lilgoose777/kantipur-news-data

Nepali Speech Dataset (YouTube-sourced) 74 labeled speech segments, split by channel (not by individual video) so the same speaker/recording can't appear in more than one split. Splits train: 74 segments validation: 0 segments test: 0 segments Transcript columns — read this before training Each segment carries three transcript variants. They are NOT interchangeable: text_original — the YouTube caption text (if any) that overlapped this segment's… See the full description on the dataset page: https://huggingface.co/datasets/lilgoose777/kantipur-news-data.