xuankai@cmu

It seems that I also met the same case with @FAhmedB , when the seq-length of features is less than the seq-length of labels.

@sw005320 I think this is interesting. What we usually did is just setting non-updated parameters to `param.requires_grad = False`. We didn't do this further step `param.grad = None`. We need...

@slSeanWU If the file is not in this repo, formatting would not be a problem.

Hi @triumph9989. It's because of out of memory. Can you confirm how much memory on your computing node and how much GPU each job can use? What s3prl upstream did...

@triumph9989 How did you run the collect-stats, locally or by some resource manage system (e.g. slurm, gridengine)? Did you estimate how much RAM each collect-stats job can reserve? In my...

> @triumph9989 How did you run the collect-stats, locally or by some resource manage system (e.g. slurm, gridengine)? Did you estimate how much RAM each collect-stats job can reserve? In...

Hi @Emrys365 I agree with the `init_param` replacements. It also happens for `asr.sh` and `enh_st.sh`. About the error of lm_dev_txt, we could also move it. One option is [here](https://github.com/espnet/espnet/blob/85bef5609cfd78b4756b52fbc3880bb0e7865517/egs2/TEMPLATE/enh_asr1/enh_asr.sh#L690). What...

@Tonyyouyou Which encoder type did you use? `rnn` or `vgg_rnn`? By default, those two encoder types downsample the input with a factor of 4. You can check the details for...

@Tonyyouyou Oh, sorry. I missed a point. You are using 44.1 kHz data. It may not be suitable to use. These frontends were trained on 16 kHz data. You may...

Hi @deeplearner01, I just want to confirm. Did you make any modification to ESPnet code? Also, can you paste the config you used?