jik876 / hifi-gan

HiFi-GAN: Generative Adversarial Networks for Efficient and High Fidelity Speech Synthesis
MIT License
1.92k stars 506 forks source link

the fmax #80

Open burcehan opened 3 years ago

burcehan commented 3 years ago

Hi ,thanks for your work, in the config_v1.json file , the sampling_rate is 22050, the fmax is 8000 as follows, why the fmax is 8000, can the fmax be 11025? Thanks "sampling_rate": 22050, "fmin": 0, "fmax": 8000,

Miralan commented 3 years ago

It's OK

jik876 commented 3 years ago

Thank you for your interest. We followed previous work for a fair comparison in quality assessment. Since setting a low fmax is similar to using a partially obscured input condition, using a higher fmax may lead to better quality. It should be the same as the fmax of the 1st stage model you are using.