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)
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_invalid_otype(dummy_lpc, otype):
pysptk.lpc2lsp(dummy_lpc, otype=otype)
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係数を線スペクトル対に変換
lsp = pysptk.lpc2lsp(lpc, otype=0, fs=fs)
# 全極フィルタの作成
synthesizer = Synthesizer(LSPDF(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, lsp)
# 音声の書き込み
y = y.astype(np.int16)
wavfile.write(OUT_WAVE_FILE, fs, y)
def __dummy_windowed_frames(source, frame_len=512, hopsize=80):
np.random.seed(98765)
n_frames = int(len(source) / hopsize) + 1
windowed = np.random.randn(n_frames, frame_len) * pysptk.blackman(frame_len)
return 0.5 * 32768.0 * windowed
# audio, _ = lbr.load("data/SF1/100008.wav", sr=None, dtype=np.float64)
audio, _ = lbr.load(lbr.util.example_audio_file(), sr=None)
hopsize = 80
windowed = __dummy_windowed_frames(
audio, frame_len=frame_length, hopsize=hop_length)
frames = lbr.util.frame(audio, frame_length=frame_length, hop_length=hop_length).astype(np.float64).T
frames *= pysptk.hamming(frame_length, normalize=0)
print(type(frames), frames.shape)
# mcep = np.apply_along_axis(pysptk.mcep, 1, frames, order=order, alpha=alpha).T
lpcs = np.apply_along_axis(pysptk.lpc, 1, windowed, order=order)
lpcs1 = np.apply_along_axis(pysptk.lpc, 1, frames, order=order)
# for frame in frames:
# print(frame.shape)
# lpc = pysptk.lpc(frame)
print(type(lpcs), lpcs.shape)
lsp = pysptk.lpc2lsp(lpcs)
print(type(lsp), lsp.shape)
def __test_invalid_otype(dummy_lpc, otype):
pysptk.lpc2lsp(dummy_lpc, otype=otype)
