CVSS

CVSS是一个大规模多语言到英语的语音到语音翻译（S2ST）数据集，包含21种语言到英语的句子级平行S2ST对。该数据集源自Common Voice语音数据集和CoVoST 2语音到文本翻译数据集，通过使用先进的TTS系统将CoVoST 2中的翻译文本合成语音。CVSS提供两种版本的英语翻译语音：CVSS-C，所有翻译语音使用单一高质量标准语音；CVSS-T，翻译语音使用与源语音相应的转移语音。此外，CVSS还提供与翻译语音发音匹配的标准化翻译文本，适用于模型训练和评估。CVSS旨在解决多语言环境下语音翻译的挑战，推动直接S2ST模型的研究。

CVSS is a large-scale multilingual-to-English speech-to-speech translation (S2ST) dataset, consisting of sentence-level parallel S2ST pairs from 21 languages to English. It is derived from the Common Voice speech dataset and the CoVoST 2 speech-to-text translation dataset, with the translated texts in CoVoST 2 being synthesized into speech via state-of-the-art TTS systems. CVSS provides two versions of English translated speech: CVSS-C, where all translated speech adopts a single high-quality standard voice; and CVSS-T, where the translated speech uses a transferred voice corresponding to the source speech. Additionally, CVSS offers standardized translated texts that match the pronunciation of the translated speech, which are suitable for model training and evaluation. CVSS aims to address the challenges of speech translation in multilingual scenarios and advance research on direct S2ST models.