Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Lin, Bin, Zhao, Bo, Wu, Boyong, Yan, Chao, Wu, Chen, Yi, Cheng, Yao, Chengyuan, Liu, Daijiao, Tian, Fei, Tian, Feng, Sun, Haiyang, Zhang, Haoyang, Zhen, Jiangjie, Gong, Jinglan, Chen, Jun, Xie, Li, Li, Peilin, Yang, Peng, Tan, Pengfei, Lin, Qingjian, Li, Runze, Hu, Shenghua, Zhou, Siyi, Qu, Wenwen, Li, Xiangyu, Zhang, Xiangyu Tony, Yang, Xuerui, Yang, Yang, Huang, Yechang, Fu, Yu, Luo, Yuchu, Li, Yuxin, Zhang, Yuxin, Sheng, Zhengyan, Li, Brian, Zeng, Chang, Zhang, Changlin, Geng, Chen, Dong, Chenghao, Feng, Chengli, Zhou, Dan, Wan, Danni, Chen, Di, Zhang, Die, Pang, Dongqing, Yang, Guanglong, Hu, Guoqiang, Zhu, Huangxi, Gao, Jianzheng, Liang, Jinghua, Wan, Jinmei, Yuan, Junjie, An, Kang, Lei, Lei, Zhong, Limin, Cai, Lun, Ren, Mengqiang, Xu, Min, Li, Mingliang, Li, Mingxiao, Wang, Na, Tong, Qiang, Huang, Qiaoling, Du, Qingfu, Wang, Rui, Zhou, Shengchen, Qiu, Shi, Peng, Shihao, Yang, Shiliang, Tu, Siqi, Deng, Tianjiao, Xu, Ting, Wang, Tong, Niu, WeiMing, Xie, Wuxun, Zhang, Xianwei, Feng, Xianyu, Liu, Xiaojia, Chen, Xing, Wu, Xiongbin, Wu, Yan, Li, Yang, Liu, Yi, Zhang, Yifan, Liu, Yile, Long, Yongshen, Luo, Yu, Ding, Yuanhao, Wang, Yuhao, Yin, Yuhe, Xu, Yunfang, Yang, Yuxiang, Huang, Zhiguo, Wu, Zhiyue, Li, Zichao, Zhou, Zichao, Jiang, Daxin, Li, Future, Yu, Gang, Zhang, Xiangyu, Zhu, Yibo
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2605.23463
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866910247738671104
author Lin, Bin
Zhao, Bo
Wu, Boyong
Yan, Chao
Wu, Chen
Yi, Cheng
Yao, Chengyuan
Liu, Daijiao
Tian, Fei
Tian, Feng
Sun, Haiyang
Zhang, Haoyang
Zhen, Jiangjie
Gong, Jinglan
Chen, Jun
Xie, Li
Li, Peilin
Yang, Peng
Tan, Pengfei
Lin, Qingjian
Li, Runze
Hu, Shenghua
Zhou, Siyi
Qu, Wenwen
Li, Xiangyu
Zhang, Xiangyu Tony
Yang, Xuerui
Yang, Yang
Huang, Yechang
Fu, Yu
Luo, Yuchu
Li, Yuxin
Zhang, Yuxin
Sheng, Zhengyan
Li, Brian
Zeng, Chang
Zhang, Changlin
Geng, Chen
Dong, Chenghao
Feng, Chengli
Zhou, Dan
Wan, Danni
Chen, Di
Zhang, Die
Pang, Dongqing
Yang, Guanglong
Hu, Guoqiang
Zhu, Huangxi
Gao, Jianzheng
Liang, Jinghua
Wan, Jinmei
Yuan, Junjie
An, Kang
Lei, Lei
Zhong, Limin
Cai, Lun
Ren, Mengqiang
Xu, Min
Li, Mingliang
Li, Mingxiao
Wang, Na
Tong, Qiang
Huang, Qiaoling
Du, Qingfu
Wang, Rui
Zhou, Shengchen
Qiu, Shi
Peng, Shihao
Yang, Shiliang
Tu, Siqi
Deng, Tianjiao
Xu, Ting
Wang, Tong
Niu, WeiMing
Xie, Wuxun
Zhang, Xianwei
Feng, Xianyu
Liu, Xiaojia
Chen, Xing
Wu, Xiongbin
Wu, Yan
Li, Yang
Liu, Yi
Zhang, Yifan
Liu, Yile
Long, Yongshen
Luo, Yu
Ding, Yuanhao
Wang, Yuhao
Yin, Yuhe
Xu, Yunfang
Yang, Yuxiang
Huang, Zhiguo
Wu, Zhiyue
Li, Zichao
Zhou, Zichao
Jiang, Daxin
Li, Future
Yu, Gang
Zhang, Xiangyu
Zhu, Yibo
author_facet Lin, Bin
Zhao, Bo
Wu, Boyong
Yan, Chao
Wu, Chen
Yi, Cheng
Yao, Chengyuan
Liu, Daijiao
Tian, Fei
Tian, Feng
Sun, Haiyang
Zhang, Haoyang
Zhen, Jiangjie
Gong, Jinglan
Chen, Jun
Xie, Li
