def __getitem__(self, idx):
if idx in self.cache_Dict.keys():
return self.cache_Dict[idx]
content_Label, content_Path, style_Label, style_Path = self.pattern_List[idx]
content_Mel = Mel_Generate(
audio= Audio_Prep(content_Path, hp_Dict['Sound']['Sample_Rate']),
sample_rate= hp_Dict['Sound']['Sample_Rate'],
num_frequency= hp_Dict['Sound']['Spectrogram_Dim'],
num_mel= hp_Dict['Sound']['Mel_Dim'],
window_length= hp_Dict['Sound']['Frame_Length'],
hop_length= hp_Dict['Sound']['Frame_Shift'],
mel_fmin= hp_Dict['Sound']['Mel_F_Min'],
mel_fmax= hp_Dict['Sound']['Mel_F_Max'],
max_abs_value= hp_Dict['Sound']['Max_Abs_Mel']
)
style_Mel = Mel_Generate(
audio= Audio_Prep(style_Path, hp_Dict['Sound']['Sample_Rate']),
sample_rate= hp_Dict['Sound']['Sample_Rate'],
num_frequency= hp_Dict['Sound']['Spectrogram_Dim'],
num_mel= hp_Dict['Sound']['Mel_Dim'],
window_length= hp_Dict['Sound']['Frame_Length'],
hop_length= hp_Dict['Sound']['Frame_Shift'],
mel_fmin= hp_Dict['Sound']['Mel_F_Min'],
mel_fmax= hp_Dict['Sound']['Mel_F_Max'],
max_abs_value= hp_Dict['Sound']['Max_Abs_Mel']
)
pattern = content_Mel, style_Mel, content_Label, style_Label
if hp_Dict['Train']['Use_Pattern_Cache']:
self.cache_Dict[idx] = pattern
return pattern
def Pattern_Generate(audio=None,
path=None,
keyword_Index_Dict=None,
top_db=60,
reverse=False,
invert=False):
audio = audio if not audio is None else Audio_Prep(
path, hp_Dict['Sound']['Sample_Rate'], top_db)
if reverse:
audio = audio[::-1]
if invert:
audio = -audio
mel = Mel_Generate(audio=audio,
sample_rate=hp_Dict['Sound']['Sample_Rate'],
num_frequency=hp_Dict['Sound']['Spectrogram_Dim'],
num_mel=hp_Dict['Sound']['Mel_Dim'],
window_length=hp_Dict['Sound']['Frame_Length'],
hop_length=hp_Dict['Sound']['Frame_Shift'],
mel_fmin=hp_Dict['Sound']['Mel_F_Min'],
mel_fmax=hp_Dict['Sound']['Mel_F_Max'],
max_abs_value=hp_Dict['Sound']['Max_Abs_Mel'])
pitch = Pitch_Generate(audio)
singer_ID = None
if not keyword_Index_Dict is None:
for keyword, index in keyword_Index_Dict.items():
if keyword in path:
singer_ID = index
break
if singer_ID is None:
raise ValueError('No keyword in keyword_Index_Dict.')
return audio, mel, pitch, singer_ID
def Pattern_Generate(path, top_db=60):
audio = Audio_Prep(path, hp_Dict['Sound']['Sample_Rate'], top_db)
mel = Mel_Generate(audio=audio,
sample_rate=hp_Dict['Sound']['Sample_Rate'],
num_frequency=hp_Dict['Sound']['Spectrogram_Dim'],
num_mel=hp_Dict['Sound']['Mel_Dim'],
window_length=hp_Dict['Sound']['Frame_Length'],
hop_length=hp_Dict['Sound']['Frame_Shift'],
mel_fmin=hp_Dict['Sound']['Mel_F_Min'],
mel_fmax=hp_Dict['Sound']['Mel_F_Max'],
max_abs_value=hp_Dict['Sound']['Max_Abs_Mel'])
return audio, mel
def Pattern_Generate(path, top_db=60):
audio = Audio_Prep(path, hp.Sound.Sample_Rate, top_db)
mel = Mel_Generate(audio=audio,
sample_rate=hp.Sound.Sample_Rate,
num_frequency=hp.Sound.Spectrogram_Dim,
num_mel=hp.Sound.Mel_Dim,
window_length=hp.Sound.Frame_Length,
hop_length=hp.Sound.Frame_Shift,
mel_fmin=hp.Sound.Mel_F_Min,
mel_fmax=hp.Sound.Mel_F_Max,
max_abs_value=hp.Sound.Max_Abs_Mel)
pitch = Pitch_Generate(audio)
return audio, mel, pitch
def __getitem__(self, idx):
label, file = self.patterns[idx]
noisy = Audio_Prep(file, sample_rate=self.sample_Rate)
return noisy, label
def __getitem__(self, idx):
audio = Audio_Prep(self.files[idx], sample_rate=self.sample_Rate)
noise = Audio_Prep(choice(self.noises), sample_rate=self.sample_Rate)
return audio, noise