Distortion of wav generated by the pre-trained model

Question

Distortion of wav generated by the pre-trained model

oshindow opened this issue a year ago · comments

hi,
I'm trying to generate wav from the pre-trained model NSPP_02100000 with the defalut config.json modifing the checkpoint_file_load and test_input_wav_dir only. The test wav is p278_002.wav in VCTK, but the generated wav was distortied. The tone of output wav is strange.

YangAi520 · Answer 1 · Thu Mar 02 2023 14:43:07 GMT+0800 (China Standard Time)

hi, I'm trying to generate wav from the pre-trained model NSPP_02100000 with the defalut config.json modifing the checkpoint_file_load and test_input_wav_dir only. The test wav is p278_002.wav in VCTK, but the generated wav was distortied. The tone of output wav is strange.

I am sorry. I made a little mistake while tidying up the code. In line 79 of model.py, the positions of pseudo_imaginary_part and pseudo_real_part need to be reversed. I have modified it and updated the code. You can try it again. The pre-trained model NSPP_02100000 is trained using the updated code, so it exists distortion when using the original code. Thanks for your question.

Wang Xintong · Answer 2 · Thu Mar 02 2023 15:09:01 GMT+0800 (China Standard Time)

Solved this problem. Amazing model!