def matlab_fn_as_matlab_input_data(self, filename):
data = load_json(os.path.join('shrp', filename))
mag, index = two_max(data['x'],
int(data['lowerbound']) - 1,
int(data['upperbound']) - 1, data['unitlen'])
np.testing.assert_array_almost_equal(mag, data['mag'])
np.testing.assert_array_almost_equal(index, int(data['index']) - 1)
def test_wavread(self):
fn = sound_file_path('beijing_f3_50_a.wav')
samples, samples_int, Fs = wavread(fn)
expected = load_json(
os.path.join('helpers', 'beijing_f3_50_a-wavread-expected'))
self.assertEqual(Fs, expected['Fs'])
self.assertAllClose(samples, expected['y'], rtol=1e-05, atol=1e-08)
def test_with_matlab_data(self):
data = load_json(os.path.join('shrp', 'shr_pitch_data'))
wav_data, wavdata_int, fps = wavread(
sound_file_path('beijing_f3_50_a.wav'))
shr, f0 = shr_pitch(wav_data, fps, 25, 1, 50, 550, 0.4, 5, 200)
np.testing.assert_array_almost_equal(f0, data['F0'])
np.testing.assert_array_almost_equal(shr, data['SHR'])
def test_with_matlab_data(self):
data = load_json(os.path.join('shrp', 'toframes_data'))
res = toframes(data['input'],
data['curpos'].astype(int)-1,
int(data['segmentlen']),
'hamm')
np.testing.assert_array_almost_equal(res, data['frames'])
def test_with_matlab_data(self):
data = load_json(os.path.join('shrp', 'GetLogSpectrum_data'))
interp_amplitude = get_log_spectrum(data['segment'],
int(data['fftlen']),
int(data['limit']) - 1,
data['logf'], data['interp_logf'])
np.testing.assert_array_almost_equal(interp_amplitude,
data['interp_amplitude'])
def matlab_fn_as_matlab_input_data(self, filename):
data = load_json(os.path.join('shrp',filename))
mag, index = two_max(data['x'],
int(data['lowerbound'])-1,
int(data['upperbound'])-1,
data['unitlen'])
np.testing.assert_array_almost_equal(mag, data['mag'])
np.testing.assert_array_almost_equal(index, int(data['index'])-1)
def test_with_min_max_pitch_not_specified(self):
data = load_json(os.path.join('shrp', 'shr_pitch_data'))
wav_data, wavdata_int, fps = wavread(sound_file_path('beijing_f3_50_a.wav'))
with self.assertRaisesRegex(ValueError, 'none or both of min_pitch, max_pitch must be specified'):
shr, f0 = shr_pitch(wav_data, fps, min_pitch=50, datalen=200)
with self.assertRaisesRegex(ValueError, 'none or both of min_pitch, max_pitch must be specified'):
shr, f0 = shr_pitch(wav_data, fps, max_pitch=550, datalen=200)
def test_with_matlab_data(self):
data = load_json(os.path.join('shrp', 'GetLogSpectrum_data'))
interp_amplitude = get_log_spectrum(
data['segment'],
int(data['fftlen']),
int(data['limit']) - 1,
data['logf'],
data['interp_logf'])
np.testing.assert_array_almost_equal(interp_amplitude,
data['interp_amplitude'])
def test_second_peak_index(self):
data = load_json(os.path.join('shrp', 'twomax_data'))
x = data['x']
for i in range(142, 146):
