TCST-6D-AD: A Six-tuple Parallel Tibetan-Chinese Speech Translation Dataset for the Amdo Dialect

We constructed the TCST-6D-AD end-to-end Tibetan-Chinese speech translation dataset. The source data was obtained through three channels: web collection, laboratory data sharing, and screening of public datasets. After processing through standardized procedures such as Tibetan-Chinese machine translation, text normalization, speech synthesis, audio preprocessing, and six-tuple modality alignment, the final dataset contains six-tuple structured data including Tibetan spoken speech and text, Tibetan written speech and text, and Chinese text and speech. The TCST-6D-AD dataset includes one wav folder, one text folder, and Hexad-Metadata.json. The wav folder has three subfolders: tw_speech (Tibetan written speech), ts_speech (Tibetan spoken speech), m_speech (Chinese speech); the text folder contains three aligned text files: tw_text.tsv (Tibetan written text), ts_text.tsv (Tibetan spoken text), m_text.tsv (Chinese text).The overall scale of the dataset is as follows: it contains 10,068 six-tuple samples, with a total data size of 6.05 GB; among them, the text folder is 7.03 MB, the metadata file is 11 MB, and the audio files are 6.03 GB. The total audio durations are: Tibetan spoken speech 1,324.95 minutes, Tibetan written speech 1,165.44 minutes, and Chinese speech 883.84 minutes.