KT-Speech-Crawler

KT-Speech-Crawler是由信息科学知识技术大学汉堡分校开发的一个自动数据集构建工具，专门用于从YouTube视频中提取语音样本以支持语音识别系统的训练。该数据集通过爬虫技术自动收集，包含约108,617条样本，涵盖了多种语音条件，如背景噪声、音乐、远距离麦克风录音以及多种口音和回声。数据集的创建过程涉及多个过滤和后处理步骤，以确保样本的质量。该数据集主要应用于解决语音识别技术中的数据稀缺问题，通过提供大量多样化的语音样本，帮助提升语音识别系统的性能。

KT-Speech-Crawler is an automated dataset construction tool developed by the Hamburg Campus of the University of Information Science and Knowledge Technology, specifically designed to extract speech samples from YouTube videos to support the training of speech recognition systems. This dataset is automatically collected via web crawling technology, containing approximately 108,617 speech samples covering various speech conditions including background noise, music, distant microphone recordings, diverse accents and echoes. Multiple filtering and post-processing steps are involved in the dataset creation process to ensure the quality of the samples. This dataset is primarily developed to address the data scarcity issue in speech recognition technology, and it helps improve the performance of speech recognition systems by providing a large volume of diverse speech samples.