Piotr Żelasko

Looks like not every training example has features extracted. Make sure you passed the path to the right cut set (with features). You can also check ‘lhotse cut describe ’...

> looks like features num is much smaller than cuts count? is that something wrong?and why it happend? I combine two sets to get the cut_train set and I found...

You either need to use keep_overlapping=False or filter out the cuts that have overlapping speech (whichever makes sense for your use case).

Some tips: - splitting cut/recording/supervision set into smaller parts can be done with `parts = cuts.split(num_parts)`, e.g.: ``` In [4]: cuts Out[4]: CutSet(len=1519) [underlying data type: ] In [8]: cuts.split(2)...

I didn’t get your question, please elaborate.

Yes, you can compute the features inside the PyTorch dataset class. See OnTheFlyFeatures or K2SpeechRecognitionDataset for some examples. You can also look up k2-fsa/icefall repo for recipes that support this.

In addition to what Desh stated: `plot/play_audio` does not actually support multi-channel data (yet). The reason it was slow for you is because matplotlib plot received a 2-d array `(num_channels,...

It looks like you might have some data loaded in memory. Can you share more about the context of your usage? Are you using webdataset/shar or functions such as move_to_memory?...

I think the issue might be that you also have audio data that is in memory and can't be stored in JSONL. You can either do `cuts = cuts.drop_recordings()` before...

In the example posted above copy_feats had worked so I can't really replicate your case. The only other thing I can think of right now is that you might have...