Piotr Żelasko

Thanks! Yes that seems to have helped. It now seems that only torch 2.3+ is compiled against numpy 2.0 so I will have to adapt Lhotse tests accordingly.

I don't believe that you can have both random access and IO efficiency on slow NFS setups. A few years ago I was playing around with Apache Arrow for audio...

Yes, please refer to this tutorial to understand how to set up multiple datasets (possibly corresponding to multiple tasks) in a single training: https://colab.research.google.com/github/lhotse-speech/lhotse/blob/master/examples/03-combining-datasets.ipynb I specifically recommend `CutSet.mux()` approach for...

Either your FLAC file was somehow corrupted (maybe during download, in which case remove the data and re-try downloading), or for some reason is not readable by soundfile. You can...

Yes, one example is given towards the end of this tutorial https://github.com/lhotse-speech/lhotse/blob/master/examples/04-lhotse-shar.ipynb if you don’t want to go into tar archives you can use any of the feature writers in...

Update to the latest lhotse version (ideally from the master branch)

We had broken transform serialization after some refactoring that I fixed fairly recently (I think 1.24.2 has those fixes already). > d["transforms"] = [t.to_dict() for t in self.transforms] > AttributeError:...

This PR seems to have slipped through. Should we merge it? @ericharper @titu1994

Cut-level transforms such as noise augmentation, reverb, speed perturbation, etc. in lhotse return copies of the original cuts, so it will be easy to obtain multiple variants of the same...

Yea, thanks. Fixed.