Shinji Watanabe

oh, i see. i overlooked it. then, that is fine.

I haven't observed such issues, as we did not usually perform batch decoding with conformer. @pengchengguo, can you answer @danpovey's questions about the masking?

zenodo issues seem to happen again. We'll work on switching HF hub as a default model sharing space.

I added a bit more details of uploading a model to the HF hub in https://github.com/espnet/espnet_model_zoo#huggingface Can you check it and try it?

Are these models using s3prl? If so, the test should include it in the installation stage. It is a little bit difficult decision as we want to make espnet_model_zoo installation...

Can you fix the conflict?

Unfortunately, espnet1 cannot be used. It is based on a very different interface. If you have issues reproducing the same results with espnet2, you can post it in the espnet...

I'd say it isn't easy. I recommend you check the espnet2 config similar to your current problem (regarding the amount of training data and language).

Which data are you using? Which language? How many hours? Then, I'll let you know which config is appropriate. As I said, it is better to make an issue in...

I see the point. I'm sorry that I don't have a nice solution about it right now, but will let you know once I come up with something. Thanks for...