# Zap the values in the second peak range to exercise the branch.
x[i] = 0.0001 * i
mag, index = two_max(x,
int(data['lowerbound']) - 1,
int(data['upperbound']) - 1, data['unitlen'])
np.testing.assert_array_almost_equal(mag,
np.append(data['mag'], 0.0145))
np.testing.assert_array_equal(index, [int(data['index']) - 1, 145])
def test_med_smooth_greater_than_zero(self):
data = load_json(os.path.join('shrp', 'shrp_data'))
with self.assertRaises(NotImplementedError):
f0_time, f0_value, shr, f0_candidates = shrp(
data['Y'],
int(data['Fs']), [int(x) for x in data['F0MinMax']],
int(data['frame_length']),
int(data['timestep']),
data['SHR_Threshold'],
data['ceiling'],
med_smooth=5,
CHECK_VOICING=False)
def test_with_matlab_data(self):
data = load_json(os.path.join('shrp', 'shrp_data'))
f0_time, f0_value, shr, f0_candidates = shrp(
data['Y'], int(data['Fs']), [int(x) for x in data['F0MinMax']],
int(data['frame_length']), int(data['timestep']),
data['SHR_Threshold'], data['ceiling'], data['med_smooth'],
data['CHECK_VOICING'])
np.testing.assert_array_almost_equal(f0_time, data['f0_time'])
np.testing.assert_array_almost_equal(f0_value, data['f0_value'])
np.testing.assert_array_almost_equal(shr, data['SHR'])
np.testing.assert_array_almost_equal(f0_candidates,
data['f0_candidates'])
def matlab_fn_as_matlab_input_data(self, filename):
data = load_json(os.path.join('shrp', filename))
peak_index, shr, shshift, index = compute_shr(
data['log_spectrum'], data['min_bin'],
data['startpos'].astype(int) - 1, data['endpos'].astype(int) - 1,
int(data['lowerbound']) - 1,
int(data['upperbound']) - 1, int(data['N']),
int(data['shift_units']), data['SHR_Threshold'])
np.testing.assert_array_equal(peak_index, int(data['peak_index']) - 1)
np.testing.assert_array_almost_equal(shr, data['SHR'])
np.testing.assert_array_almost_equal(shshift, data['shshift'])
np.testing.assert_array_almost_equal(index, data['index'] - 1)
def test_second_peak_index(self):
data = load_json(os.path.join('shrp', 'twomax_data'))
x = data['x']
for i in range(142, 146):
# Zap the values in the second peak range to exercise the branch.
x[i] = 0.0001*i
mag, index = two_max(x,
int(data['lowerbound'])-1,
int(data['upperbound'])-1,
data['unitlen'])
np.testing.assert_array_almost_equal(mag,
np.append(data['mag'], 0.0145))
np.testing.assert_array_equal(index, [int(data['index'])-1, 145])
def test_med_smooth_greater_than_zero(self):
data = load_json(os.path.join('shrp', 'shrp_data'))
with self.assertRaises(NotImplementedError):
f0_time, f0_value, shr, f0_candidates = shrp(
data['Y'],
int(data['Fs']),
[int(x) for x in data['F0MinMax']],
int(data['frame_length']),
int(data['timestep']),
data['SHR_Threshold'],
data['ceiling'],
med_smooth=5,
CHECK_VOICING=False)
def test_with_min_max_pitch_not_specified(self):
data = load_json(os.path.join('shrp', 'shr_pitch_data'))
wav_data, wavdata_int, fps = wavread(
sound_file_path('beijing_f3_50_a.wav'))
with self.assertRaisesRegex(
ValueError,
'none or both of min_pitch, max_pitch must be specified'):
shr, f0 = shr_pitch(wav_data, fps, min_pitch=50, datalen=200)
with self.assertRaisesRegex(
ValueError,
'none or both of min_pitch, max_pitch must be specified'):
shr, f0 = shr_pitch(wav_data, fps, max_pitch=550, datalen=200)
def test_resample_data_against_matlab(self):
