def test_mono_audio_file_construction(self, shared_datadir):
file_name = (shared_datadir / 'audio/test_sine.wav').resolve()
audio = AudioBuffer(file_name)
assert audio.get_audio().shape == (44100,)
assert audio.get_sample_rate() == 44100
assert audio.channels == 1
assert audio.file_name == file_name
def test_list_construction_stereo(self):
samples = [[1,2,3,4,5],[6,7,8,9,10]]
audio = AudioBuffer(samples, 44100)
assert np.array_equal(audio.get_audio(), samples)
assert audio.get_sample_rate() == 44100
assert audio.channels == 2
assert audio.file_name == ''
def test_array_construction_mono(self):
samples = np.array([1,2,3,4,5,6,7,8,9,10])
audio = AudioBuffer(samples, 44100)
assert np.array_equal(audio.get_audio(), samples)
assert audio.get_sample_rate() == 44100
assert audio.channels == 1
assert audio.file_name == ''
def test_replace_audio(self):
samples = np.array([[1,2,3,4,5],[6,7,8,9,10]])
audio = AudioBuffer(samples, 44100)
samplesB = np.array([10,9,8,7,6,5,4,3,2,1])
audio.replace_audio_data(samplesB, 96000)
assert np.array_equal(audio.get_audio(), samplesB)
assert audio.channels == 1
def test_plot_spectrogram(self):
samples = self.make_test_sine(44100, 150, 44100)
audio = AudioBuffer(samples, 44100)
fig = plt.figure()
ax = audio.plot_spectrogram()
plt.plot()
return fig
def test_sine_wave_stft_complex(self, shared_datadir):
file_name = (shared_datadir / 'audio/test_sine.wav').resolve()
audio = AudioBuffer()
audio.load(file_name)
stft = STFT()
features = stft(audio)
assert features.shape == (513, 87)
assert isinstance(features[0][0], np.complex64)
def test_sine_wave_stft_complex_time_major(self, shared_datadir):
file_name = (shared_datadir / 'audio/test_sine.wav').resolve()
audio = AudioBuffer()
audio.load(file_name)
stft = STFT(time_major=True, fft_size=2048, hop_size=1024)
features = stft(audio)
assert features.shape == (44, 1025)
assert isinstance(features[0][0], np.complex64)
def test_save_mono_create_dir(self, tmp_path):
samples = self.make_test_sine(44100, 440, 44100)
audio = AudioBuffer(samples, 44100)
file_name = (tmp_path / 'temp_dir/test_save.wav').resolve()
audio.save(file_name)
audioReload = AudioBuffer(file_name)
np.testing.assert_array_almost_equal(audio.get_audio(), audioReload.get_audio())
assert audioReload.get_sample_rate() == 44100
assert audioReload.channels == 1
assert audioReload.file_name == file_name
def test_sine_extraction_scale(self):
frame_size = 1024
hop_size = 512
bin_freq = 44100. / float(frame_size)
num_frames = 10
batch_size = 10
spectral = SpectralSummarized()
features = np.ndarray((
10,
22,
))
for i in range(batch_size):
sine = utils.make_test_cosine(22050, bin_freq * (i + 1), 44100)
audio = AudioBuffer(sine, 44100)
features[i] = spectral(audio)
assert features.shape == (10, 22)
scaled = spectral.fit_scaler(features)
assert scaled.mean() == pytest.approx(0.)
assert scaled.std() == pytest.approx(1.)
np.testing.assert_array_almost_equal(scaled.mean(0), np.zeros(22))
np.testing.assert_array_almost_equal(scaled.std(0), np.ones(22))
def test_sine_mfcc_time_major(self):
sine = utils.make_test_sine(2048, 440, 44100)
audio = AudioBuffer(sine, 44100)
mfcc = MFCC(hop_size=512, frame_size=1024, time_major=True)
features = mfcc(audio)
assert features.shape == (5, 20)
def test_load_folder_with_non_audio_stereo(self, shared_datadir):
file_name = (shared_datadir / 'audio').resolve()
audio = AudioBuffer.load_folder(file_name, mono=False)
assert len(audio) == 2
assert audio[0].get_audio().shape == (44100,)
assert audio[0].get_sample_rate() == 44100
assert audio[0].file_name == 'test_sine.wav'
assert audio[1].get_audio().shape == (2,44100)
assert audio[1].get_sample_rate() == 44100
assert audio[1].file_name == 'test_sine_stereo.wav'
def test_sine_wave_fft_magnitude_longer(self):
bin_freq = 44100. / 4096.
sine = utils.make_test_sine(4096, bin_freq * 10, 44100)
audio = AudioBuffer(sine, 44100)
fft = FFT(output='magnitude')
features = fft(audio)
expected = np.zeros(2049)
expected[10] = 2048.
