CN-Celeb: Multi-genre speaker recognition

Research on speaker recognition is extending to address the vulnerability in wild conditions, among which genre mismatch perhaps most challenging, for instance, enrollment with reading speech while testing conversational or singing audio. This leads complex and composite inter-session variations, both intrinsic (i.e., speaking style, physiological status) extrinsic recording device, background noise). Unfortunately, few existing multi-genre corpora are not only limited size but also recorded under controlled cannot support conclusive research problem. In this work, we firstly publish CN-Celeb, a large-scale corpus that includes in-the-wild utterances of 3000 speakers 11 different genres. Secondly, using dataset, conduct comprehensive study phenomenon, particular impact challenge performance gain when new dataset used training.