def __test(order):
a1 = pysptk.lpc(x, order, use_scipy=False)
a2 = pysptk.lpc(x, order, use_scipy=True)
a3 = pysptk.levdur(pysptk.acorr(x, order), use_scipy=False)
a4 = pysptk.levdur(pysptk.acorr(x, order), use_scipy=True)
assert np.all(np.isfinite(a1))
assert np.allclose(a1, a2)
assert np.allclose(a1, a3)
assert np.allclose(a1, a4)
def test_lpc(order):
x = windowed_dummy_data(1024)
a1 = pysptk.lpc(x, order, use_scipy=False)
a2 = pysptk.lpc(x, order, use_scipy=True)
a3 = pysptk.levdur(pysptk.acorr(x, order), use_scipy=False)
a4 = pysptk.levdur(pysptk.acorr(x, order), use_scipy=True)
assert np.all(np.isfinite(a1))
assert np.allclose(a1, a2)
assert np.allclose(a1, a3)
assert np.allclose(a1, a4)
def task3():
def lpc(data,order=10):
data=data.astype("int64")
acf=sig.correlate(in1=data, in2=data, mode="full",method="direct")
mid=len(acf)//2
matrix=np.array([acf[mid-i:mid-i+order] for i in range(order)])
print(matrix.dtype)
gammavec=acf[mid+1:mid+1+order]
sol=la.solve(matrix,gammavec)
print(sol)
return sol
rate,data_a=wav.read(f"oving8/vowels/a.wav")
_, data_ae=wav.read("oving8/vowels/ae.wav")
a = lpc(data_a)
b = lpc(data_ae)
filtered_a=sig.lfilter(b, a, data_ae)
wav.write("filtered_a.wav",rate,filtered_a)
def test_lpc():
# frame -l 512 -p 80 < test16k.float | window -l 512 | dmp +f | awk \
# '{print $2}' > test16k_windowed.txt
frames = (np.loadtxt(join(DATA_DIR, "test16k_windowed.txt")).reshape(
759, 512).astype(np.float64))
# frame -l 512 -p 80 < test16k.float | window -l 512 | lpc -m 25 -l 512 > test16k.lpc
lpc = (np.fromfile(join(DATA_DIR, "test16k.lpc"),
np.float32).reshape(759, 26).astype(np.float64))
lpc_hat = pysptk.lpc(frames, order=25)
# yeah may have a bug...
assert np.allclose(lpc, lpc_hat, atol=1e-1)
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)
lpc = pysptk.lpc(windowed, filt.order)
lsp = pysptk.lpc2lsp(lpc)
# make sure lsp has loggain
lsp[:, 0] = np.log(lsp[:, 0])
# synthesis
synthesizer = Synthesizer(filt, hopsize)
y = synthesizer.synthesis(source, lsp)
assert np.all(np.isfinite(y))
def test_lpc2lsp():
for order in [15, 20, 25, 30]:
yield __test_transform_base, pysptk.lpc2lsp, order
def __test_invalid_otype(dummy_lpc, otype):
pysptk.lpc2lsp(dummy_lpc, otype=otype)
np.random.seed(98765)
dummy_lpc = pysptk.lpc(np.random.rand(512), 21)
# invalid otype
yield raises(ValueError)(__test_invalid_otype), dummy_lpc, 2
yield raises(ValueError)(__test_invalid_otype), dummy_lpc, 3
lsp1 = pysptk.lpc2lsp(dummy_lpc, otype=2, fs=16000)
lsp2 = pysptk.lpc2lsp(dummy_lpc, otype=3, fs=16)
assert np.allclose(lsp1, lsp2)
# loggain
lsp3 = pysptk.lpc2lsp(dummy_lpc, otype=3, fs=16, loggain=True)
assert lsp3[0] == np.log(lsp2[0])
def test_lpc2lsp():
for order in [15, 20, 25, 30]:
yield __test_transform_base, pysptk.lpc2lsp, order
def __test_invalid_otype(dummy_lpc, otype):
