hf-audio/asr-leaderboard-longform
收藏资源简介:
该数据集名为ASR Leaderboard: Longform Test Sets,包含三个长格式自动语音识别(ASR)基准测试集:Earnings-21、Earnings-22和TED-LIUM。这些数据集用于评估在现实条件下(如长时间音频片段、重叠说话人和特定领域语言)的长格式ASR模型性能。数据集以标准化的Parquet格式提供,包含音频、文本以及根据数据集不同而异的额外元数据。README还详细说明了每个数据集的领域、时长、说话风格、许可证信息,以及使用示例、数据字段、数据准备、评估方法和许可信息。
This dataset is titled ASR Leaderboard: Longform Test Sets, which includes three long-form automatic speech recognition (ASR) benchmark datasets: Earnings-21, Earnings-22, and TED-LIUM. These datasets are used to evaluate the performance of long-form ASR models under real-world conditions such as long-duration audio clips, overlapping speakers and domain-specific languages. The datasets are provided in a standardized Parquet format, containing audio, text and additional metadata that varies across different datasets. The README also details the domain, duration, speaking style, license information of each dataset, as well as usage examples, data fields, data preparation, evaluation methods and licensing information.




