mythicinfinity/Libriheavy-HQ

Libriheavy-HQ数据集是Libri-Light的高质量版本，其音频文件被替换为librivox提供的最高质量版本，无需重新编码。在大多数情况下，这意味着源音频从64kbps的.mp3升级到128kbps的.mp3。目前，small子集中提供了500小时的音频，未来将添加更多子集。该数据集适用于文本到语音、文本到音频和自动语音识别等任务。

Libriheavy-HQ is a high-quality speech recognition dataset based on the Libri-Light dataset, using the highest quality audio files without re-encoding. The dataset includes multiple subsets, with the currently available small subset containing 509 hours of speech data from 417 speakers. The purpose of the dataset is to provide a high-quality resource for text-to-speech, text-to-audio, and automatic speech recognition tasks.