Running inference on an assortment of FastSpeech2 variants with a new vocoder.. after 3 hours of sifting through the stack trace, realised that I forgot to `unsqueeze` my input vector! *cries in torch*

→ View original post on X — @reach_vb