parler-tts/mls-eng-speaker-descriptions

该数据集是对多语言LibriSpeech（MLS）数据集中英语子集的注释。MLS数据集是一个适用于语音研究的大型多语言语料库，包含来自LibriVox的有声读物，涵盖8种语言。该数据集特别提供了对英语MLS的注释，包括说话者和话语特征的自然语言描述。这些注释是通过Data-Speech仓库生成的。该数据集与原始版本和LibriTTS-R一起用于训练Parler-TTS模型。数据集的特征包括音频路径、时间戳、文本、音频时长、说话者ID、书籍ID、信噪比、语音质量等。数据集分为开发集、测试集和训练集，分别包含不同数量的样本。

This dataset consists of annotations for the English subset of the Multilingual LibriSpeech (MLS) dataset, focusing on natural language descriptions of speaker and utterance characteristics. The MLS dataset is a large multilingual corpus suitable for speech research, derived from audiobooks from LibriVox, containing 8 languages with approximately 44.5K hours of English and a total of about 6K hours for other languages. These annotations were generated using the Data-Speech repository and were used to train the Parler-TTS models.