Audio Demo Page

Samples from Voice Conversion System

Source Samples are from Librispeech dev-clean dataset.

Comparison to other models with different chunk sizes.

HiFi-GAN V1 causal is checkpoint g_012400000.pt with zero padding.

HiFi-GAN V1 causal ctx is checkpoint g_01800000.pt with retaining context.

Model 9600 4800 3200 1600
HiFi-GAN V1 orignal
HiFi-GAN V1 causal
HiFi-GAN V1 causal ctx