Text to Speech Distribution Score 2 (TTSDS2)

TTSDS2是一个用于评估文本到语音系统质量的资源，它通过比较合成语音与真实语音的分布相似度来评估系统。该数据集包含来自YouTube和LibriVox等多个来源的数据，涉及14种语言。TTSDS2使用多种感知因素，如说话人身份、清晰度和韵律，通过比较这些因素的分布来评估合成语音的质量。此外，该数据集还提供了一个不断更新的基准，用于14种语言的文本到语音系统。

TTSDS2 is a dedicated resource for evaluating the quality of text-to-speech (TTS) systems, which assesses such systems by measuring the distributional similarity between synthesized speech and natural human speech. This dataset contains data from multiple sources including YouTube and LibriVox, covering 14 distinct languages. TTSDS2 utilizes multiple perceptual factors such as speaker identity, intelligibility, and prosody, and evaluates the quality of synthesized speech by comparing the distributions of these factors between synthetic and natural speech. Furthermore, this dataset provides a continuously updated benchmark for text-to-speech systems across 14 languages.