spoken-squad
收藏资源简介:
该数据集包含两个配置版本:默认配置和预处理配置。默认配置是一个多模态数据集,包含音频数据及其对应的文本转录、问题列表以及主题和段落标识符。音频采样率为16kHz,每个样本包括音频文件、真实转录文本(gt_transcript)、多个问题(questions)以及用于组织的topic_id和paragraph_id。数据集分为训练集(16035个样本)和测试集(1896个样本),适用于语音识别、音频转录和问答相关任务。预处理配置仅包含训练集的文本嵌入向量特征,每个嵌入向量为384维浮点数,适用于需要预计算文本表示的下游任务。
This dataset includes two configuration versions: the default configuration and the preprocessed configuration. The default configuration is a multimodal dataset containing audio data along with corresponding text transcriptions, question lists, and topic and paragraph identifiers. The audio sampling rate is 16kHz, and each sample consists of an audio file, ground truth transcription text (gt_transcript), multiple questions (questions), and organizational identifiers such as topic_id and paragraph_id. The dataset is divided into a training set (16,035 samples) and a test set (1,896 samples), making it suitable for tasks related to speech recognition, audio transcription, and question answering. The preprocessed configuration only contains text embedding vector features for the training set, with each embedding vector being a 384-dimensional floating-point number, applicable to downstream tasks that require precomputed text representations.
数据集概述
基本信息
- 数据集名称:spoken-squad
- 数据集地址:https://huggingface.co/datasets/alextsigilis/spoken-squad
- 数据集总大小:约34.04 GB(下载大小约34.04 GB)
配置(Configs)
1. default(默认配置)
- 数据集大小:约34.05 GB
- 数据划分:
- 训练集(train):16,035 个样本,约30.31 GB
- 测试集(test):1,896 个样本,约3.73 GB
- 数据特征:
topic_id:整型(int64),话题IDparagraph_id:整型(int64),段落IDaudio:音频数据,采样率为16000 Hzgt_transcript:字符串(string),真实转录文本questions:字符串列表(list of string),相关问题
- 数据文件位置:
- 测试集:
data/test-* - 训练集:
data/train-*
- 测试集:
2. preprocessed(预处理配置)
- 数据集大小:约21.48 GB
- 数据划分:
- 训练集(train):16,001 个样本,约21.48 GB
- 数据特征:
text_embedding:浮点16类型(float16)的列表,长度为384,表示文本嵌入向量
- 数据文件位置:
- 训练集:
preprocessed/train-*
- 训练集:




