cifkao/open-asr-leaderboard

ESB测试集：Parquet格式和排序版数据集是一个包含多个语音识别测试集的集合，这些测试集经过处理并按音频长度排序。数据集包括AMI、Common Voice、Earnings22、GigaSpeech、LibriSpeech、Speech Robust Bench、SPGISpeech、TED-LIUM、Urgent2025 Nonblind、Voxpopuli和Voxpopuli Cleaned AA等配置。每个配置包含音频、数据集名称、文本转录、唯一ID和音频长度（秒）等特征。该数据集设计用于ASR系统的训练和评估脚本，音频文件在访问时自动解码和重采样。README还提供了每个数据集的数据字段、准备、访问和许可的详细信息，以及使用示例和引用信息。

The ESB Test Sets: Parquet & Sorted dataset is a collection of test sets from various speech recognition datasets, processed and sorted by audio length. It includes configurations for datasets such as AMI, Common Voice, Earnings22, GigaSpeech, LibriSpeech, Speech Robust Bench, SPGISpeech, TED-LIUM, Urgent2025 Nonblind, Voxpopuli, and Voxpopuli Cleaned AA. Each configuration features audio, dataset name, text transcription, unique ID, and audio length in seconds. The dataset is designed for easy access and use in training and evaluation scripts for ASR systems, with audio files decoded and resampled on the fly. The README provides detailed information on data fields, preparation, access, and licensing for each included dataset, along with example usage and citation information.