Changed format of input file to csv with headers (required)#14
Changed format of input file to csv with headers (required)#14
Conversation
| @@ -1,3 +1,4 @@ | |||
| mels|pitch|text | |||
| wavs/LJ045-0096.wav|pitch/LJ045-0096.pt|Mrs. De Mohrenschildt thought that Oswald, | |||
There was a problem hiding this comment.
these files should then also be updated to use absolute paths
|
|
||
| assert not (load_pitch_from_disk and self.pitch_tmp_dir is not None) | ||
|
|
||
| if len(self.audiopaths_and_text[0]) < expected_columns: |
There was a problem hiding this comment.
should we still do some kind of check on expected number of columns?
There was a problem hiding this comment.
oh lol, I commented on this across 2 PRs
| pitchpath = self.audiopaths_and_text[index]['pitch'] | ||
| pitch = torch.load(pitchpath) | ||
| if self.pitch_mean is not None: | ||
| assert self.pitch_std is not None |
There was a problem hiding this comment.
commenting here because I can't do it lower down but what about the TTSCollate call function?
I guess that's to do with the return type if getitem returning a tuple?
| audiopath, *fields = self.audiopaths_and_text[index] | ||
| audiopath = self.audiopaths_and_text[index]['mels'] | ||
|
|
||
| if self.n_speakers > 1: |
There was a problem hiding this comment.
just out of interest, where is this spk used? Afaik it's not currently used here, though I imagine the intention is to have different mean and std?
|
@johannahom I'll test just this branch in an existing setup (without the conditioning), however that still leaves the issue of what paths to use in these paths shared by everyone |
|
works |
2 participants
See new input file as examples.