# XXX: This test fails because SciPy is using a different algorithm
# for resampling than Matlab. When the SciPy and Matlab
# resampled data are plotted against each other, they look
# reasonably similar, so the fact that this test fails is
# probably okay.
for fn in wav_fns:
t = self.tmpdir()
tmp_path = os.path.join(t, os.path.basename(fn))
shutil.copy(fn, tmp_path)
s = SoundFile(tmp_path, resample_freq=16000)
resample_fn = os.path.splitext(os.path.basename(fn))[0] + '-matlab-resample'
data = load_json(os.path.join('soundfile', 'resample', resample_fn))
self.assertEqual(len(s.wavdata_rs), len(data['y_rs']))
self.assertAllClose(s.wavdata_rs, data['y_rs'], rtol=1e-01)
def matlab_fn_as_matlab_input_data(self, filename):
data = load_json(os.path.join('shrp', filename))
peak_index, shr, shshift, index = compute_shr(
data['log_spectrum'],
data['min_bin'],
data['startpos'].astype(int)-1,
data['endpos'].astype(int)-1,
int(data['lowerbound'])-1,
int(data['upperbound'])-1,
int(data['N']),
int(data['shift_units']),
data['SHR_Threshold'])
np.testing.assert_array_equal(peak_index, int(data['peak_index'])-1)
np.testing.assert_array_almost_equal(shr, data['SHR'])
np.testing.assert_array_almost_equal(shshift, data['shshift'])
np.testing.assert_array_almost_equal(index, data['index']-1)
def test_with_matlab_data(self):
data = load_json(os.path.join('shrp', 'shrp_data'))
f0_time, f0_value, shr, f0_candidates = shrp(
data['Y'],
int(data['Fs']),
[int(x) for x in data['F0MinMax']],
int(data['frame_length']),
int(data['timestep']),
data['SHR_Threshold'],
data['ceiling'],
data['med_smooth'],
data['CHECK_VOICING'])
np.testing.assert_array_almost_equal(f0_time, data['f0_time'])
np.testing.assert_array_almost_equal(f0_value, data['f0_value'])
np.testing.assert_array_almost_equal(shr, data['SHR'])
np.testing.assert_array_almost_equal(f0_candidates,
data['f0_candidates'])
def test_wavread(self):
fn = sound_file_path('beijing_f3_50_a.wav')
samples, samples_int, Fs = wavread(fn)
expected = load_json(os.path.join('helpers', 'beijing_f3_50_a-wavread-expected'))
self.assertEqual(Fs, expected['Fs'])
self.assertAllClose(samples, expected['y'], rtol=1e-05, atol=1e-08)
def test_with_matlab_data(self):
data = load_json(os.path.join('shrp', 'toframes_data'))
res = toframes(data['input'], data['curpos'].astype(int) - 1,
int(data['segmentlen']), 'hamm')
np.testing.assert_array_almost_equal(res, data['frames'])
def main(wav_dir, fs_rs):
"""Compare original data vs resampled data for all wav files in wav_dir,
where resampling frequency is given in Hz by fs_rs
"""
# Find all .wav files in test/data directory
wav_files = glob.glob(os.path.join(wav_dir, '*.wav'))
for wav_file in wav_files:
print('Processing wav file {}'.format(wav_file))
# y is data points, fs is sampling frequency
y, y_int, fs = wavread(wav_file)
# ns is number of samples
ns = len(y)
period = 1.0 / fs
# Time points corresponding to samples
t = np.arange(0, ns*period, period)
# Resample to 16 kHz
period_rs = 1.0 / fs_rs
# Number of points in resample
ns_rs = np.int_(np.ceil(ns * fs_rs / fs))
# Do resample
y_rs, t_rs = resample(y, ns_rs, t)
y_h, t_h = resample(y, ns_rs, t, window='hamming')
# Number of points to plot
n = 1000
# Start plotting from this data point
# Choose random starting point
s = randint(10000, 35015-n)
fig = plt.figure()
fig.suptitle(os.path.basename(wav_file))
# Compare original data with resampled ata
ax = plt.subplot(2,1,1)
ax.plot(t, y, 'b-', markersize=1)
ax.plot(t_rs, y_rs, 'ro', markersize=3)
