Is it OK that the training loss kept around 5.8 when steps goes up to 250K？

[butterl]

Hi , I used thchs30 dataset to train wavenet and the loss kept around 5.8 when steps goes up to 200K, I'm not sure if this is normal and if not， any suggestion to make the loss down？

The evaluation wav got may noise in it and if this wave a good one ? ( I think the wav different need to be less， but the below seem not fine)

[begeekmyfriend]

I am afraid you need to wait until 800K steps #41

[butterl]

Any hparam suggestion？ I only changed the worker num and batch size to fit memory

[WendongGan]

I train it on the data of Ljspeech. After trained over 600k, the loss decrease slowly. I want to know how low can the loss be.

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.