CN-Celeb is a multi-genre dataset covering 11 different genres in real world, collected from multiple Chinese open media sources.
3,000
SpeakersCN-Celeb contains speech from Chinese celebrities.
600,000 +
UtterancesCN-Celeb covers multiple genres of speech, including entertainment, interview, singing, play, movie, vlog, live broadcast, speech, drama, recitation and advertisement.
1,200 +
Hours
CN-Celeb consists of complex long-short challenge which meets the scenarios of most
real applications.
Download
The dataset consists of two subsets, CN-Celeb1 and CN-Celeb2. For each subset, we provide audio files and speaker meta-data. There is no overlap between the two subsets. CN-Celeb1 contains more than 125,000 utterances from 997 Chinese celebrities, and CN-Celeb2 contains more than 520,000 utterances from 1,996 Chinese celebrities.
License
All the resources contained in the dataset are free for research institutes and individuals. The copyright remains with the original owners of the audio/video. No commerical usage is permitted.
Publications
Publications based on the dataset welcome to cite the following papers:
Y.Fan, J.W.Kang, L.T.Li, K.C.Li, H.L.Chen, S.T.Cheng, P.Y.Zhang, Z.Y.Zhou, Y.Q.Cai, D.Wang*
CN-Celeb: A Challenging Chinese Speaker Recognition Dataset, ICASSP, 2020
L.T.Li, R.Q.Liu, J.W.Kang, Y.Fan, H.Cui, Y.Q.Cai, R.Vipperla, T.F.Zheng, D.Wang*
CN-Celeb: multi-genre speaker recognition, Speech Communication, 2022
Challenge
We are hosting the first CN-Celeb Speaker Recognition Challenge (CNSRC) at Odyssey 2022 (The Speaker and Language Recognition Workshop). CNSRC aims to evaluate how well the current speaker recognition methods work in real world scenarios, usually with in-the-wild complexity and real-time processing speed. CNSRC consists of two parts, an evaluation challenge and an accompanying workshop. The challenge website can be found here and the workshop website can be found here .
Acknowledgements
This work is supported by the National Natural Science Foundation of China (NSFC) under Grants No.61633013 and No.62171250.
CN-Celeb-AV is a multi-genre audio-visual person recognition dataset covering 11 different genres in the real world, collected from multiple Chinese open media sources.
1,136
SpeakersCN-Celeb-AV contains speech from Chinese celebrities.
419,000 +
UtterancesCN-Celeb-AV covers multiple genres of speech, including entertainment, interview, singing, play, movie, vlog, live broadcast, speech, drama, recitation and advertisement.
660 +
HoursCN-Celeb-AV consists of both full-modality and partial-modality challenges which meet the scenarios of most real applications.
Video
Audio
Dev-F:689
A development set with full-modality information, contains both audio and visual information.
Eval-F:197
An evaluation set with full-modality information, contains both audio and visual information.
Eval-P:250
An evaluation set with partial-modality information, contains some segments whose audio or visual information is corrupted or fully lost.
Download
The dataset consists of three subsets, Dev-F, Eval-F and Eval-P. For each subset, we provide video and audio files and speaker meta-data. There is no overlap among the three subsets. Dev-F contains more than 93,000 segments from 689 Chinese celebrities, Eval-F contains more than 17,000 segments from 197 Chinese celebrities, and Eval-P contains more than 307,900 segments from 250 Chinese celebrities.
License
All the resources contained in the dataset are free for research institutes and individuals. The copyright remains with the original owners of the audio/video.
No commerical usage is permitted.
Please register and log in to the CN-Celeb system, and then submit the data license to request the data.
Acknowledgements
This work is supported by the National Natural Science Foundation of China (NSFC) under Grants No.62171250.
CN-CVS is a large-scale continuous visual-speech dataset in Mandarin Chinese consisting of short clips collected from TV news and Internet speech shows.
2,500 +
Speakers
CN-CVS contains speech from speakers
ranging in age from teenagers to seniors.
200k +
UtterancesCN-CVS covers complex environmental factors. For instance, different videos are recorded with diverse cameras; in the same video, the angle and distance of the camera may change.
300 +
HoursCN-CVS involves two parts: CN-CVS/News which consists of TV news, and CN-CVS/Speech which consists of speech shows from online media.
Download
CN-CVS involves two parts: CN-CVS/News and CN-CVS/Speech. The former is relatively constrained in speaking style and camera setting while the latter involves more real-life complexity. CN-CVS/News contains 13,016 segments from 28 speakers with 35 hours in total. CN-CVS/Speech contains 193,245 utterances from 2,529 speakers with 273 hours in total.
License
All the resources contained in the dataset are free for research institutes and individuals. The copyright remains with the original owners of the audio/video.
No commerical usage is permitted.
Please register and log in to the CN-Celeb system, and then submit the data license to request the data.
Acknowledgements
This work is supported by the National Natural Science Foundation of China (NSFC) under Grants No.62171250.