declare-lab/tango

Downloading AudioCaps data

fyell opened this issue · 2 comments

fyell commented

Hi,

I'm trying to download the AudioCaps data in order to train the Tango model. However, I'm not seeing any instructions in the AudioCaps repository on how to download it. Can you share any scripts or instructions on how to download and format the audio to train Tango?

Thanks!

You need to use something like youtube-dl to download the audio files from youtube.

Otherwise, you may want to download the WavCaps dataset and extract the audios from the zip files. This will include AudioCaps among several other datasets.

The dataset already provides a ChatGPT-generated caption for each audio file, but you can probably map the audio files to the original AudioCaps captions using the filenames.