def test_stft():
"""Test STFT calculation"""
assert_raises(ValueError, stft, 0)
assert_raises(ValueError, stft, [], window='foo')
assert_raises(ValueError, stft, [[]])
result = stft([])
assert np.allclose(result, np.zeros_like(result))
n_fft = 256
step = 128
for n_samples, n_estimates in ((256, 1), (383, 1), (384, 2), (511, 2),
(512, 3)):
result = stft(np.ones(n_samples), n_fft=n_fft, step=step, window=None)
assert result.shape[1] == n_estimates
expected = np.zeros(n_fft // 2 + 1)
expected[0] = 1
for res in result.T:
assert np.allclose(expected, np.abs(res))
assert np.allclose(expected, np.abs(res))
for n_pts, last_freq in zip((256, 255), (500., 498.)):
freqs = fft_freqs(n_pts, 1000)
assert freqs[0] == 0
assert np.allclose(freqs[-1], last_freq, atol=1e-1)
def test_stft():
"""Test STFT calculation"""
assert_raises(ValueError, stft, 0)
assert_raises(ValueError, stft, [], window='foo')
assert_raises(ValueError, stft, [[]])
result = stft([])
assert np.allclose(result, np.zeros_like(result))
n_fft = 256
step = 128
for n_samples, n_estimates in ((256, 1),
(383, 1), (384, 2),
(511, 2), (512, 3)):
result = stft(np.ones(n_samples), n_fft=n_fft, step=step, window=None)
assert result.shape[1] == n_estimates
expected = np.zeros(n_fft // 2 + 1)
expected[0] = 1
for res in result.T:
assert np.allclose(expected, np.abs(res))
assert np.allclose(expected, np.abs(res))
for n_pts, last_freq in zip((256, 255), (500., 498.)):
freqs = fft_freqs(n_pts, 1000)
assert freqs[0] == 0
assert np.allclose(freqs[-1], last_freq, atol=1e-1)
def _calculate_spectrogram(self):
if self._x is not None:
x = self._x
nan_mean = np.nanmean(x)
idx = np.isnan(x)
x[idx] = nan_mean
data = stft(x, self._n_fft, self._step, self._fs, self._window)
data = np.abs(data)
data = 20 * np.log10(data) if self._color_scale == 'log' else data
if self._normalize:
for i in range(data.shape[0]):
data[i, :] -= np.mean(data[i, :])
data[i, :] /= np.std(data[i, :])
return data
else:
return None
def _do_spec(self, min_time, max_time):
if min_time < 0:
min_time = 0
if max_time - min_time > 30:
step = 0.1
else:
step = self.step
step = (max_time - min_time) / 1000
min_samp = int(min_time * self._sr)
max_samp = int(max_time * self._sr)
self.xscale = step
step_samp = int(step * self._sr)
#if step_samp < 28:
# step_samp = 28
# step = step_samp / self._sr
if self._n_fft < 32:
self._n_fft = 32
#self._n_fft = 512
data = stft(self._signal[min_samp:max_samp], self._n_fft, step_samp, self._sr, self._window)
data = np.abs(data)
data = 20 * np.log10(data) if self._color_scale == 'log' else data
return data
