How can I use a tacotron2 or fastpitch model with Riva?

I have riva running on my AGX Xavier, and it is sounding fantastic! But I’d like to use one of the models I have pretrained with my own voice. I have signed up for the Riva custom voice early access, but in the meantime I already have a couple of models I can train with fastpitch for example.

I saw in the documentation that it is possible, but I have no idea where to put the model files or access them.
Any help is much appreciated! Thanks!

Hi @jcwscience

Thanks for your interest in Riva,

I will check with my team and provide information on how to use the model in Riva


You can follow the instructions from Custom Models — NVIDIA Riva to run inference using a custom model.

Thanks, I’ll take a look

How would I convert a tacotron2 file to a .Nemo file then?

Hi @jcwscience

I will check with the team,
but guessing the model is available as .nemo

I would not recommend using the Tacotron2 model as it will be removed in the end of October Riva release.