assert features.shape == (2049, )
assert isinstance(features[0], np.float32)
np.testing.assert_array_almost_equal(features, expected)
def test_sine_wave_fft_power(self):
bin_freq = 44100. / 1024.
sine = utils.make_test_sine(1024, bin_freq * 10, 44100)
audio = AudioBuffer(sine, 44100)
fft = FFT(output='power')
features = fft(audio)
expected = np.zeros(513)
expected[10] = 512. * 512.
assert features.shape == (513, )
assert isinstance(features[0], np.float32)
np.testing.assert_array_almost_equal(features, expected)
def match_from_file(self, path):
"""
Load audio file from disk and perform sound matching on it
Args:
filepath (str): location of audio file on disk
Returns:
:ref:`AudioBuffer <audio_buffer>`: audio output from synthesizer after sound matching
"""
target = AudioBuffer(path, self.features.sample_rate)
return self.match(target)
def test_stereo_audio_file_load(self, shared_datadir):
file_name = (shared_datadir / 'audio/test_sine_stereo.wav').resolve()
audio = AudioBuffer()
audio.load(file_name, mono=False)
assert audio.get_audio().shape == (2, 44100)
assert audio.get_sample_rate() == 44100
assert audio.channels == 2
assert audio.file_name == file_name
def test_audio_file_load_resample(self, shared_datadir):
file_name = (shared_datadir / 'audio/test_sine.wav').resolve()
audio = AudioBuffer()
audio.load(file_name, 88200)
assert audio.get_audio().shape == (88200,)
assert audio.get_sample_rate() == 88200
assert audio.channels == 1
assert audio.file_name == file_name
def test_load_folder_resample(self, shared_datadir):
file_name = (shared_datadir / 'audio/test_folder').resolve()
audio = AudioBuffer.load_folder(file_name, sample_rate=22050)
assert len(audio) == 3
assert audio[0].get_audio().shape == (22050,)
assert audio[0].get_sample_rate() == 22050
assert audio[0].file_name == 'test_1.wav'
assert audio[1].get_audio().shape == (22050,)
assert audio[1].get_sample_rate() == 22050
assert audio[1].file_name == 'test_2.wav'
assert audio[2].get_audio().shape == (22050,)
assert audio[2].get_sample_rate() == 22050
assert audio[2].file_name == 'test_3.wav'
def test_batch_scale_magnitude(self):
bin_freq = 44100. / 1024.
feature_batch = np.zeros((10, 513))
fft = FFT(output='magnitude', scale_axis=None)
for i in range(10):
bin = (i + 1) * 2
sine = utils.make_test_cosine(1024, bin_freq * bin, 44100)
audio = AudioBuffer(sine, 44100)
feature_batch[i] = fft(audio)
scaled = fft.fit_scaler(feature_batch)
expected_mean = (512.0 * 10) / (513.0 * 10)
assert fft.scaler.mean == pytest.approx(expected_mean)
def get_audio(self):
"""
Return monophonic audio from rendered patch
:return: An audio buffer of the rendered patch
:rtype: :class:`spiegelib.core.audio_buffer.AudioBuffer`
"""
if self.rendered_patch:
audio = AudioBuffer(self.engine.get_audio_frames(),
self.sample_rate)
return audio
else:
raise Exception(
'Patch must be rendered before audio can be retrieved')
def test_sine_mfcc_scale(self):
batch_size = 10
mfcc = MFCC(hop_size=512, frame_size=1024, scale_axis=(0, 2))
features = np.zeros((batch_size, 20, 5))
for i in range(batch_size):
sine = utils.make_test_sine(2048, 100 + (50 * i), 44100)
audio = AudioBuffer(sine, 44100)
features[i] = mfcc(audio)
assert features.shape == (10, 20, 5)
scaled = mfcc.fit_scaler(features)
assert scaled.mean() == pytest.approx(0.)
assert scaled.std() == pytest.approx(1.)
np.testing.assert_array_almost_equal(scaled.mean((0, 2)), np.zeros(20))
np.testing.assert_array_almost_equal(scaled.std((0, 2)), np.ones(20))
def test_sine_wave_stft_power(self):
bin_freq = 44100. / 1024.
sine = utils.make_test_cosine(1024 + 1, bin_freq * 10, 44100)
audio = AudioBuffer(sine, 44100)
stft = STFT(output='power', fft_size=1024)
features = stft(audio)
expected = np.zeros((513, 3))
# amplitude is spread out over neighbouring bins
expected[9, :] = 128. * 128.
expected[10, :] = 256. * 256.
expected[11, :] = 128. * 128.
assert features.shape == (513, 3)
assert isinstance(features[0][0], np.float32)
np.testing.assert_array_almost_equal(features, expected)
def test_sine_wave_fft_power_phase(self):
bin_freq = 44100. / 1024.