pysptk.lpc2lsp(dummy_lpc, otype=otype)
np.random.seed(98765)
dummy_lpc = pysptk.lpc(np.random.rand(512), 21)
# invalid otype
yield raises(ValueError)(__test_invalid_otype), dummy_lpc, 2
yield raises(ValueError)(__test_invalid_otype), dummy_lpc, 3
lsp1 = pysptk.lpc2lsp(dummy_lpc, otype=2, fs=16000)
lsp2 = pysptk.lpc2lsp(dummy_lpc, otype=3, fs=16)
assert np.allclose(lsp1, lsp2)
# loggain
lsp3 = pysptk.lpc2lsp(dummy_lpc, otype=3, fs=16, loggain=True)
assert lsp3[0] == np.log(lsp2[0])
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)
lpc = pysptk.lpc(windowed, filt.order)
lpc[:, 0] = 0
b = -lpc
# 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))
import pysptk
from scipy.io import wavfile
import librosa.util
frame_length = 2048
hop_length = 512
order = 20
path = '../ETTS_newdata/data/wav/lmy00001.wav'
# LPC
sr, x = wavfile.read(path)
x = x.astype(np.float64)
librosa.util.valid_audio(x)
x = np.pad(x, int(frame_length // 2), mode='reflect')
frames = librosa.util.frame(x,
frame_length=frame_length,
hop_length=hop_length).astype(np.float64).T
frames *= pysptk.blackman(frame_length)
lpc = pysptk.lpc(frames, order)
lpc[:, 0] = np.log(lpc[:, 0])
#MFCC
y, sr = librosa.load(path)
y = y.astype(np.float64)
mfcc = librosa.feature.mfcc(y=y,
sr=sr,
n_fft=frame_length,
hop_length=hop_length)
print(d)
hop_length=HOP_LENGTH).astype(np.float64).T
frames *= pysptk.blackman(FRAME_LENGTH) # 窓掛け（ブラックマン窓）
# ピッチ抽出
pitch = pysptk.swipe(x,
fs=fs,
hopsize=HOP_LENGTH,
min=MIN_F0,
max=MAX_F0,
otype="pitch")
# 励振源信号(声帯音源)の生成
source_excitation = pysptk.excite(pitch, HOP_LENGTH)
# 線形予測分析による線形予測符号化(LPC)係数の抽出
lpc = pysptk.lpc(frames, ORDER)
lpc[:, 0] = np.log(lpc[:, 0])
# LPC係数をPARCOR係数に変換
parcor = pysptk.lpc2par(lpc)
# 全極フィルタの作成
synthesizer = Synthesizer(AllPoleLatticeDF(order=ORDER), HOP_LENGTH)
# 励振源信号でフィルタを駆動して音声を合成
y = synthesizer.synthesis(source_excitation, parcor)
# 音声の書き込み
y = y.astype(np.int16)
wavfile.write(OUT_WAVE_FILE, fs, y)
def test_lpc_failure():
pysptk.lpc(np.zeros(256), 40, use_scipy=False)
def __test_min_det(min_det):
pysptk.lpc(x, min_det=min_det, use_scipy=False)
def __test(order):
a = pysptk.lpc(x, order)
assert np.all(np.isfinite(a))
def test_lpc_failure():
with pytest.raises(RuntimeError):
pysptk.lpc(np.zeros(256), 40, use_scipy=False)
print(p)
x[-4] = 0
x[-3] = 0
x[-2] = 0
x[-1] = 0
sumi = 0
for j in range(101):
for i in range(4):
sumi += p * x[j]
print(x[j] - sumi)
np.random.seed(1)
N = 10
b1 = np.random.rand(N)
b2 = np.random.rand(N)
X = np.column_stack([x, x])
X -= X.mean(axis=0)
fact = N - 1
by_hand = np.dot(X.T, X.conj()) / fact
print(by_hand)
import seaborn
import scipy
import pysptk
from pysptk.synthesis import AllPoleDF
lpc = pysptk.lpc(x, 4)
print(lpc)
def test_lpc_failure():
pysptk.lpc(np.zeros(256), 40)
def __test_min_det(min_det):
pysptk.lpc(x, min_det=min_det)
