def file_to_sac(input_file):
audio,fs = sf.read(input_file)
vocals = np.array(audio[:,1])
f0 = pitch.extract_f0_sac(vocals, config.fs, 0.00580498866)
y=69+12*np.log2(f0/440)
# y = hertz_to_new_base(f0)
nans, x= nan_helper(y)
naners=np.isinf(y)
y[nans]= np.interp(x(nans), x(~nans), y[~nans])
# y=[float(x-(min_note-1))/float(max_note-(min_note-1)) for x in y]
y=np.array(y).reshape([len(y),1])
guy=np.array(naners).reshape([len(y),1])
y=np.concatenate((y,guy),axis=-1)
return y
def stft_to_feats(vocals, fs, mode=config.comp_mode):
feats=pw.wav2world(vocals,fs,frame_period=5.80498866)
ap = feats[2].reshape([feats[1].shape[0],feats[1].shape[1]]).astype(np.float32)
ap = 10.*np.log10(ap**2)
harm=10*np.log10(feats[1].reshape([feats[2].shape[0],feats[2].shape[1]]))
f0 = pitch.extract_f0_sac(vocals, fs, 0.00580498866)
y=69+12*np.log2(f0/440)
# import pdb;pdb.set_trace()
# y = hertz_to_new_base(f0)
nans, x= nan_helper(y)
naners=np.isinf(y)
y[nans]= np.interp(x(nans), x(~nans), y[~nans])
# y=[float(x-(min_note-1))/float(max_note-(min_note-1)) for x in y]
y=np.array(y).reshape([len(y),1])
guy=np.array(naners).reshape([len(y),1])
y=np.concatenate((y,guy),axis=-1)
if mode == 'mfsc':
harmy=sp_to_mfsc(harm,60,0.45)
apy=sp_to_mfsc(ap,4,0.45)
elif mode == 'mgc':
harmy=sp_to_mgc(harm,60,0.45)
apy=sp_to_mgc(ap,4,0.45)
# import pdb;pdb.set_trace()
out_feats=np.concatenate((harmy,apy,y.reshape((-1,2))),axis=1)
# harm_in=mgc_to_sp(harmy, 1025, 0.45)
# ap_in=mgc_to_sp(apy, 1025, 0.45)
return out_feats