def test_add_noise_rms():
x, fs = wavread('tests/sp10.wav')
n, _ = wavread('tests/ssn.wav')
y, n_scaled = add_noise(x, n, fs, 5.0)
snr = rms_energy(x) - rms_energy(n_scaled)
assert np.isclose(snr, 5.0)
def test_add_noise_p56():
x, fs = wavread('tests/sp10.wav')
n, _ = wavread('tests/ssn.wav')
y, n_scaled = add_noise(x, n, fs, 5.0, speech_energy='P.56')
snr = asl_meter(x, fs) - rms_energy(n_scaled)
assert np.isclose(snr, 5.0)
def test_add_reverb():
x, fs = wavread('tests/sp10.wav')
r, _ = wavread('tests/rir.wav')
y = add_reverb(x, r, fs, speech_energy='P.56')
y_ref, _ = wavread('tests/sp10_reverb_ref.wav')
assert np.allclose(y, y_ref, atol=2e-3)
def generate_condition(self,
snrs,
noise,
output_dir,
reverb=None,
files_per_condition=None):
if noise not in self.noise.keys():
raise ValueError('noise not in dataset')
if type(snrs) is not list:
snrs = [snrs]
n, nfs = wavread(self.noise[noise])
if reverb is not None:
r, rfs = wavread(self.reverb[reverb])
condition_name = '{}_{}'.format(reverb, noise)
else:
condition_name = noise
if not os.path.isdir(output_dir):
os.mkdir(output_dir)
# FIXME: avoid overwriting an existing folder?
try:
for snr in snrs:
os.mkdir(
os.path.join(output_dir,
'{}_{}dB'.format(condition_name, snr)))
except OSError:
print('Condition folder already exists!')
for snr in snrs:
if files_per_condition is not None:
speech_files = np.random.choice(self.speech,
files_per_condition,
replace=False).tolist()
else:
speech_files = self.speech
for f in tqdm(speech_files, desc='{}dB'.format(snr)):
x, fs = wavread(f)
if fs != nfs:
raise ValueError(
'Speech file and noise file have different fs!')
if reverb is not None:
if fs != rfs:
raise ValueError(
'Speech file and reverb file have different fs!')
x = add_reverb(x, r, fs, speech_energy=self.speech_energy)
y = add_noise(x, n, fs, snr,
speech_energy=self.speech_energy)[0]
wavwrite(
os.path.join(output_dir,
'{}_{}dB'.format(condition_name, snr),
os.path.basename(f)), y, fs)
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
from scipy.fftpack import fft, ifft
from scipy import signal
from maracas.utils import wavread
from maracas import add_reverb
import numpy as np
import matplotlib.pyplot as plt
x, fs = wavread('sp10.wav')
assert np.allclose(fft(ifft(x)), x, atol=1e-12)
f, t, Sxx = signal.spectrogram(x, fs)
print('Shapes: {} and {}'.format(x.shape, (f.shape, t.shape, Sxx.shape)))
plt.pcolormesh(t, f, Sxx)
plt.ylabel('Frequency [Hz]')
plt.xlabel('Time [sec]')
plt.show()
r, _ = wavread('rir.wav')
n_x = add_reverb(x, r, fs, speech_energy='P.56')
f, t, Sxx = signal.spectrogram(n_x, fs)
print('Shapes: {} and {}'.format(x.shape, (f.shape, t.shape, Sxx.shape)))
plt.pcolormesh(t, f, Sxx)
plt.ylabel('Frequency [Hz]')
plt.xlabel('Time [sec]')
def test_asl_meter():
x, fs = wavread('tests/sp10.wav')
assert np.isclose(asl_meter(x, fs), -25.631381743520010)