def uttloader(scp_config, reader_kwargs, loader_kwargs, train=True):
mix_reader = SpectrogramReader(scp_config['mixture'], **reader_kwargs)
target_reader = [
SpectrogramReader(scp_config[spk_key], **reader_kwargs)
for spk_key in scp_config if spk_key[:3] == 'spk'
]
dataset = Dataset(mix_reader, target_reader)
# modify shuffle status
loader_kwargs["shuffle"] = train
# validate perutt if needed
# if not train:
# loader_kwargs["batch_size"] = 1
# if validate, do not shuffle
utt_loader = DataLoader(dataset, **loader_kwargs)
return utt_loader
def run(args):
reader_kwargs = {
"frame_length": args.frame_length,
"frame_shift": args.frame_shift,
"window": args.window,
"center": False,
"apply_abs": True,
"apply_log": args.apply_log,
"apply_pow": args.apply_pow
}
num_bins = nfft(args.frame_length) // 2 + 1
reader = SpectrogramReader(args.wave_scp, **reader_kwargs)
mean = np.zeros(num_bins)
std = np.zeros(num_bins)
num_frames = 0
# D(X) = E(X^2) - E(X)^2
for _, spectrogram in tqdm.tqdm(reader):
num_frames += spectrogram.shape[0]
mean += np.sum(spectrogram, 0)
std += np.sum(spectrogram**2, 0)
mean = mean / num_frames
std = np.sqrt(std / num_frames - mean**2)
with open(args.cmvn_dst, "wb") as f:
cmvn_dict = {"mean": mean, "std": std}
pickle.dump(cmvn_dict, f)
print("Totally processed {} frames".format(num_frames))
print("Global mean: {}".format(mean))
print("Global std: {}".format(std))
def run(args):
# return complex result
reader_kwargs = {
"frame_length": args.frame_length,
"frame_shift": args.frame_shift,
"window": args.window,
"center": True
}
print(
"Using {} Mask".format("Ratio" if not args.psm else "Phase Sensitive"))
mixture_reader = SpectrogramReader(args.mix_scp,
**reader_kwargs,
return_samps=True)
targets_reader = [
SpectrogramReader(scp, **reader_kwargs) for scp in args.ref_scp
]
num_utts = 0
for key, packed in mixture_reader:
samps, mixture = packed
norm = np.linalg.norm(samps, np.inf)
skip = False
for reader in targets_reader:
if key not in reader:
print("Skip utterance {}, missing targets".format(key))
skip = True
break
if skip:
continue
num_utts += 1
if not num_utts % 1000:
print("Processed {} utterance...".format(num_utts))
targets_list = [reader[key] for reader in targets_reader]
spk_masks = compute_mask(mixture, targets_list, args.psm)
for index, mask in enumerate(spk_masks):
istft(os.path.join(args.dump_dir,
'{}.spk{}.wav'.format(key, index + 1)),
mixture * mask,
**reader_kwargs,
norm=norm,
fs=8000,
nsamps=samps.size)
print("Processed {} utterance!".format(num_utts))
def uttloader(scp_config, reader_kwargs, loader_kwargs, train=True):
mix_reader = SpectrogramReader(scp_config['mixture'], **reader_kwargs)
target_reader = [
SpectrogramReader(scp_config[spk_key], **reader_kwargs)
for spk_key in scp_config if spk_key[:3] == 'spk'
]
dataset = Datasets(mix_reader, target_reader, mvn_dict= loader_kwargs["mvn_dict"])
# modify shuffle status
loader_kwargs["shuffle"] = train
# validate perutt if needed
# if not train:
# loader_kwargs["batch_size"] = 1
# if validate, do not shuffle
#utt_loader = DataLoaders(dataset, **loader_kwargs)
utt_loader = DataLoader(dataset, batch_size=40,shuffle=loader_kwargs['shuffle'],
num_workers=10, sampler=None,drop_last=True,
collate_fn=_collate)
# mvn_dict= loader_kwargs["mvn_dict"])
return utt_loader
def run(args):
num_bins, conf_dict = parse_yaml(args.train_conf)
reader = SpectrogramReader(args.wave_scp,
**conf_dict["spectrogram_reader"])
mean = np.zeros(num_bins)
std = np.zeros(num_bins)
num_frames = 0
# D(X) = E(X^2) - E(X)^2
for _, spectrogram in tqdm.tqdm(reader):
num_frames += spectrogram.shape[0]
mean += np.sum(spectrogram, 0)
std += np.sum(spectrogram**2, 0)
mean = mean / num_frames
std = np.sqrt(std / num_frames - mean**2)
with open(args.cmvn_dst, "wb") as f:
cmvn_dict = {"mean": mean, "std": std}
pickle.dump(cmvn_dict, f)
print("Totally processed {} frames".format(num_frames))
print("Global mean: {}".format(mean))
print("Global std: {}".format(std))