Only using the pretrained model of HiFiGAN Singing

[powei-C]

Hi, Thank you for sharing the pre-trained model of HiFiGAN Singing! I would like to use only the vocoder part for another research, so I extracted the "hifigan.py" file and its imported files and modified them for inference.

However, after loading "model_ckpt_steps_1512000.ckpt" and "config.yaml" files and following the procedure in "hifigan.py", the inference "wav_out" remains silence (I write the file with scipy.io.wavfile). I use your " wav2spec" function to generate mel-spec, then use "y = self.model(c).view(-1)" branch to inverse the mel-spec. Since there's no error occurred, I wonder if there's another required config file (except for "config.yaml" in 0109_hifigan_bigpopcs_hop128.zip ) to load for inference? Or do you have any suggestions to figure out this issue?

Thank you very much.