sample.py

import sys
sys.path.append('..')

import speechset
sys.path.pop()


# construct data reader
lj = speechset.datasets.LJSpeech('D:\\dataset\\LJSpeech-1.1')

# construct configuration
config = speechset.Config()

# construct acoustic model
acoustic = speechset.AcousticDataset(lj, config)

# indexing
text, mel = acoustic[0]
print(text.shape, mel.shape)

# split sample
testset = acoustic.split(1000)
# back pack
text, mel, textlen, mellen = acoustic[0:3]
print(text.shape, mel.shape, textlen.shape, mellen.shape)
# iteration
text, mel = next(iter(testset))
print(text.shape, mel.shape)

# construct vocoder model
vocoder = speechset.VocoderDataset(lj, config)

# unpack
mel, audio, mellen, audiolen = vocoder[0:3]
print(mel.shape, audio.shape, mellen.shape, audiolen.shape)

# speaker id support
libri = speechset.datasets.LibriTTS('D:\\dataset\\LibriTTS\\test-clean')

# construct multi speaker
acoustic = speechset.utils.IDWrapper(
    speechset.AcousticDataset(libri, config))

# unpack
sid, text, mel, textlen, mellen = acoustic[0:3]
print(sid.shape, text.shape, mel.shape, textlen.shape, mellen.shape)

# construct concatenated reader
concat = speechset.datasets.ConcatReader([libri, libri])
print(len(concat.speakers()))

# construct wav model
wavset = speechset.utils.IDWrapper(speechset.WavDataset(concat))

# unpack
sid, speeches = wavset[0]
print(sid, speeches.shape)

# dump
DUMP_PATH = 'D:\\dataset\\LibriTTS\\test-clean-dump'
speechset.utils.mp_dump(acoustic, DUMP_PATH, 4)

# load dump
class WrappedAcoustic(speechset.utils.IDWrapper):
    def __init__(self, *args, **kwargs):
        super().__init__(speechset.AcousticDataset(*args, **kwargs))

dumped = speechset.utils.DumpDataset(WrappedAcoustic, DUMP_PATH)

# unpack
sid, text, mel, textlen, mellen = dumped[0:3]
print(sid.shape, text.shape, mel.shape, textlen.shape, mellen.shape)