def __call__(self, src: str) -> tuple:
speech, sr = sf.read(src)
lowcut, highcut, order = self.calc_params()
filtered_speech = butter_bandpass_filter(speech, lowcut, highcut, sr,
order)
filtered_normalized = normalize(filtered_speech)
return filtered_normalized, sr
def __call__(self, src) -> tuple:
impulse_response, sr_imp = self.load_impulse()
speech, sr_speech = sf.read(src, always_2d=True)
wet_speech = self.convolve_impulse_response(speech, impulse_response, sr_speech, sr_imp)
wet_speech = wet_speech * self.cfg.artifact_scaling_factor
mixed = np.add(speech, wet_speech)
mixed = normalize(mixed)
mono_mixed = stereo_to_mono(mixed)
return mono_mixed, sr_speech
def mix_samples(sample_a: np.ndarray, sample_b: np.ndarray, trim_to: str = "a") -> np.ndarray:
"""
@param sample_a: first audio-sample
@param sample_b: second audio-sample
@param trim_to: resulting length will be length of sample_a or sample_b,
if the sample not specified is shorter than the specified one,
it will be repeated to match the specified samples length
@return: sum of both samples, normalized to avoid clipping
"""
if trim_to == "b":
sample_a, sample_b = sample_b, sample_a
sample_b = match_length(sample_b, sample_a.shape[0])
add = np.add(sample_a, sample_b)
return normalize(add)
def read_audio(filename, scipy_=True, normalization=True):
if scipy_:
audio, sample_rate = wavread(filename)
audio = audio.astype('float32')
if audio.ndim >= 2:
audio = np.mean(audio, 1)
else:
with soundfile.SoundFile(filename) as sound_file:
audio = sound_file.read(dtype='float32')
sample_rate = sound_file.samplerate
if audio.ndim >= 2:
audio = np.mean(audio, 1)
if normalization:
audio = normalize(audio, sample_rate)
return audio, sample_rate
def test_normalize(self):
sample = 2 * np.random.random_sample(128)
res = normalize(sample)
assert ((res >= -1).all() and (res <= 1).all())