How to handle feature-only datasets?

[t13m]

Wondering is there any way to handle datasets where no audio but their features are available? For example, AudioSet provides audio features extracted from youtube videos.

I'm new to lhotse. If I understand the document correctly, a typical lhotse workflow starts from constructing the recording manifest, and audio source seems required there. Am I missing something?

[desh2608]

Yes, a Cut in Lhotse can be defined using any combination of Recording, Feature, or Supervision. See the CutSet.from_manifests() method here:

lhotse/lhotse/cut/set.py

Line 310 in 89ca0e6

def from_manifests(

.

Once you have created the CutSet using the features, you can use them in the same way as you would use any regular cut set.

[desh2608]

The only thing you would need to do is to store the features as a FeatureSet. I would suggest looking at this class:

lhotse/lhotse/features/io.py

Line 258 in 89ca0e6

class LilcomFilesWriter(FeaturesWriter):

[t13m]

OK, that's exactly what I want to know. Thank you! I'll look into it.