A dataset of Mongolian, Tibetan and Uyghur speech fragments based on voice activity detection

Based on the speech data in Mongolian, Tibetan, and Uyghur speech data from Chinese minority regions in 2015, we adopted a double threshold Voice Activity Detection method with short-time energy and short-time zero-crossing rate to obtain multiple voice fragments of each sentence speech. The result dataset contains 1657 Mongolian speech fragments, 666 Tibetan speech fragments and 756 Uygur speech fragments. The total volume of the data is about 111 MB. Through automatic software segmentation and multiple auditing and proofreading by language experts, we have obtained high-quality voice fragment data of Mongolian, Tibetan and Uygur, which can be applied to minority speech recognition, voice activity detection, speech enhancement, speech synthesis and language teaching.