def Mel_Generate(path, top_db=60):
sig = librosa.core.load(path, sr=hp_Dict['Sound']['Sample_Rate'])[0]
sig = preemphasis(sig)
sig = librosa.effects.trim(
sig, top_db=top_db, frame_length=32, hop_length=16)[0] * 0.99
return np.transpose(
melspectrogram(y=sig,
num_freq=hp_Dict['Sound']['Spectrogram_Dim'],
hop_length=hp_Dict['Sound']['Frame_Shift'],
win_length=hp_Dict['Sound']['Frame_Length'],
num_mels=hp_Dict['Sound']['Mel_Dim'],
sample_rate=hp_Dict['Sound']['Sample_Rate'],
max_abs_value=hp_Dict['Sound']['Max_Abs_Mel']).astype(
np.float32))
def Spectrogram_Generate(path, top_db=60, range_Ignore=False):
sig = librosa.core.load(path, sr=hp_Dict['Sound']['Sample_Rate'])[0]
sig = preemphasis(sig)
sig = librosa.effects.trim(
sig, top_db=top_db, frame_length=32, hop_length=16)[0] * 0.99
sig = inv_preemphasis(sig)
sig_Length = sig.shape[0] / hp_Dict['Sound']['Sample_Rate'] * 1000 #ms
if not range_Ignore and (sig_Length < hp_Dict['Train']['Min_Wav_Length'] or
sig_Length > hp_Dict['Train']['Max_Wav_Length']):
return None
return np.transpose(
spectrogram(y=sig,
num_freq=hp_Dict['Sound']['Spectrogram_Dim'],
frame_shift_ms=hp_Dict['Sound']['Frame_Shift'],
frame_length_ms=hp_Dict['Sound']['Frame_Length'],
sample_rate=hp_Dict['Sound']['Sample_Rate']).astype(
np.float32))