Output of the discriminators.

jik876 / hifi-gan

HiFi-GAN: Generative Adversarial Networks for Efficient and High Fidelity Speech Synthesis

MIT License

1.97k stars 507 forks source link

Closed helloybz closed 3 years ago

helloybz commented 3 years ago

Hi.

I can't get how it is guaranteed that the discriminators' outputs are in (0, 1)???

In either DiscriminatorS in hifi-gan/models.py or DiscriminatorP in hifi-gan/models.py, I cannot find sigmoid-like stuffs.

But while computing losses of the discriminators in your implementations, it seems the outputs are treated as it is probabilities.

helloybz commented 3 years ago

Ok, now i've read the LSGAN paper.