Pad CutSet with silence?

I’m wondering whether there is an easy way to pad all the cuts in a CutSet with a fixed amount of silence on the sides, as in extend()? This is for things like debugging edge issues with models, or decoding and making sure final symbols are output. I tried extend_by(), but it appears to be a no-op, e.g. a.extend_by(0.05, preserve_id=True).fill_supervisions() seemed to return the original CutSet (possibly because it only extended to the edge of the original recordings?) I ended up doing this with something like:

 load_manifest(self.args.manifest_dir / "cuts_test-clean.json.gz").pad(direction='both').fill_supervisions()

just for testing purposes, which isn’t ideal because it make all utterances as long as the longest one, which is not what I really wanted.

[desh2608]

I guess we could add a boolean argument pad_silence to the extend_by() function which specifies whether to pad the cut with silence if not enough context is present.

[desh2608]

Closing via https://github.com/lhotse-speech/lhotse/pull/816.