NVIDIA / flowtron

Flowtron is an auto-regressive flow-based generative network for text to speech synthesis with control over speech variation and style transfer
https://nv-adlr.github.io/Flowtron
Apache License 2.0
887 stars 177 forks source link

incompatibility with pytorch >= 1.7.0 #129

Open deepglugs opened 3 years ago

deepglugs commented 3 years ago

Upgraded to 1.9.0 and flowtron inference stopped producing anything but noise. After downgrading to pytorch 1.6.0, it started working again. Anyone else seeing this?

rui-lin commented 3 years ago

This version worked for me torch 1.8.1+cu111

ddascalescu commented 2 years ago

Found the same issue. Both torch versions 1.10.2+cu111 and 1.8.1+cu111 produced no speech output, whereas version 1.6.0+cu101 produced speech correctly.