def CPC_feature_function(x):
CPC_features = buildFeature(featureMaker,
x,
seqNorm=args.seq_norm,
strict=args.strict,
maxSizeSeq=args.max_size_seq)
return CPC_features.squeeze(0).float().cpu().numpy()
def cpc_feature_function(x):
if args.nobatch is False:
return buildFeature_batch(featureMaker,
x,
seqNorm=False,
strict=args.strict,
maxSizeSeq=args.max_size_seq,
batch_size=args.batch_size)
else:
return buildFeature(featureMaker,
x,
seqNorm=False,
strict=args.strict)
def feature_function(x):
if args.nobatch is False:
res0 = buildFeature_batch(featureMaker, x,
seqNorm=False,
strict=args.strict,
maxSizeSeq=args.max_size_seq,
batch_size=args.batch_size)
if args.norm_vec_len:
# [!] we actually used CPC_audio/scripts/quantize_audio.py for that in the end
res0Lengths = torch.sqrt((res0*res0).sum(2))
res0 = res0 / res0Lengths.view(*(res0Lengths.shape), 1)
return res0
else:
res0 = buildFeature(featureMaker, x,
seqNorm=False,
strict=args.strict)
if args.norm_vec_len:
# [!] we actually used CPC_audio/scripts/quantize_audio.py for that in the end
res0Lengths = torch.sqrt((res0*res0).sum(2))
res0 = res0 / res0Lengths.view(*(res0Lengths.shape), 1)
return res0
def cpc_feature_function(x):
if args.nobatch is False:
res0 = buildFeature_batch(featureMaker,
x,
seqNorm=False,
strict=args.strict,
maxSizeSeq=args.max_size_seq,
batch_size=args.batch_size)
if args.norm_vec_len:
res0Lengths = torch.sqrt((res0 * res0).sum(2))
res0 = res0 / res0Lengths.view(*(res0Lengths.shape), 1)
return res0
else:
res0 = buildFeature(featureMaker,
x,
seqNorm=False,
strict=args.strict)
if args.norm_vec_len:
res0Lengths = torch.sqrt((res0 * res0).sum(2))
res0 = res0 / res0Lengths.view(*(res0Lengths.shape), 1)
return res0
def feature_function(x): return buildFeature(feature_maker, x,
seqNorm=args.seq_norm,
strict=args.strict,
maxSizeSeq=args.max_size_seq)
elif args.load == 'from_pre_computed':
def buildAllFeature(featureMaker,
pathDB,
pathOut,
seqList,
stepSize=0.01,
strict=False,
maxSizeSeq=64000,
format='fea',
seqNorm=False):
totSeqs = len(seqList)
startStep = stepSize / 2
bar = progressbar.ProgressBar(maxval=totSeqs)
bar.start()
for nseq, seqPath in enumerate(seqList):
bar.update(nseq)
feature = buildFeature(featureMaker,
os.path.join(pathDB, seqPath),
strict=strict or seqNorm,
maxSizeSeq=maxSizeSeq,
seqNorm=seqNorm)
_, nSteps, hiddenSize = feature.size()
outName = os.path.basename(os.path.splitext(seqPath)[0]) + f'.{format}'
fname = os.path.join(pathOut, outName)
if format == 'npz':
time = [startStep + step * stepSize for step in range(nSteps)]
values = feature.squeeze(0).float().cpu().numpy()
totTime = np.array([stepSize * nSteps], dtype=np.float32)
with open(fname, 'wb') as f:
np.savez(f, time=time, features=values, totTime=totTime)
elif format == 'npy':
time = [startStep + step * stepSize for step in range(nSteps)]
values = feature.squeeze(0).float().cpu().numpy()
with open(fname, 'wb') as f:
np.save(f, values)
elif format == 'af':
import arrayfire as af
time = [startStep + step * stepSize for step in range(nSteps)]
values = feature.squeeze(0).float().cpu().numpy()
totTime = np.array([stepSize * nSteps], dtype=np.float32)
af.save_array("time", af.Array(time, dtype=af.Dtype.f32), fname)
af.save_array("totTime",
af.interop.from_ndarray(totTime),
fname,
append=True)
af.save_array("features",
af.interop.from_ndarray(values),
fname,
append=True)
else:
with open(fname, 'w') as f:
_, nSteps, hiddenSize = feature.size()
for step in range(nSteps):
line = [startStep + step * stepSize] + \
feature[0, step, :].tolist()
line = [str(x) for x in line]
linestr = ' '.join(line) + '\n'
f.write(linestr)
bar.finish()
def feature_function(x):
return buildFeature(feature_maker,
x,
seqNorm=args.seq_norm,
strict=args.strict,
maxSizeSeq=args.max_size_seq)