MAGICDATA introduction
Contains 755 hours of voice data, which is mainly recording data of mobile terminals. 1080 speakers from different key regions in China were invited to participate in the recording. Sentence transcription accuracy is greater than 98%. Recordings were made in a quiet indoor environment. The database is divided into training set, validation set and test set with a ratio of 51:1:2. Details such as speech data encoding and speaker information are stored in metadata files. The fields of recorded texts are diversified, including interactive question and answer, music search, SNS information, home command and control, etc. Segmented transcripts are also provided. The corpus is designed to support researchers in speech recognition, machine translation, speaker recognition, and other speech-related fields. Therefore, the corpus is completely free for academic use.