SR Evaluation Data

CN-Celeb

# Accuracy (ACC)
# Equal Error Rate (EER)

Data Description

CN-Celeb contains a voiceprint database of 1,000 Chinese celebrities with 130,109 Chinese utterances (including singers, actors, rappers, etc.) from interviews, music performances, and film works, totaling 273.72 hours of audio.
This dataset includes 11 types of real-world scenarios, covering complexities such as noise, channel effects, and pronunciation variations. It is particularly suitable for studying speaker recognition technologies in complex scenarios.

Dataset Composition and Specifications

Source Data Volume

The training set contains 111,260 speech samples from 800 speakers, and the test set contains 18,849 speech samples from 200 speakers.

Evaluation Data Volume

The test set, sourced from CN-Celeb, contains 18,849 speech samples.

Data Fields

For the ASI (Automatic Speaker Identification) task, both the training and test sets are stored in .txt files, with each line containing two fields: the audio file index and the speaker label.

*.txt: audio_file_path speaker_id

For the ASV (Automatic Speaker Verification) task, the training and validation sets use the same format. However, the test set is saved in .txt files with only the audio file index.

*.txt: audio_file_path

Dataset Example

downstream/spkrec/dataset/data/id00013/singing-03-036.flac id00013

Evaluation Metrics

• Accuracy (ACC) is used for the speaker identification task.
• Equal Error Rate (EER) is used for the speaker verification task.

Paper Citations

@inproceedings{fan2020cn,
  title={CN-CELEB: a challenging Chinese speaker recognition dataset},
  author={Fan, Yue and Kang, JW and Li, LT and Li, KC and Chen, HL and Cheng, ST and Zhang, PY and Zhou, ZY and Cai, YQ and Wang, Dong},
  booktitle={ICASSP 2020-2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
  pages={7604--7608},
  year={2020},
  organization={IEEE}
}

@misc{li2020cn,
  title={CN-Celeb: multi-genre speaker recognition},
  author={Lantian Li and Ruiqi Liu and Jiawen Kang and Yue Fan and Hao Cui and Yunqi Cai and Ravichander Vipperla and Thomas Fang Zheng and Dong Wang},
  year={2020},
  eprint={2012.12468},
  archivePrefix={arXiv},
  primaryClass={eess.AS}
 }

Open Source License

Attribution-ShareAlike 4.0 International