UserLibri

UserLibri数据集是由谷歌公司创建，旨在支持移动设备上的语音模型个性化研究。该数据集基于LibriSpeech，包含了从Project Gutenberg获取的个性化文本数据，用于训练个性化语言模型以提高语音识别性能。数据集包含107个用户，每个用户都有一组音频-文本对和额外的文本数据。创建过程中，数据集通过标准化文本文件编码和去除非内容信息来处理原始PG书籍文本文件。该数据集主要应用于解决移动设备上语音识别的个性化问题，通过利用设备上的文本数据来学习用户偏好的词汇或短语，从而提高语音识别的准确性。

The UserLibri dataset was developed by Google to support research on personalized speech models for mobile devices. It is built upon LibriSpeech and includes personalized text data sourced from Project Gutenberg, which is used to train personalized language models to enhance speech recognition performance. The dataset contains 107 users, each provided with a set of audio-text pairs and additional text data. During its creation, the original PG book text files were processed via standardized text encoding and the removal of non-content information. The primary application of this dataset is to address the personalized speech recognition problem on mobile devices: by leveraging on-device text data to learn a user's preferred vocabulary or phrases, thereby improving the accuracy of speech recognition.