This dataset is specifically tailored to capture the nuances of speech from the elderly and children, two demographic groups with distinct vocal characteristics. This dataset is recorded using desktop equipment to ensure high audio quality, and all recordings take place in a quiet environment to minimize background noise. The corpus includes read speech, which is beneficial for training speech recognition models on clear and deliberate pronunciations. The gender balance in the dataset ensures that the recognition system can accurately interpret both male and female voices. Furthermore, the speakers are drawn from the seven major Chinese dialect regions, providing a diverse and balanced distribution of accents and speech patterns. For the children's recordings, the dataset includes speech samples from interactive car control systems, children's audiobooks, children's video content, and music featuring children's songs and popular tunes from platforms like TikTok. The elderly recordings cover similar domains with a focus on applications and content that cater to their preferences, such as car control, map navigation, audiobooks with programs selected for an older audience, and music that includes selections favored by the elderly. This comprehensive approach ensures that the speech recognition system can effectively adapt to the unique speech traits of these age groups across various contexts and dialects.
This dataset was recorded in a quiet office/home environment, with the participation of 200 speakers, including 123 males and 77 females. All speakers who took part in the recording were professionally screened to ensure standardized pronunciation and clear articulation. The recorded text materials cover information such as news.