Ego4d icon indicating copy to clipboard operation
Ego4d copied to clipboard

Downloading audio only in Ego-Exo4D?

Open juliawilkins opened this issue 1 year ago • 6 comments

Hi, thank you so much for the great contributions to this space. I was wondering if there is any way to download only the audio (no video), either take- or unprocessed-level via a CLI argument. I have not found it in the CLI docs. Thank you!

juliawilkins avatar Jan 09 '24 20:01 juliawilkins

This is not currently possible. Unfortunately, you will have to download the videos and extract the audio yourself as of right now. For the egocentric audio you need to download the raw VRS files --parts capture_raw_vrs. There will soon be an update, such that the amount of data you have to download will be reduced, but don't expect this right now.

Some questions:

  • What's your use case? Are you after narrate & act transcriptions?
  • Do you need audio from all cameras, just the egocentric or exocentric?

miguelmartin75 avatar Jan 10 '24 03:01 miguelmartin75

Thanks so much for the quick response @miguelmartin75. This makes sense.

  • My use case is from a purely audio perspective, looking for real-world, "general sounds" multichannel audio.
  • When you say "all cameras" what do you mean exactly? Would that be different from ego+exo all audio? And just to clarify, the egocentric audio is the 7-channel Aria glasses-based audio, and the exo is stereo?

Question: what audio would I get at with capture_raw_stitched_videos instead of raw vrs?

juliawilkins avatar Jan 10 '24 15:01 juliawilkins

When you say "all cameras" what do you mean exactly

All devices for the capture, i.e. the egocentric aria and exocentric gopros (and for some small amount of captures, the egocentric gopros).

And just to clarify, the egocentric audio is the 7-channel Aria glasses-based audio, and the exo is stereo?

This is correct.

Question: what audio would I get at with capture_raw_stitched_videos instead of raw vrs?

This is the GoPro recordings (mostly exo). The raw VRS is strictly the egocentric aria device.

miguelmartin75 avatar Jan 16 '24 20:01 miguelmartin75

Hi @miguelmartin75, where is the stereo audio from the GoPro exo recordings? When downloading with --views exo, the cam01.mp4 etc. go-pro mp4s don't seem to have audio. Are there any flags to add to get the audio here, or better yet audio-only? Thanks!

juliawilkins avatar Feb 07 '24 14:02 juliawilkins

Hi, I have the same/similar issue, we can get egocentric pre-extracted audio from --parts take_audio, however, the exocentric .mp4s do not have any audio, how can I obtain the audio for the exocentric videos?

I believe this is also related to this question in the forum:

wisdomikezogwo avatar May 23 '24 02:05 wisdomikezogwo

Unfortunately we do not provide exocentric audio yet. We have this on our backlog cc @devanshk

miguelmartin75 avatar Jun 08 '24 02:06 miguelmartin75