TibAdD-D2Std ST:Tibetan Amdo Dialect Style Transformation Dataset

Among them, 15,000 data points are sourced from internal public data of the laboratory, while the remaining 45,299 data points are collected from public live broadcasts on short video platforms. Systematic preprocessing, manual annotation, and other data processing tasks are conducted according to standard data processing rules, and the data is anonymized.

Ultimately, high-quality parallel data for the Anduo dialect of Tibetan is constructed. This dataset is stored in JSON format, with each element in the array representing an independent JSON object corresponding to an alignment entry of a set of spoken language text and its standard written language text. Each object contains four fields: unique ID, spoken language, written language, and data source. All objects constitute a JSON array, facilitating program parsing, batch processing, and subsequent stylistic transformation modeling work.