arXiv cs.CL· Wanyi Ning, Yinshang Guo, Haitao Qian, Jiyuan Cheng, Wei Zhou, Weiyuan Feng, Yufei Zhang·· 13 小时前AI 评分32
FormalASR：端到端中文口语转正式文本，0.6B 与 1.7B 两个版本
FormalASR: End-to-End Spoken Chinese to Formal Text
AI 导读
FormalASR 推出 0.6B 和 1.7B 两个紧凑端到端模型，可将中文口语直接转写为正式书面文本，部署时无需后处理 LLM。团队基于 LLM 改写与质量过滤构建 WenetSpeech-Formal 和 Speechio-Formal 两个大规模数据集，并对 Qwen3-ASR 进行监督微调。
正文 · 原文
Abstract:Automatic speech recognition (ASR) systems are typically optimized for verbatim transcription, which preserves disfluencies, filler words, and informal spoken structures that are often unsuitable for downstream writing-oriented applications. A common workaround is a two-stage ASR+LLM pipeline for post-editing, but this design increases latency and memory cost and is difficult to deploy on-device. We present FormalASR, two compact end-to-end models with 0.6B and 1.7B parameters that directly transcribe spoken Chinese into formal written text. To enable this setting, we build WenetSpeech-Formal and Speechio-Formal, two large-scale spoken-to-formal datasets constructed by LLM-based rewriting and quality filtering. We then fine-tune Qwen3-ASR at both model scales with supervised fine-tuning. Experiments on WenetSpeech-Formal and Speechio-Formal show that FormalASR achieves up to 37.4% relative CER reduction over verbatim baselines, while also improving ROUGE-L and BERTScore. FormalASR requires no post-processing LLM at deployment time, providing a lightweight, on-device solution for spoken-to-formal transcription.
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2605.19266 [cs.CL] |
| (or arXiv:2605.19266v4 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2605.19266 arXiv-issued DOI via DataCite |
Submission history
From: Wanyi Ning [view email]
[v1]
Tue, 19 May 2026 02:27:27 UTC (1,244 KB)
[v2]
Sat, 6 Jun 2026 08:09:32 UTC (1,244 KB)
[v3]
Fri, 26 Jun 2026 09:28:46 UTC (1,244 KB)
[v4]
Thu, 22 Oct 2026 08:03:44 UTC (1,319 KB)
来源：arXiv cs.CL · arxiv.org