AISHELL-3 High-fidelity Chinese Speech Database

The AISHELL-3 Chinese Mandarin Speech Database has a speech duration of 85 hours and a total of 88,035 sentences, and can be used as a multi-speaker synthesis system.The recording process was conducted in a quiet indoor environment using a high-fidelity microphone (44.1kHz, 16bit). The dataset was recorded by 218 speakers from different accent regions in China. Professional voice proofreaders performed pinyin and rhythm annotations and passed strict quality inspections. The accuracy of the phonetic characters in this database is above 98%.