I am trying to generate longer sequences with this repository. I am working on Geforce RTX 2080 TI, and getting everything to work well for training.
For inference, I am trying to feed in sequence of mels, but as soon as I go past 6 or 7 seconds, I automatically get out of memory. Looking through the code this is probably caused by the fact that the entire output sequence is initialised before generation starts, whereas since Wavenet is autoregressive, this is not necessary.
Any ideas how to scale this repo to generate longer sequences?
Hello,
I am trying to generate longer sequences with this repository. I am working on Geforce RTX 2080 TI, and getting everything to work well for training.
For inference, I am trying to feed in sequence of mels, but as soon as I go past 6 or 7 seconds, I automatically get out of memory. Looking through the code this is probably caused by the fact that the entire output sequence is initialised before generation starts, whereas since Wavenet is autoregressive, this is not necessary.
Any ideas how to scale this repo to generate longer sequences?