The HI-MIA dataset was used in the 2019 AISHELL Speaker Verification Challenge.It was extracted from a larger database called AISHELL-WakeUp-1.

The dataset is divided into the HI-MIA dataset and the training set, which contains the Chinese and English wake-up words "Hi, Mia". The data is collected in a real home environment using a microphone array and a Hi-Fi microphone.The paperThe collection process and development of the baseline system are described. The data used in the challenge is extracted from 1 Hi-Fi microphone and 16-channel circular microphone array of 1/3/5 meters. The content is the wake-up word in Chinese. The whole collection is divided into train (254 people), dev (42 people) and test (44 people) subsets. The test subset provides paired target/non-target answers to evaluate the verification results.

The AISHELL-WakeUp-1 voice database contains 3,936,003 wake-up word voices, totaling 1561.12 hours.The recording languages are Chinese and English; the recording region is China. The recording text is the wake-up word "hello, Mia". This dataset invited 254 speakers to participate in the recording. The recording process was set up in a real home environment, with 7 recording positions, using 6 circular 16-channel PDM microphone array recording boards for far-talk pickup (16kHz, 16bit) and 1 high-fidelity microphone for close-talk pickup (44.1kHz, 16bit). This database has been transcribed and annotated by professional voice proofreaders and passed strict quality inspection, with a word accuracy of 100%. It can be used for research such as voiceprint recognition and voice wake-up recognition.