Imaginary value for fretchet distance

[ranjana-creator]

I used the original audio vs generated audio metric on songs generated using SVC. I am aware that this is not the intended use but I was curious on my way to developing a metric for singing voices with the same lyrics. Can you provide me insight into what an imaginary value means in this case? Is it something wrong with my audios or is it because the FD works only for TTS?