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

fixed issue with the 'add another target audio to artificially create longer sequences' for HDF5 just duplicating the utterance initially sampled

Browse Source
This commit is contained in:
mrq 2023-10-18 20:38:33 -05:00
parent 09cda7d3f9
commit 3195026dba
2 changed files with 5 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
vall_e/config.py
View File

@ -134,7 +134,7 @@ class Dataset:
phones_range: list[int] = field(default_factory=lambda: [4, 256]) phones_range: list[int] = field(default_factory=lambda: [4, 256])
duration_range: list[float] = field(default_factory=lambda: [1.0, 12.0]) duration_range: list[float] = field(default_factory=lambda: [1.0, 12.0])
min_utterances: int = 0 min_utterances: int = 2
random_utterance: float = 1.0 random_utterance: float = 1.0
max_prompts: int = 3 max_prompts: int = 3

17
vall_e/data.py
View File

@ -214,15 +214,13 @@ class Dataset(_Dataset):
spkr = cfg.get_spkr( data_dir / "dummy" ) spkr = cfg.get_spkr( data_dir / "dummy" )
spkr_group = cfg.get_spkr_group( data_dir / "dummy" ) spkr_group = cfg.get_spkr_group( data_dir / "dummy" )
if len(self.paths_by_spkr_name[spkr]) < cfg.dataset.min_utterances:
continue
if spkr_group not in self.spkrs_by_spkr_group: if spkr_group not in self.spkrs_by_spkr_group:
self.spkrs_by_spkr_group[spkr_group] = [] self.spkrs_by_spkr_group[spkr_group] = []
self.spkrs_by_spkr_group[spkr_group].append( spkr ) self.spkrs_by_spkr_group[spkr_group].append( spkr )
self.spkr_groups = list(self.spkrs_by_spkr_group.keys()) self.spkr_groups = list(self.spkrs_by_spkr_group.keys())
self.spkr_samplers = { name: Sampler( [*set(speakers)], keep_all=True ) for name, speakers in self.spkrs_by_spkr_group.items() } self.spkr_samplers = { name: Sampler( [*set(speakers)], keep_all=True ) for name, speakers in self.spkrs_by_spkr_group.items() }
if cfg.dataset.sample_type == "path": if cfg.dataset.sample_type == "path":
@ -385,15 +383,8 @@ class Dataset(_Dataset):
spkr_group = self.spkr_groups[index] spkr_group = self.spkr_groups[index]
spkr_group_id = self.spkr_group_symmap[spkr_group] spkr_group_id = self.spkr_group_symmap[spkr_group]
spkr_name = self.spkr_samplers[spkr_group].sample() spkr_name = self.spkr_samplers[spkr_group].sample()
if spkr_name in self.spkr_symmap: spkr_id = self.spkr_symmap[spkr_name]
spkr_id = self.spkr_symmap[spkr_name] path = self.samplers[spkr_name].sample()
else:
spkr_id = -1
try:
path = self.samplers[spkr_name].sample()
except Exception as e:
print( "ERROR", spkr_group, spkr_name )
raise e
elif cfg.dataset.sample_type == "speaker": elif cfg.dataset.sample_type == "speaker":
spkr_name = self.spkrs[index] spkr_name = self.spkrs[index]
spkr_id = self.spkr_symmap[spkr_name] spkr_id = self.spkr_symmap[spkr_name]
@ -432,7 +423,7 @@ class Dataset(_Dataset):
sampled_path = random.choice(choices) sampled_path = random.choice(choices)
choices = [*(set(choices) - {sampled_path})] choices = [*(set(choices) - {sampled_path})]
if cfg.dataset.use_hdf5: if cfg.dataset.use_hdf5:
key = _get_hdf5_path(path) key = _get_hdf5_path(sampled_path)
txt = cfg.hdf5[key]["text"][:] txt = cfg.hdf5[key]["text"][:]
qnt = cfg.hdf5[key]["audio"][:, :] qnt = cfg.hdf5[key]["audio"][:, :]