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

could have sworn i had 'vall_e.emb.process --dtype' working, also possible RAM optimization so I can stop locking up my server when firing four encoding processes

Browse Source
This commit is contained in:
mrq 2025-02-07 18:52:19 -06:00
parent 47eb498046
commit ed94b261dc
1 changed files with 7 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
vall_e/emb/process.py
View File

@ -25,11 +25,13 @@ from .qnt import encode as quantize, encode_batch as quantize_batch
def pad(num, zeroes): def pad(num, zeroes):
return str(num).zfill(zeroes+1) return str(num).zfill(zeroes+1)
def load_audio( path, device=None ): def load_audio( path, device=None, dtype=None ):
waveform, sr = torchaudio.load( path ) waveform, sr = torchaudio.load( path )
if waveform.shape[0] > 1: if waveform.shape[0] > 1:
# mix channels # mix channels
waveform = torch.mean(waveform, dim=0, keepdim=True) waveform = torch.mean(waveform, dim=0, keepdim=True)
if dtype is not None:
waveform = waveform.to(dtype)
if device is not None: if device is not None:
waveform = waveform.to(device) waveform = waveform.to(device)
return waveform, sr return waveform, sr
@ -190,7 +192,7 @@ def process(
cfg.inference.weight_dtype = dtype # "bfloat16" cfg.inference.weight_dtype = dtype # "bfloat16"
cfg.inference.amp = amp # False cfg.inference.amp = amp # False
dtype = None dtype = cfg.inference.dtype if not amp else None
output_dataset = f"{output_dataset}/{'2' if cfg.sample_rate == 24_000 else '4'}{'8' if cfg.sample_rate == 48_000 else '4'}KHz-{cfg.audio_backend}" # "training" output_dataset = f"{output_dataset}/{'2' if cfg.sample_rate == 24_000 else '4'}{'8' if cfg.sample_rate == 48_000 else '4'}KHz-{cfg.audio_backend}" # "training"
@ -245,7 +247,7 @@ def process(
if outpath.exists(): if outpath.exists():
continue continue
waveform, sample_rate = load_audio( inpath ) waveform, sample_rate = load_audio( inpath, dtype=dtype )
qnt = quantize(waveform, sr=sample_rate, device=device) qnt = quantize(waveform, sr=sample_rate, device=device)
process_job(outpath, waveform, sample_rate) process_job(outpath, waveform, sample_rate)
@ -296,7 +298,7 @@ def process(
# 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 ) waveform, sample_rate = load_audio( inpath, dtype=dtype )
if max_duration and waveform.shape[-1] / sample_rate > max_duration: if max_duration and waveform.shape[-1] / sample_rate > max_duration:
continue continue
@ -325,7 +327,7 @@ def process(
# 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 ) waveform, sample_rate = load_audio( inpath, dtype=dtype )
start = int((segment['start']-0.05) * sample_rate) start = int((segment['start']-0.05) * sample_rate)
end = int((segment['end']+0.5) * sample_rate) end = int((segment['end']+0.5) * sample_rate)