Shinji Watanabe

@espnetUser and @D-Keqi, I'm not sure I can merge this until we could confirm the performance. Is it ongoing?

Once your PR becomes a good shape, please ping me. I'll review it again.

Please fixe the CI issue https://github.com/espnet/espnet/actions/runs/6345118958/job/17236560016

@soumimaiti, can you continue this PR?

Sorry for my late response. This PR looks very good to me!

@YosukeHiguchi, can you review this PR?

Yes, I recognized it. This is just for an assignment. If you finish it, please ping us.

Please increase the batch_bins times multiplied by the number of GPUs (i.e., `batch_bins` = 10,000,000 in your case). (This is the common usage for multiple GPUs in various frameworks)

> If I double batch_bins to 10 million for dual GPUs, it would take even longer to complete each epoch This would be due to some inefficient GPU computing. -...

Thanks for the detailed report. So, the GPU usage was also always busy, right? (GPU utilizations in `nvidia-smi` was always over 70%) I think @kan-bayashi or @soumimaiti can answer it.