def pysptk_featurize(audiofile):
labels = list()
features = list()
fs, x = wavfile.read(audiofile)
f0_swipe = pysptk.swipe(x.astype(np.float64),
fs=fs,
hopsize=80,
min=60,
max=200,
otype="f0")
features = features + stats(f0_swipe)
labels = stats_labels('f0_swipe', labels)
f0_rapt = pysptk.rapt(x.astype(np.float32),
fs=fs,
hopsize=80,
min=60,
max=200,
otype="f0")
features = features + stats(f0_rapt)
labels = stats_labels('f0_rapt', labels)
mgc = pysptk.mgcep(xw, 20, 0.0, 0.0)
features = features + stats(mgc)
labels = stats_labels('mel-spectrum envelope', labels)
return features, labels
def readmgc(filename):
# all parameters can adjust by yourself :)
try:
sr, x = wavfile.read(filename)
if(sr != 44100):
data, samplerate = soundfile.read(filename)
soundfile.write(filename, data, 44100, subtype='PCM_16')
except ValueError:
data, samplerate = soundfile.read(filename)
soundfile.write(filename, data, 44100, subtype='PCM_16')
sr, x = wavfile.read(filename)
assert sr == 44100
x = x.astype(np.float64)
frame_length = 1024
hop_length = 256
# Windowing
frames = librosa.util.frame(x, frame_length=frame_length, hop_length=hop_length).astype(np.float64).T
frames *= pysptk.blackman(frame_length)
assert frames.shape[1] == frame_length
# Order of mel-cepstrum
order = 25
alpha = 0.41
stage = 5
gamma = -1.0 / stage
mgc = pysptk.mgcep(frames, order, alpha, gamma)
mgc = mgc.reshape(-1, order + 1)
print("mgc of {} is ok!".format(filename))
return mgc
def extract_mgc(amp_sp: np.array,
fs: int = None,
num_coded_sps: int = 60,
mgc_alpha: float = None) -> np.array:
"""Extract MGC from the amplitude spectrum from SPTK."""
if mgc_alpha is None:
assert fs is not None, "Either sampling rate or mgc alpha has to be given."
mgc_alpha = AudioProcessing.fs_to_mgc_alpha(fs)
mgc = pysptk.mgcep(amp_sp,
order=num_coded_sps - 1,
alpha=mgc_alpha,
gamma=AudioProcessing.mgc_gamma,
eps=1.0e-8,
min_det=0.0,
etype=1,
itype=3)
return mgc.astype(np.float32, copy=False)
def __test_synthesis(filt):
# dummy source excitation
source = __dummy_source()
hopsize = 80
# dummy filter coef.
windowed = __dummy_windowed_frames(source,
frame_len=512,
hopsize=hopsize)
gamma = -1.0 / filt.stage
mgc = pysptk.mgcep(windowed, filt.order, filt.alpha, gamma)
b = pysptk.mgc2b(mgc, filt.alpha, gamma)
# synthesis
synthesizer = Synthesizer(filt, hopsize)
y = synthesizer.synthesis(source, b)
assert np.all(np.isfinite(y))
# transpose
synthesizer = Synthesizer(filt, hopsize, transpose=True)
y = synthesizer.synthesis(source, b)
assert np.all(np.isfinite(y))
def readmgc(filename):
# all parameters can adjust by yourself :)
sr, x = wavfile.read(filename)
assert sr == 16000 # Frequency sampling value
x = x.astype(np.float64)
frame_length = 1024
hop_length = 256
# Windowing
frames = librosa.util.frame(x,
frame_length=frame_length,
hop_length=hop_length).astype(np.float64).T
frames *= pysptk.blackman(frame_length)
assert frames.shape[1] == frame_length
# Order of mel-cepstrum
order = 25
alpha = 0.41
stage = 5
gamma = -1.0 / stage
mgc = pysptk.mgcep(frames, order, alpha, gamma)
mgc = mgc.reshape(-1, order + 1)
print("mgc of {} is ok!".format(filename))
return mgc
def test_mgcep_failure():
pysptk.mgcep(np.ones(256))
def test_mgcep_failure():
with pytest.raises(RuntimeError):
pysptk.mgcep(np.ones(256))
def test_mgcep_failure():
pysptk.mgcep(np.ones(256))
def __test_min_det(min_det):
pysptk.mgcep(x, min_det=min_det)
def __test_itype(itype=0):
pysptk.mgcep(x, itype=itype)
def __test_gamma(gamma):
pysptk.mgcep(x, gamma=gamma)
def __test(order, alpha, gamma):
mgc = pysptk.mgcep(x, order, alpha, gamma)
assert np.all(np.isfinite(mgc))
# Clean and gen should have the same length, and be 1D
dis = stoi.stoi(gtaud, degaud, sr, extended=False)
st.append(dis)
#Compute MCD
num=np.int(np.shape(degaud)[0]/1024)-5
#print(np.shape(gtaud), np.shape(degaud), sr, num)
costFn=logSpecDbDist
frame=1
nat=[]
for ij in range(1,num):
x=gtaud[1024*ij:1024*(ij+1)]
#print(np.shape(x),x)
try:
mgc_temp = pysptk.mgcep(x)
except:
print("Error at: ", ij)
else:
nat.append(mgc_temp)
if 1:
#print(np.shape(mgc_temp))
#degaud=np.reshape(degaud[0:shp],(shp,1))
synth=[]
for kidx in range(1,num):
try:
x=degaud[1024*kidx:1024*(kidx+1)]
#print(np.shape(x),x)
mgc_temp = pysptk.mgcep(x)
except:
def mel_generalized_cepstrum(frames):
mgc = ps.mgcep(frames, ORDER, ALPHA, GAMMA)
return mgc
def __test(order, alpha, gamma):
mgc = pysptk.mgcep(x, order, alpha, gamma)
assert np.all(np.isfinite(mgc))
def __test_gamma(gamma):
pysptk.mgcep(x, gamma=gamma)
def __test_itype(itype=0):
pysptk.mgcep(x, itype=itype)
def __test_eps(etype=0, eps=0.0):
pysptk.mgcep(x, etype=etype, eps=eps)
def __test_eps(etype=0, eps=0.0):
pysptk.mgcep(x, etype=etype, eps=eps)
def __test_otype(otype=0):
pysptk.mgcep(x, otype=otype)
def __test_min_det(min_det):
pysptk.mgcep(x, min_det=min_det)
def __test_otype(otype=0):
pysptk.mgcep(x, otype=otype)
def test_mgcep(order, alpha, gamma):
x = windowed_dummy_data(1024)
mgc = pysptk.mgcep(x, order, alpha, gamma)
assert np.all(np.isfinite(mgc))
