def __init__(self, mix_scp, s1_scp, s2_scp):
"""Initialize DataReader. (2 mixtures)
Args:
mix_scp: scp file for mixed waves (KALDI format)
"""
check(mix_scp, s1_scp, s2_scp)
self.key = read_key(mix_scp)
self.mix_path = read_path(mix_scp)
self.s1_path = read_path(s1_scp)
self.s2_path = read_path(s2_scp)
def __init__(self,
mix_c1_scp,
s1_c1_scp,
s2_c1_scp,
mix_c2_scp,
s1_c2_scp,
s2_c2_scp,
sample_rate,
sample_clip_size=4):
"""Initialize the TimeDomainDateset. (2 mixtures)
Args:
mix_scp: scp file for mixed waves (KALDI format)
s1_scp: scp file for speaker 1
s2_scp: scp file for speaker 2
sample_clip_size: segmental length (default: 4s)
"""
check(mix_c1_scp, s1_c1_scp, s2_c1_scp, mix_c2_scp, s1_c2_scp,
s2_c2_scp)
self.sample_rate = sample_rate
self.sample_clip_size = sample_clip_size
self.segment_length = self.sample_rate * self.sample_clip_size
self.mix_c1_path = read_path(mix_c1_scp)
self.s1_c1_path = read_path(s1_c1_scp)
self.s2_c1_path = read_path(s2_c1_scp)
self.mix_c2_path = read_path(mix_c2_scp)
self.s1_c2_path = read_path(s1_c2_scp)
self.s2_c2_path = read_path(s2_c2_scp)
self.retrieve_index = []
for i in range(len(self.mix_c1_path)):
sample_size = len(wavread(self.mix_c1_path[i])[0])
if sample_size < self.segment_length:
# wave length is smaller than segmental length
if sample_size * 2 < self.segment_length:
continue
self.retrieve_index.append((i, -1))
else:
# Cut wave into clips and restore the retrieve index
sample_index = 0
while sample_index + self.segment_length < sample_size:
self.retrieve_index.append((i, sample_index))
sample_index += self.segment_length
if sample_index != sample_size - 1:
self.retrieve_index.append(
(i, sample_size - self.segment_length))