GigaST
收藏资源简介:
GigaST是由字节跳动创建的一个大规模伪语音到文本翻译(ST)数据集,基于GigaSpeech英语ASR数据集,将其转录翻译成德语和中文。该数据集通过强大的机器翻译系统翻译训练集,人工翻译测试集,显著提升了MuST-C英语-德语基准测试集的性能。GigaST数据集不仅包含英语-中文(En-Zh)和英语-德语(En-De)翻译方向,还通过发布训练脚本,便于复现其系统,旨在推动语音翻译领域的研究。
GigaST is a large-scale pseudo speech-to-text translation (ST) dataset developed by ByteDance. Built upon the GigaSpeech English ASR dataset, it translates the transcripts from the source dataset into German and Chinese. For the training split, the dataset uses powerful machine translation systems, while the test split is manually translated, which significantly improves the performance on the MuST-C English-German benchmark. The GigaST dataset supports two translation directions: English-Chinese (En-Zh) and English-German (En-De). Additionally, it releases training scripts to facilitate the reproduction of its experimental system, aiming to advance research in the field of speech translation.




