Alright7398/modded-distill-wavlm-base

该数据集包含LibriSpeech语料库在16 kHz采样率下的音频数据，以及通过冻结的microsoft/wavlm-base模型预计算的最后一层隐藏状态表示。这是一个预计算的教师缓存和原始音频字节数据集，并非通用的语音基准测试分割。数据集结构包括训练集（train/）和评估集（eval/），分别对应LibriSpeech的960小时训练语料（train-clean-100、train-clean-360、train-other-500）和开发集（dev-clean）。每个数据行包含音频键、分割标签、原始音频字节、音频长度（以样本数表示）以及教师模型的最后一层隐藏状态（以float16列表形式存储）。数据集适用于特征提取和表示学习任务，使用Lance格式存储，需通过Lance读取器加载。Lance tables of LibriSpeech utterances at 16 kHz with cached last-layer representations from frozen microsoft/wavlm-base. This release is a precomputed teacher cache plus raw audio bytes, not a general-purpose speech benchmark split. The dataset includes train and eval directories, corresponding to the LibriSpeech 960-hour training corpora (train-clean-100, train-clean-360, train-other-500) and dev-clean set, respectively. Each row contains an utterance key, split tag, raw audio bytes, audio length in samples, and flattened last hidden states from WavLM-Base for valid frames (stored as float16 list). It is designed for feature extraction and representation learning, stored in Lance format and requires a Lance reader for loading.