BidirLM/librispeech_contrastive

该数据集是BidirLM-Omni集合的一部分，专门针对音频模态，基于Librispeech语料库的子样本。它包含音频文件（采样率为16kHz）、说话者ID、章节ID、唯一标识符和锚点信息，用于训练BidirLM-Omni模型。数据集分为训练集（104,014个示例）和验证集（2,703个示例），总大小约为24.27GB。

This dataset is part of the BidirLM-Omni Collection, specifically a sub-sample for the audio modality based on Librispeech. It includes audio files (with a sampling rate of 16kHz), speaker IDs, chapter IDs, unique identifiers, and anchor information, used for training the BidirLM-Omni models. The dataset is split into train (104,014 examples) and validation (2,703 examples) sets, with a total size of approximately 24.27GB.