def __init__(self, training_files, segment_length, filter_length,
hop_length, win_length, sampling_rate, mel_fmin, mel_fmax):
if os.path.isfile(str(training_files)):
self.audio_files = files_to_list(training_files)
else:
self.audio_files = []
random.seed(1234)
random.shuffle(self.audio_files)
self.stft = TacotronSTFT(filter_length=filter_length,
hop_length=hop_length,
win_length=win_length,
sampling_rate=sampling_rate,
mel_fmin=mel_fmin, mel_fmax=mel_fmax)
self.segment_length = segment_length
self.sampling_rate = sampling_rate
import unidecode
import yaml
import librosa
from waveglow import inference as waveglow
from melgan import inference as melgan
from mellotron import inference as mellotron
from utils.argutils import locals2dict
from mellotron.layers import TacotronSTFT
from mellotron.hparams import create_hparams
# 用griffinlim声码器
_hparams = create_hparams()
_stft = TacotronSTFT(_hparams.filter_length, _hparams.hop_length,
_hparams.win_length, _hparams.n_mel_channels,
_hparams.sampling_rate, _hparams.mel_fmin,
_hparams.mel_fmax)
_use_waveglow = 0
_device = 'cuda' if torch.cuda.is_available() else 'cpu'
filename_formatter_re = re.compile(r'[\s\\/:*?"<>|\']+')
def plot_mel_alignment_gate_audio(mel,
alignment,
gate,
audio,
figsize=(16, 16)):
fig, axes = plt.subplots(4, 1, figsize=figsize)
axes = axes.flatten()