ecker/vall-e
5
9
Fork 0
Code Issues 8 Pull Requests Packages Projects Releases Wiki Activity

Forbid the prompt and response audio to be the same

Browse Source
This commit is contained in:
enhuiz 2023-01-14 17:09:25 +08:00
parent be4619cddf
commit 1c5d49f9c1
1 changed files with 11 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
vall_e/data.py
View File

@ -90,15 +90,21 @@ class VALLEDatset(Dataset):
self._head = None
self.min_phones = min_phones
self.max_phones = max_phones
self.paths = [
path for path in paths if _validate(path, self.min_phones, self.max_phones)
]
self.spkr_symmap = spkr_symmap or self._get_spkr_symmap()
self.phone_symmap = phone_symmap or self._get_phone_symmap()
self.training = training
self.paths_by_spkr_name = self._get_paths_by_spkr_name(extra_paths_by_spkr_name)
self.paths = [
p for p in self.paths if len(self.paths_by_spkr_name[_get_spkr_name(p)]) > 1
]
if training:
self.sampler = Sampler(self.paths, [_get_spkr_name])
else:
@ -128,18 +134,16 @@ class VALLEDatset(Dataset):
def _get_spkr_symmap(self):
return {s: i for i, s in enumerate(self.spkrs)}
def sample_prompts(self, spkr_name, better_not):
def sample_prompts(self, spkr_name, ignore):
prom_list = []
choices = set(self.paths_by_spkr_name[spkr_name]) - {better_not}
choices = set(self.paths_by_spkr_name[spkr_name]) - {ignore}
choices = [*choices]
if len(choices) == 0:
_logger.info(
f"Failed to find another different utterance for {spkr_name}, "
"using the same audio as prompt."
raise ValueError(
f"Failed to find another different utterance for {spkr_name}."
)
choices = [better_not]
for _ in range(cfg.max_prompts):
path = random.choice(choices)
@ -160,7 +164,7 @@ class VALLEDatset(Dataset):
spkr_name = _get_spkr_name(path)
text = torch.tensor([*map(self.phone_symmap.get, _get_phones(path))])
proms = self.sample_prompts(spkr_name, better_not=path)
proms = self.sample_prompts(spkr_name, ignore=path)
resps = _load_quants(path)
resp = resps[..., 0]