Li, Peilin
Yang, Peng
Tan, Pengfei
Lin, Qingjian
Li, Runze
Hu, Shenghua
Zhou, Siyi
Qu, Wenwen
Li, Xiangyu
Zhang, Xiangyu Tony
Yang, Xuerui
Yang, Yang
Huang, Yechang
Fu, Yu
Luo, Yuchu
Li, Yuxin
Zhang, Yuxin
Sheng, Zhengyan
Li, Brian
Zeng, Chang
Zhang, Changlin
Geng, Chen
Dong, Chenghao
Feng, Chengli
Zhou, Dan
Wan, Danni
Chen, Di
Zhang, Die
Pang, Dongqing
Yang, Guanglong
Hu, Guoqiang
Zhu, Huangxi
Gao, Jianzheng
Liang, Jinghua
Wan, Jinmei
Yuan, Junjie
An, Kang
Lei, Lei
Zhong, Limin
Cai, Lun
Ren, Mengqiang
Xu, Min
Li, Mingliang
Li, Mingxiao
Wang, Na
Tong, Qiang
Huang, Qiaoling
Du, Qingfu
Wang, Rui
Zhou, Shengchen
Qiu, Shi
Peng, Shihao
Yang, Shiliang
Tu, Siqi
Deng, Tianjiao
Xu, Ting
Wang, Tong
Niu, WeiMing
Xie, Wuxun
Zhang, Xianwei
Feng, Xianyu
Liu, Xiaojia
Chen, Xing
Wu, Xiongbin
Wu, Yan
Li, Yang
Liu, Yi
Zhang, Yifan
Liu, Yile
Long, Yongshen
Luo, Yu
Ding, Yuanhao
Wang, Yuhao
Yin, Yuhe
Xu, Yunfang
Yang, Yuxiang
Huang, Zhiguo
Wu, Zhiyue
Li, Zichao
Zhou, Zichao
Jiang, Daxin
Li, Future
Yu, Gang
Zhang, Xiangyu
Zhu, Yibo
contents Unified audio-language modeling has emerged as a prominent trend in modern speech systems, promising to bring the reasoning capabilities of large language models to auditory tasks. However, existing unified foundations often struggle to match the depth of specialized systems across automatic speech recognition (ASR), text-to-speech synthesis (TTS), and realtime spoken interaction. Bridging this gap remains an open challenge. This report presents StepAudio 2.5, a unified audio-language foundation model that matches or exceeds specialized systems across all three capabilities. Rather than treating these tasks as architecturally distinct, we operate on the premise that once text and audio share a multimodal representational space, task specialization becomes a matter of operational regimes: data construction, optimization targets, and decoding constraints. Guided by this insight, we advance the post-training paradigm from standard supervised learning to task-tailored Reinforcement Learning from Human Feedback (RLHF), using it as the primary mechanism to define complex optimization targets. We leverage this RLHF-centric alignment, alongside specialized decoding, to shape a shared backbone into three distinct operational modes. Concretely, the ASR branch advances transcription efficiency via verifiable multi-token decoding; the TTS branch achieves controllable, expressive synthesis through preference-based RLHF and context-rich supervision; and the Realtime branch realizes low-latency, persona-consistent dialogue via generative reward modeling within an RLHF framework. On standard benchmarks, StepAudio 2.5 achieves state-of-the-art results across ASR, TTS, and Realtime, demonstrating that a singular audio-language foundation can successfully internalize the distinct deployment objectives of speech understanding, generation, and live interaction.
