Parcollet Titouan

Hi, this should certainly be managed at the Kaldi level as labels and features are generated with Kaldi !

The .pkl that is saved is pretty standard ! You can load it with torch.load()

Hi ! Isn't it simply a problem with the path of the test dataset in the config file ?

Is the problem happening if you use the validation or training set as the test set?

@mravanelli ok to merge that? Seems ok to me.

Hi, sorry for the late reply, we are very very busy with SpeechBrain and PyTorch-Kaldi won't be maintained anymore very soon. You can easily wrap the pytorch transformers to a...

Hi ! So this is quite a hard problem in itself. 2K hours is a lot, and 10 days of training on a Single GPU sounds reasonable to me. You...

How many GPUs do you have ?

Well, yes. The model is stored each epoch. So if the training stops at epoch 13 (let's say at 50%) then it should restart at epoch 13 0%

It actually depends where this print comes from. It can be the sequence length (as sequence are ordered, smaller ones start first) or sequence_lenght * batch_size (often done before a...