When I run inference in Jupyter and my trained model (90 speakers, 45k iters) works, but not exacly how it should.
When I tested it with suprised.txt and her datased... I mean I get voices, but alot of speakers sounds like speaker 4 IDs before it. I mean ID 86 sounds like ID 82, but ID 0 or 1 sounds good.
I think the issue is with dataset, because I know I should use my train.txt with all my IDs... Yeah, but what about dataset? Is there some code to use my model as a dataset? Because Colab Flowtron Speech doesnt need any wavs, theyre used for training, which I did already...
How to set my flowtron.pt as a dataset in Jupyter, is there any code for this?
Of course I tried to put all my wavs to data folder with train.txt with all IDs, and dataloader sees correct amount of speakers, but... It cannot load, I just wait and Kernel is busy all the time, because dataset is in gigabytes, so yeah...
Hi,
When I run inference in Jupyter and my trained model (90 speakers, 45k iters) works, but not exacly how it should. When I tested it with suprised.txt and her datased... I mean I get voices, but alot of speakers sounds like speaker 4 IDs before it. I mean ID 86 sounds like ID 82, but ID 0 or 1 sounds good.
I think the issue is with dataset, because I know I should use my train.txt with all my IDs... Yeah, but what about dataset? Is there some code to use my model as a dataset? Because Colab Flowtron Speech doesnt need any wavs, theyre used for training, which I did already... How to set my flowtron.pt as a dataset in Jupyter, is there any code for this?
Of course I tried to put all my wavs to data folder with train.txt with all IDs, and dataloader sees correct amount of speakers, but... It cannot load, I just wait and Kernel is busy all the time, because dataset is in gigabytes, so yeah...
Please help me.