format Preprint
id arxiv_https___arxiv_org_abs_2605_23463
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle StepAudio 2.5 Technical Report
Lin, Bin
Zhao, Bo
Wu, Boyong
Yan, Chao
Wu, Chen
Yi, Cheng
Yao, Chengyuan
Liu, Daijiao
Tian, Fei
Tian, Feng
Sun, Haiyang
Zhang, Haoyang
Zhen, Jiangjie
Gong, Jinglan
Chen, Jun
Xie, Li
Li, Peilin
Yang, Peng
Tan, Pengfei
Lin, Qingjian
Li, Runze
Hu, Shenghua
Zhou, Siyi
Qu, Wenwen
Li, Xiangyu
Zhang, Xiangyu Tony
Yang, Xuerui
Yang, Yang
Huang, Yechang
Fu, Yu
Luo, Yuchu
Li, Yuxin
Zhang, Yuxin
Sheng, Zhengyan
Li, Brian
Zeng, Chang
Zhang, Changlin
Geng, Chen
Dong, Chenghao
Feng, Chengli
Zhou, Dan
Wan, Danni
Chen, Di
Zhang, Die
Pang, Dongqing
Yang, Guanglong
Hu, Guoqiang
Zhu, Huangxi
Gao, Jianzheng
Liang, Jinghua
Wan, Jinmei
Yuan, Junjie
An, Kang
Lei, Lei
Zhong, Limin
Cai, Lun
Ren, Mengqiang
Xu, Min
Li, Mingliang
Li, Mingxiao
Wang, Na
Tong, Qiang
Huang, Qiaoling
Du, Qingfu
Wang, Rui
Zhou, Shengchen
Qiu, Shi
Peng, Shihao
Yang, Shiliang
Tu, Siqi
Deng, Tianjiao
Xu, Ting
Wang, Tong
Niu, WeiMing
Xie, Wuxun
Zhang, Xianwei
Feng, Xianyu
Liu, Xiaojia
Chen, Xing
Wu, Xiongbin
Wu, Yan
Li, Yang
Liu, Yi
Zhang, Yifan
Liu, Yile
Long, Yongshen
Luo, Yu
Ding, Yuanhao
Wang, Yuhao
Yin, Yuhe
Xu, Yunfang
Yang, Yuxiang
Huang, Zhiguo
Wu, Zhiyue
Li, Zichao
Zhou, Zichao
Jiang, Daxin
Li, Future
Yu, Gang
Zhang, Xiangyu
Zhu, Yibo
Audio and Speech Processing
Unified audio-language modeling has emerged as a prominent trend in modern speech systems, promising to bring the reasoning capabilities of large language models to auditory tasks. However, existing unified foundations often struggle to match the depth of specialized systems across automatic speech recognition (ASR), text-to-speech synthesis (TTS), and realtime spoken interaction. Bridging this gap remains an open challenge. This report presents StepAudio 2.5, a unified audio-language foundation model that matches or exceeds specialized systems across all three capabilities. Rather than treating these tasks as architecturally distinct, we operate on the premise that once text and audio share a multimodal representational space, task specialization becomes a matter of operational regimes: data construction, optimization targets, and decoding constraints. Guided by this insight, we advance the post-training paradigm from standard supervised learning to task-tailored Reinforcement Learning from Human Feedback (RLHF), using it as the primary mechanism to define complex optimization targets. We leverage this RLHF-centric alignment, alongside specialized decoding, to shape a shared backbone into three distinct operational modes. Concretely, the ASR branch advances transcription efficiency via verifiable multi-token decoding; the TTS branch achieves controllable, expressive synthesis through preference-based RLHF and context-rich supervision; and the Realtime branch realizes low-latency, persona-consistent dialogue via generative reward modeling within an RLHF framework. On standard benchmarks, StepAudio 2.5 achieves state-of-the-art results across ASR, TTS, and Realtime, demonstrating that a singular audio-language foundation can successfully internalize the distinct deployment objectives of speech understanding, generation, and live interaction.
title StepAudio 2.5 Technical Report
topic Audio and Speech Processing
url https://arxiv.org/abs/2605.23463