Error Explainable Benchmark (EEB) dataset

Error Explainable Benchmark (EEB) 数据集是由韩国高丽大学计算机科学与工程系开发，旨在解决自动语音识别（ASR）模型在实际应用中的诊断和验证问题。该数据集考虑了语音和文本两个层面，通过细分错误类型，帮助更细致地理解模型的不足。数据集的构建基于现有的韩国语法错误校正（GEC）数据集，并考虑了语音环境，以实现更贴近实际的评估。EEB数据集的应用领域主要集中在ASR模型的性能诊断和提升，旨在通过详细分析错误类型，提高模型的识别准确性和用户友好性，从而增强终端用户的满意度。

Error Explainable Benchmark (EEB) dataset was developed by the Department of Computer Science and Engineering, Korea University, South Korea. It is designed to address the diagnosis and validation challenges of automatic speech recognition (ASR) models in real-world applications. This dataset covers both speech and text modalities, and subdivides error categories to enable a more granular comprehension of model shortcomings. The construction of EEB is based on existing Korean grammatical error correction (GEC) datasets, while incorporating realistic speech environments to facilitate evaluations that closely align with real-world usage scenarios. The primary application scope of the EEB dataset centers on performance diagnosis and enhancement of ASR models, aiming to improve model recognition accuracy and user-friendliness via detailed error type analysis, thereby enhancing end-user satisfaction.