class LibriSequence:
def __init__(self):
self.path = config.path
self.specpath = config.specpath
self.ls = LibriSpeech()
self.ls.load()
self.batchsize = 128
def sequence(self, type="train"):
array = self.ls.info[type][::-1]
def get(ix):
reader,book,i = array[ix]
file = self.ls.data[reader][book][i]
buf, _ = sf.read(join(self.path, file['path']))
trans = file['trans']
wb = whole_buffer()
wb.params.spectrum_range = config.librispeech_range
return trans, wb.all(buf)
def get2(ix):
reader,book,i = array[ix]
file = self.ls.data[reader][book][i]
path = file['path']
if path.endswith(".flac"):
path = path[:-5] + ".npy"
buf = numpy.load(join(self.specpath, path)).astype(numpy.float32)/4096
return file['trans'], buf
return SequenceFromLibriSpeech(array, self.batchsize, get2)
def _random_test_buf():
ls = LibriSpeech()
ls.load()
file = ls.uniform_test()
trans = file['trans']
fn = join(config.path, file['path'])
buf = _buf_from_file(fn)
buf = _buffer_block([buf])
return trans, buf
def convert_all_flac():
path = config.path
dest = "../speechify_dat/spectrogram"
ls = LibriSpeech()
ls.load()
f = FilesToSpectrogram()
for reader, r_ in ls.data.items():
mkdir(dest, reader)
for book, b_ in r_.items():
mkdir(dest, reader, book)
for i, file in enumerate(b_):
npzf = join(dest, reader, book, "%s-%s-%04d.npy" % (reader, book, i))
print("Creating %s" % npzf)
f.write_out_spectrogram(npzf, join(path, file['path']))
class Sampler:
def __init__(self, home=False):
self.path = config.path
self.ls = LibriSpeech()
self.ls.load()
self.wb = whole_buffer()
self.wb.params.spectrum_range = config.librispeech_range
def rand(self):
file = self.ls.uniform_train() #file = self.ls.uniform_random()
buf, _ = sf.read(join(self.path, file['path']))
return file, self.wb.all(buf)
def whatev(self):
pass
class LibriSequence:
def __init__(self):
self.path = config.path
self.ls = LibriSpeech()
self.ls.load()
self.batchsize = 4
def sequence(self, type="train"):
def get(ix):
reader,book,i = self.ls.info[type][ix]
file = self.ls.data[reader][book][i]
buf, _ = sf.read(join(self.path, file['path']))
trans = file['trans']
wb = whole_buffer()
wb.params.spectrum_range = config.librispeech_range
return trans, wb.all(buf)
return SequenceFromLibriSpeech(self.ls.info[type], self.batchsize, get)
def __init__(self, home=False):
self.path = config.path
self.ls = LibriSpeech()
self.ls.load()
self.wb = whole_buffer()
self.wb.params.spectrum_range = config.librispeech_range
from util.data import LibriSpeech
ls = LibriSpeech()
ls.make("../speechify_dat/combined")
ls.train_test_split()
ls.dump()
def __init__(self):
self.path = config.path
self.ls = LibriSpeech()
self.ls.load()
self.batchsize = 4