用一个模型搞定多部分口语评分，又快又准。
One Whisper to Grade Them All
- 单个Whisper-small模型处理全部口语回答，无需分题建模
- 测试误差仅0.384，比文本基线低13%，参数量仅168M
- 只需44.8%说话人数据就能达到高分，适合数据少的场景
我们提出一种高效端到端的全自动口语评估方法，用于多部分第二语言测试的综合评分，专为2025年Speak & Improve挑战设计。系统核心创新在于使用单一Whisper-small编码器处理全部四段口语回答，通过轻量级聚合器整合信息并预测总分。该架构无需转写和每部分单独建模，显著降低推理时间，使大规模语言学习系统中的自动评分成为可能。系统在测试中达到0.384的均方根误差（RMSE），优于文本基线的0.44，且参数量最多168M（约为Whisper-small的70%）。此外，我们提出一种数据采样策略，仅需44.8%的说话人数据即可达到0.383的RMSE，证明了对不平衡类别的改进及强大的数据效率。
原文摘要 · Abstract (English)
We present an efficient end-to-end approach for holistic Automatic Speaking Assessment (ASA) of multi-part second-language tests, developed for the 2025 Speak & Improve Challenge. Our system's main novelty is the ability to process all four spoken responses with a single Whisper-small encoder, combine all information via a lightweight aggregator, and predict the final score. This architecture removes the need for transcription and per-part models, cuts inference time, and makes ASA practical for large-scale Computer-Assisted Language Learning systems. Our system achieved a Root Mean Squared Error (RMSE) of 0.384, outperforming the text-based baseline (0.44) while using at most 168M parameters (about 70% of Whisper-small). Furthermore, we propose a data sampling strategy, allowing the model to train on only 44.8% of the speakers in the corpus and still reach 0.383 RMSE, demonstrating improved performance on imbalanced classes and strong data efficiency.
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。