# Using a cosine here to get zero initial phase
sine = utils.make_test_cosine(1024, bin_freq * 10, 44100)
audio = AudioBuffer(sine, 44100)
fft = FFT(output='power_phase')
features = fft(audio)
expected = np.zeros((513, 2))
expected[10][0] = 512. * 512.
assert features.shape == (513, 2)
assert isinstance(features[0][0], np.float32)
assert isinstance(features[0][1], np.float32)
# Assert correct magnitude
np.testing.assert_array_almost_equal(features[:, 0], expected[:, 0])
# Uncertain about phase, except for 10th bin
assert features[10][1] == pytest.approx(0.)
def test_sine_wave_stft_magnitude_scaled(self):
bin_freq = 44100. / 1024.
length = (1024. * 10) + 1
stft = STFT(output='magnitude',
fft_size=1024,
hop_size=512,
time_major=True,
scale_axis=None)
batch_size = 10
features = np.zeros((10, 21, 513))
for i in range(batch_size):
sine = utils.make_test_cosine(int(length), bin_freq * (i + 10),
44100)
audio = AudioBuffer(sine, 44100)
features[i] = stft(audio)
scaled = stft.fit_scaler(features)
assert scaled.shape == (10, 21, 513)
assert scaled.mean() == pytest.approx(0.)
assert scaled.std() == pytest.approx(1.)
def test_sine_extraction(self):
frame_size = 1024
hop_size = 512
bin_freq = 44100. / float(frame_size)
sine = utils.make_test_cosine(1024 + 1, bin_freq * 10, 44100)
audio = AudioBuffer(sine, 44100)
spectral = SpectralSummarized(frame_size=1024, hop_size=512)
features = spectral(audio)
assert features.shape == (22, )
assert features[0] == pytest.approx(bin_freq * 10)
assert features[1] == pytest.approx(0.)
assert features[2] == pytest.approx(30.452547928239625)
assert features[3] == pytest.approx(0.)
assert features[4] == pytest.approx(0.)
assert features[5] == pytest.approx(0.)
assert features[6] == pytest.approx(bin_freq * 11)
assert features[7] == pytest.approx(0.)
# Spectral flatness subbands
assert features[8] == pytest.approx(44.0823996531185)
assert features[9] == pytest.approx(0.)
assert features[10] == pytest.approx(44.0823996531185)
assert features[11] == pytest.approx(0.)
assert features[12] == pytest.approx(124.0823996531185)
assert features[13] == pytest.approx(0.)
assert features[14] == pytest.approx(44.0823996531185)
assert features[15] == pytest.approx(0.)
assert features[16] == pytest.approx(44.0823996531185)
assert features[17] == pytest.approx(0.)
assert features[18] == pytest.approx(44.0823996531185)
assert features[19] == pytest.approx(0.)
assert features[20] == pytest.approx(44.0823996531185)
assert features[21] == pytest.approx(0.)
def test_peak_normalize(self):
samples = self.make_test_sine(44100, 100)
samplesHalfAmp = samples * 0.5
samplesNorm = AudioBuffer.peak_normalize(samples)
np.testing.assert_array_equal(samples, samplesNorm)
def test_empty_construction(self):
audio = AudioBuffer()
assert audio.get_audio() == None
assert audio.get_sample_rate() == None
assert audio.channels == 0
assert audio.file_name == ''
def test_save_mono_normalize(self, tmp_path):
samples = self.make_test_sine(44100, 400, 44100)
samplesHalfAmp = samples * 0.5
audio = AudioBuffer(samples, 44100)
audioNormed = AudioBuffer(samples, 44100)
file_name = (tmp_path / 'test_save.wav').resolve()
audio.save(file_name, normalize=True)
audioReload = AudioBuffer(file_name)
# Make sure it looks correct
np.testing.assert_array_almost_equal(audioNormed.get_audio(), audioReload.get_audio(), decimal=5)
assert audioReload.get_audio().shape == (44100,)
assert audioReload.get_sample_rate() == 44100
assert audioReload.channels == 1
assert audioReload.file_name == file_name
def test_array_construction_sample_rate_exception(self):
samples = np.array([1,2,3,4,5,6,7,8,9,10])
with pytest.raises(Exception) as exc_info:
audio = AudioBuffer(samples)
assert exc_info.type is Exception
assert exc_info.value.args[0] == 'Sample rate is required when initializing with audio data'
def test_peak_normalize_no_change(self):
samples = self.make_test_sine(44100, 100)
samplesNorm = AudioBuffer.peak_normalize(samples)
np.testing.assert_array_almost_equal(samples, samplesNorm, decimal=5)
def test_load_folder_exception(self, shared_datadir):
file_name = (shared_datadir / 'audio/non_existent').resolve()
with pytest.raises(Exception) as exc_info:
audio = AudioBuffer.load_folder(file_name)
assert exc_info.type is ValueError
assert exc_info.value.args[0] == '%s is not a directory' % file_name
