XBMU-AMDO31: A Speech Recognition Dataset for the Amdo Dialect of Tibetan

The speech dataset was collected in the Xiahe region of Gansu Province, China, encompassing 31 hours of recordings from 66 native Tibetan speakers, along with their corresponding transcriptions. The dataset comprises 33 males and 33 females. The speech dataset is divided into training, development, and test sets. The training set comprises 18,505 sentences recorded from 54 speakers; the development set includes 2,045 sentences from 6 speakers; and the test set contains 2,035 sentences from 6 speakers.