Piotr Żelasko
Piotr Żelasko
In the dataset class, instead of calling `AudioSamples` / `PrecomputedFeatures` / `OnTheFlyFeatures`, simply do the following: ```python video_features, video_features_lens = lhotse.dataset.collation.collate_custom_field(cuts, "video_features") batch["inputs"] = video_features ``` You might need to...
Very interesting, thank you for debugging. LGTM
Can you fix the failing checks?
Regarding the first issue it looks like I haven't updated `CutPairsSampler` properly with latest changes. I'll take a look. Regarding the other question, You might want to use `DynamicCutSampler` or...
The MLS recipe was the first one we added for very large datasets, and it's implemented less efficiently than others. You'd need to modify it to use incremental manifest writers...
But yeah generally expect it to take a while as English MLS is quite sizeable. It may be possible to implement it differently to accommodate distributed compute environments and speed...
It depends on whether this kind of supervision is acceptable for your experimental setup. In most cases if you have a supervision that began 9s before the cut, a huge...
You’d need to modify the CPP code here to return indices instead/in addition to symbols, and adjust the CPP-Python interfaces. It’s a worthwhile addition but I don’t have the bandwidth...