Simple python script for downloading AVSpeech Dataset.
It downloads only the relevant part of the video and saves it in a mp4 container.
Tries to download 720p/360p videos with 25fps and audio at 44.1kHz.
Files names are <yt_id>_<start_time>_<end_time>.mp4
Assumptions/Limitations:
avspeech_train.csvandavspeech_test.csvare in the same directory as the download.py script.- creates the output folder in the currect directory based on the train/test set. For now you have to change the directories in the code if required.
- the script creates a file called
badfiles_train.txtwhich lists the youtube id's of the deleted/private videos which are no longer available for download.
Usage:
python downloader.py train
Replace train with test if you want to download the test set.
Dependencies:
conda install -c conda-forge ffmpeg
conda install -c conda-forge youtube-dl
pip install ffmpeg-python
NOTE:
This is a very simple script. I haven't take too much care, just wrote it quickly to download the dataset. Let me know if you have any problems using it, will try to fix it. Feel free to send in pull requests if you would like to improve it.
Cheers!