def formant(self, val, f0_v):
'''
Change formant.
val : formant rate
f0_v: f0 rate
'''
f_rate = self.audio.frame_rate
np_arr = np.array(self.audio.get_array_of_samples(),
dtype=np.float64) # pydub --> np.array(float64) 変換
# print(np_arr, f_rate)
_f0_val, _time = pyworld.dio(np_arr, f_rate) # 基本周波数
spct = pyworld.cheaptrick(np_arr, _f0_val, _time, f_rate) # スペクトル包絡
aper = pyworld.d4c(np_arr, _f0_val, _time, f_rate) # 非周期性指標
spct_b = np.zeros_like(spct)
for i in range(spct_b.shape[1]):
spct_b[:, i] = spct[:, int(i / val)]
ef_audio = pyworld.synthesize(_f0_val * f0_v, spct_b, aper, f_rate)
ef_audio = ef_audio.astype(np.int16).tobytes()
# print(ef_audio)
# print(type(ef_audio))
new_audio = AudioSegment(
ef_audio,
sample_width=self.audio.sample_width,
frame_rate=f_rate,
channels=self.audio.channels,
)
self.audio = new_audio
return self
def low_pass_filter(self, fp, fs, g_pass, g_stop):
'''
Butterworth filter (low pass)
https://docs.scipy.org/doc/scipy/reference/generated/scipy.signal.butter.html
https://watlab-blog.com/2019/04/30/scipy-lowpass/
'''
samples = np.array(self.audio.get_array_of_samples())
channels = self.audio.channels
frame_rate = self.audio.frame_rate
# print(f'samples dim: {samples.ndim} / shape: {samples.shape}\nchannels: {channels}\nfame_rate: {frame_rate}')
fn = frame_rate / 2
wp = fp / fn
ws = fs / fn
N, Wn = signal.buttord(wp, ws, g_pass, g_stop)
b, a = signal.butter(N, Wn, 'low')
sample_res = np.zeros(samples.shape[0], dtype=np.float64)
for i in range(channels):
sample_m = samples[i::channels]
sample_res[i::channels] = signal.filtfilt(
b, a, sample_m) # チャンネルごとにフィルタを適用
# print(samples[1000000:1000010])
# print(sample_res[1000000:1000010])
# print(f'samples dim: {samples.ndim} / shape: {sample_res.shape}\nwidth: {self.audio.sample_width}')
res = AudioSegment(
sample_res.astype(np.int16).tobytes(),
sample_width=self.audio.sample_width,
frame_rate=frame_rate,
channels=channels,
)
return res