"10.1109\/ICASSP48485.2024.10447004"},{"key":"ref33","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2021.emnlp-main.541"},{"key":"ref34","doi-asserted-by":"publisher","DOI":"10.1109\/taslp.2021.3138674"},{"key":"ref35","doi-asserted-by":"publisher","DOI":"10.21437\/Interspeech.2023-1428"},{"key":"ref36","doi-asserted-by":"publisher","DOI":"10.21437\/Interspeech.2022-9996"},{"key":"ref37","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2024.acl-long.352"},{"key":"ref38","article-title":"Smartfrz: An efficient training framework using attention-based layer freezing","volume-title":"International Conference on Learning Representations (ICLR)","author":"Li"},{"key":"ref39","doi-asserted-by":"publisher","DOI":"10.1109\/ICASSP43922.2022.9746682"},{"key":"ref40","doi-asserted-by":"publisher","DOI":"10.21437\/Interspeech.2021-1397"},{"key":"ref41","doi-asserted-by":"publisher","DOI":"10.21437\/Interspeech.2021-1965"},{"key":"ref42","doi-asserted-by":"publisher","DOI":"10.1145\/1143844.1143891"}],"event":{"name":"ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)","location":"Hyderabad, India","start":{"date-parts":[[2025,4,6]]},"end":{"date-parts":[[2025,4,11]]}},"container-title":["ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)"],"original-title":[],"link":[{"URL":"http:\/\/xplorestaging.ieee.org\/ielx8\/10887540\/10887541\/10890533.pdf?arnumber=10890533","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,3,25]],"date-time":"2026-03-25T05:24:48Z","timestamp":1774416288000},"score":1,"resource":{"primary":{"URL":"https:\/\/ieeexplore.ieee.org\/document\/10890533\/"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2025,4,6]]},"references-count":42,"URL":"https:\/\/doi.org\/10.1109\/icassp49660.2025.10890533","relation":{},"subject":[],"published":{"date-parts":[[2025,4,6]]}}}