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

possible logic optimization so I don't spend another 15 minutes simply iterating back to the point I was at in vall_e.emb.process

Browse Source
This commit is contained in:
mrq 2025-02-16 11:34:05 -06:00
parent 0dc49ef4d5
commit 8f86cf0e4e
1 changed files with 13 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

24
vall_e/emb/process.py
View File

@ -322,13 +322,12 @@ def process(
if len(text) == 0 or outpath.exists(): if len(text) == 0 or outpath.exists():
continue continue
# audio not already loaded, load it if max_duration:
if waveform is None: info = torchaudio.info( inpath )
waveform, sample_rate = load_audio( inpath, dtype=dtype ) if info.num_frames / info.sample_rate > max_duration:
continue
if max_duration and waveform.shape[-1] / sample_rate > max_duration:
continue
waveform, sample_rate = load_audio( inpath, dtype=dtype )
jobs.append(( outpath, waveform, sample_rate, text, language )) jobs.append(( outpath, waveform, sample_rate, text, language ))
else: else:
i = 0 i = 0
@ -351,15 +350,18 @@ def process(
if len(text) == 0 or outpath.exists(): if len(text) == 0 or outpath.exists():
continue continue
start = (segment['start']-0.05)
end = (segment['end']+0.5)
if max_duration and end - start > max_duration:
continue
# audio not already loaded, load it # audio not already loaded, load it
if waveform is None: if waveform is None:
waveform, sample_rate = load_audio( inpath, dtype=dtype ) waveform, sample_rate = load_audio( inpath, dtype=dtype )
start = int((segment['start']-0.05) * sample_rate) start = int(start * sample_rate)
end = int((segment['end']+0.5) * sample_rate) end = int(end * sample_rate)
if max_duration and (end - start) / sample_rate > max_duration:
continue
if not presliced: if not presliced:
if start < 0: if start < 0: