Is there a simplified training method？

[zhjygit]

Is there a simplified training method？
My GPU 4060ti has 16 GB of video memory and more than 100 GB of datasets, can I save time during training by simplifying the steps or reducing the number of datasets I train? In the training phase, if it is for a certain language, you only need to train the u-set, right?

[zhjygit]

Downloaded Chinese datasets, which are video, audio, and train_csv val.csv, and there are a large number of mp4 format videos in the video and audio folders, and the length of each video is less than 3 seconds. In this case, it seems that it is not possible to directly use ./data_processing_pipeline.sh for data processing, and all videos will be automatically deleted, which may not meet some of the reading requirements of AVREARDER, and the video length does not meet the requirements. If it is training in Chinese, can you provide some dataset links and training steps?