LibriTTS

“LibriTTS 是一个多语者英语语料库，以 24kHz 采样率读取约 585 小时的英语语音，由 Heiga Zen 在 Google Speech 和 Google Brain 团队成员的协助下编写。LibriTTS 语料库专为 TTS 研究而设计。它源自 LibriSpeech 语料库的原始资料（来自 LibriVox 的 mp3 音频文件和来自 Project Gutenberg 的文本文件）。与 LibriSpeech 语料库的主要区别如下： 音频文件采用 24kHz 采样率。语音分割为句子中断。包括原始文本和规范化文本。可以提取上下文信息（例如，相邻句子）。排除具有显着背景噪声的话语。

LibriTTS is a multi-speaker English speech corpus containing approximately 585 hours of English speech sampled at 24 kHz, developed by Heiga Zen with the assistance of Google Speech and Google Brain team members. The LibriTTS corpus is specifically designed for TTS research. It is derived from the original sources of the LibriSpeech corpus: MP3 audio files from LibriVox and text files from Project Gutenberg. The main differences compared to the LibriSpeech corpus are as follows:
- Audio files are sampled at 24 kHz.
- Speech is segmented at sentence boundaries.
- Both raw and normalized text are included.
- Contextual information (e.g., adjacent sentences) can be extracted.
- Utterances with significant background noise are excluded.

搜集汇总

背景与挑战

背景概述

LibriTTS是一个专为文本到语音（TTS）研究设计的多语者英语语料库，包含约585小时的24kHz采样率英语语音，源自LibriSpeech语料库并进行了句子分割和文本规范化处理，以提升语音合成质量。该数据集由Google于2019年发布，适用于音频识别和语音合成任务，采用CC BY 4.0许可协议。