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

tweaks

Browse Source
This commit is contained in:
mrq 2025-02-10 20:53:08 -06:00
parent 276a2342a4
commit d6a679ca5c
2 changed files with 16 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
vall_e/data.py
View File

@ -749,11 +749,7 @@ def _load_paths_from_metadata(group_name, type="training", validate=False):
metadata = {}
if cfg.dataset.use_metadata and metadata_path.exists():
#metadata = json.loads(open( metadata_path, "r", encoding="utf-8" ).read())
try:
metadata = json_read( metadata_path )
except Exception as e:
return []
metadata = json_read( metadata_path )
if len(metadata) == 0:
return _fn( data_dir, type if cfg.dataset.use_hdf5 else _get_artifact_extension(), validate )
@ -765,8 +761,11 @@ def _load_paths_from_metadata(group_name, type="training", validate=False):
k = key(id, entry)
# double check if in HDF5
# this might be slow
"""
if cfg.dataset.use_hdf5 and k not in cfg.hdf5:
return False
"""
# add to duration bucket
if type not in _durations_map:
@ -1201,7 +1200,16 @@ class Dataset(_Dataset):
index = reference_metadata["similar"][offset]
name = metadata_keys[index]
return root / name
path = root / name
if cfg.dataset.use_hdf5:
key = _get_hdf5_path(path)
if key not in cfg.hdf5[key]:
return None
elif not path.exists():
return None
return path
def sample_prompts(self, spkr_name, reference, should_trim=True):
# return no prompt if explicitly requested for who knows why
@ -1243,6 +1251,7 @@ class Dataset(_Dataset):
path = random.choice(choices)
else:
path = random.choice(choices)
if cfg.dataset.use_hdf5:
key = _get_hdf5_path(path)
qnt = torch.from_numpy(cfg.hdf5[key]["audio"][:, :]).to(torch.int16)

2
vall_e/emb/process.py
View File

@ -246,7 +246,7 @@ def process(
outfolder.mkdir(parents=True, exist_ok=True)
if speaker_id in audio_only:
for filename in sorted(os.listdir(f'./{input_audio}/{group_name}/{speaker_id}/')):
for filename in tqdm(sorted(os.listdir(f'./{input_audio}/{group_name}/{speaker_id}/')), desc=f"Processing {speaker_id}"):
inpath = Path(f'./{input_audio}/{group_name}/{speaker_id}/{filename}')
outpath = Path(f'./{output_dataset}/{group_name}/{speaker_id}/{filename}').with_suffix(audio_extension)