ax.set_xlim([s*period, (s+n)*period])
ax.set_ylabel('Amplitude')
ax.set_title('Normal resample')
# Compare original data with resampling + Hamming window
ax = plt.subplot(2,1,2)
ax.plot(t, y, 'b-', markersize=1)
ax.plot(t_h, y_h, 'ro', markersize=3)
ax.set_xlim([s*period, (s+n)*period])
ax.set_ylabel('Amplitude')
ax.set_xlabel('Time (s)')
ax.set_title('Hamming window')
plt.savefig(os.path.splitext(os.path.basename(wav_file))[0] + '.pdf')
fn = os.path.splitext(os.path.basename(wav_file))[0] + '-matlab-resample'
data = load_json(os.path.join('soundfile', fn))
y_rs_matlab = data['y_rs']
# Compare SciPy resampled data with Matlab resampled data
plt.figure()
plt.plot(np.linspace(0, period * ns, len(y_rs_matlab)), y_rs_matlab, 'b-', markersize=1)
plt.plot(t_rs, y_rs, 'ro', markersize=3)
plt.xlim(s*period, (s+n)*period)
plt.title(os.path.basename(wav_file) + ' - Matlab comparison')
plt.xlabel('Time (s)')
plt.ylabel('Amplitude')
plt.savefig(os.path.splitext(os.path.basename(wav_file))[0] + '-matlab.pdf')
def test_with_matlab_data(self):
data = load_json(os.path.join('shrp', 'shr_pitch_data'))
wav_data, wavdata_int, fps = wavread(sound_file_path('beijing_f3_50_a.wav'))
shr, f0 = shr_pitch(wav_data, fps, 25, 1, 50, 550, 0.4, 5, 200)
np.testing.assert_array_almost_equal(f0, data['F0'])
np.testing.assert_array_almost_equal(shr, data['SHR'])
def main(wav_dir, fs_rs):
"""Compare original data vs resampled data for all wav files in wav_dir,
where resampling frequency is given in Hz by fs_rs
"""
# Find all .wav files in test/data directory
wav_files = glob.glob(os.path.join(wav_dir, '*.wav'))
for wav_file in wav_files:
print('Processing wav file {}'.format(wav_file))
# y is data points, fs is sampling frequency
y, y_int, fs = wavread(wav_file)
# ns is number of samples
ns = len(y)
period = 1.0 / fs
# Time points corresponding to samples
t = np.arange(0, ns * period, period)
# Resample to 16 kHz
period_rs = 1.0 / fs_rs
# Number of points in resample
ns_rs = np.int_(np.ceil(ns * fs_rs / fs))
# Do resample
y_rs, t_rs = resample(y, ns_rs, t)
y_h, t_h = resample(y, ns_rs, t, window='hamming')
# Number of points to plot
n = 1000
# Start plotting from this data point
# Choose random starting point
s = randint(10000, 35015 - n)
fig = plt.figure()
fig.suptitle(os.path.basename(wav_file))
# Compare original data with resampled ata
ax = plt.subplot(2, 1, 1)
ax.plot(t, y, 'b-', markersize=1)
ax.plot(t_rs, y_rs, 'ro', markersize=3)
ax.set_xlim([s * period, (s + n) * period])
ax.set_ylabel('Amplitude')
ax.set_title('Normal resample')
# Compare original data with resampling + Hamming window
ax = plt.subplot(2, 1, 2)
ax.plot(t, y, 'b-', markersize=1)
ax.plot(t_h, y_h, 'ro', markersize=3)
ax.set_xlim([s * period, (s + n) * period])
ax.set_ylabel('Amplitude')
ax.set_xlabel('Time (s)')
ax.set_title('Hamming window')
plt.savefig(os.path.splitext(os.path.basename(wav_file))[0] + '.pdf')
fn = os.path.splitext(
os.path.basename(wav_file))[0] + '-matlab-resample'
data = load_json(os.path.join('soundfile', fn))
y_rs_matlab = data['y_rs']
# Compare SciPy resampled data with Matlab resampled data
plt.figure()
plt.plot(np.linspace(0, period * ns, len(y_rs_matlab)),
y_rs_matlab,
'b-',
markersize=1)
plt.plot(t_rs, y_rs, 'ro', markersize=3)
plt.xlim(s * period, (s + n) * period)
plt.title(os.path.basename(wav_file) + ' - Matlab comparison')
plt.xlabel('Time (s)')
plt.ylabel('Amplitude')
plt.savefig(
os.path.splitext(os.path.basename(wav_file))[0] + '-matlab.pdf')