Heng-Jui Chang

Results 5 comments of Heng-Jui Chang

Perhaps you can add `CUDA_VISIBLE_DEVICES=1` before the python script when executing, where the number `1` can be changed to your GPU's ID.

You may refer to the paper [Connectionist Temporal Classification: Labelling Unsegmented Sequence Data with Recurrent Neural Networks](https://www.cs.toronto.edu/~graves/icml_2006.pdf) or [this website](https://distill.pub/2017/ctc/) to better understand CTC loss.

You can generate vocabulary for different languages with `util/generate_vocab_file.py`.

I think that `test_path` is an argument defined by yourself since I cannot find it in the original code. The original code already has the functionality of testing the trained...

You can find them here: https://github.com/s3prl/s3prl/blob/26113e9febbaaee6311f8c2cdd9475dc495acce9/s3prl/pretrain/distiller/pretrain_expert.py#L330-L337