Code repository for the Cantonese In-car Audio-Visual Speech Recognition (CI-AVSR) dataset.

HLTCHKUST, updated 🕥 2022-07-14 08:16:09


Repository for the paper CI-AVSR: A Cantonese Audio-Visual Speech Dataset for In-car Command Recognition and the corresponding new dataset, which is accepted in LREC 2022.

If you find our dataset or code useful, please cite this paper, thanks!

  title={CI-AVSR: A Cantonese Audio-Visual Speech Dataset for In-car Command Recognition},
  author={Wenliang Dai and Samuel Cahyawijaya and Tiezheng Yu and Elham J. Barezi and Peng Xu and Cheuk Tung Shadow Yiu and Rita Frieske and Holy Lovenia and Genta Indra Winata and Qifeng Chen and Xiaojuan Ma and Bertram E. Shi and Pascale Fung},


Version 1.0

For details of the dataset, please refer to the paper.

Clean Sets

The originally collected data, including train_clean.csv, valid_clean.csv, test_clean.csv.

Noise Augmented Sets

The noise augmented data, including train_noisy.csv, valid_noisy.csv, test_noisy.csv. In addition, there is also a out-of-domain test set test_noisy_ood.csv to evaluate the generalization of models, i.e. these noises are not in the training set.

Data Download

[CI-AVSR Dataset] We provide the processed CI-AVSR dataset with audios, image frames (25 per seconds), annotations, and augmentation.

[Raw Videos] Just in case, we also provide the unprocessed raw videos (you may not need to use them).


A question about the text data

opened on 2023-03-29 07:50:34 by MayWang1007

Thanks for sharing your wok.Here is my question: I want to use the to train the model and have downloaded the whole dataset you provided.But I didn't find any text data in the dataset.could you pls give me more details about it?thanks a lot image

how to preprocess the data for model training?

opened on 2022-08-02 04:54:17 by yan159yan

Good work for the visual-audio data. is there any parameter configuration for the ""?