Why using z but not flow(z_p, spk_emb) for decoder ?

jaywalnut310 / vits

VITS: Conditional Variational Autoencoder with Adversarial Learning for End-to-End Text-to-Speech

https://jaywalnut310.github.io/vits-demo/index.html

MIT License

6.48k stars 1.21k forks source link

Why using z but not flow(z_p, spk_emb) for decoder ? #167

Open thanhkm opened 11 months ago

thanhkm commented 11 months ago

Hello, thank you for your great project!

I wonder is there any underlying reason to use z instead of z_p + spk_emb for decoder? The second schema could be post_encoder -> z - spk_emb -> z_p -> z_p + spk_emb -> z' -> wav. Will it make the flow more robust in the inference step?

Best regards