通过分阶段训练提升语音模型的推理与自纠错能力，显著降低识别错误率。
Seewo's Submission to MLC-SLM: Lessons learned from Speech Reasoning Language Models
- 采用渐进式学习+思维链增强+可验证奖励强化，构建多阶段训练框架
- 在挑战测试集上，语音识别错误率降至11.57%，对话语音分割识别错率17.67%
- 适合关注语音理解中逻辑推理与自我修正机制的研究者和开发者
本文介绍了赛沃（Seewo）针对多语言对话语音语言模型挑战赛（MLC-SLM）两个赛道的系统方案，分别解决自动语音识别（ASR）与带说话人分割的语音识别（SD-ASR）问题。提出一种多阶段训练流程，显式增强语音语言模型在语音识别中的推理与自纠错能力。方法结合课程学习实现能力渐进提升、思维链数据增强促进中间反思、以及基于可验证奖励的强化学习（RLVR）实现奖励驱动的自纠错优化。实验结果表明，该方法在官方基准之上取得显著改进。在测试集上，最优系统在第一赛道获得11.57%的词错误率（WER）与字符错误率（CER），第二赛道获得17.67%的tcpWER与tcpCER。全面消融实验验证了各模块在挑战约束下的有效性。
原文摘要 · Abstract (English)
This paper presents Seewo's systems for both tracks of the Multilingual Conversational Speech Language Model Challenge (MLC-SLM), addressing automatic speech recognition (ASR) and speaker diarization with ASR (SD-ASR). We introduce a multi-stage training pipeline that explicitly enhances reasoning and self-correction in speech language models for ASR. Our approach combines curriculum learning for progressive capability acquisition, Chain-of-Thought data augmentation to foster intermediate reflection, and Reinforcement Learning with Verifiable Rewards (RLVR) to further refine self-correction through reward-driven optimization. This approach achieves substantial improvements over the official challenge baselines. On the evaluation set, our best system attains a WER/CER of 11.57% for Track 1 and a tcpWER/tcpCER of 17.67% for Track 2. Comprehensive ablation studies demonstrate the effectiveness of each component under challenge constraints.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。