def test_get_raw(self):
vector = np.arange(1, 7)
audio = FramedAudio(vector, 44100, 4, 2)
raw = audio.get_raw()
self.assertTrue(np.array_equal(raw, vector))
def test_get_frame(self):
vector = np.arange(1, 7)
audio = FramedAudio(vector, 44100, 4, 2)
frame = audio.get_frame(0)
expected = np.array([1,2,3,4])
self.assertTrue(np.array_equal(frame, expected))
frame = audio.get_frame(1)
expected = np.array([3,4,5,6])
self.assertTrue(np.array_equal(frame, expected))
def test_from_file(self):
test_path = os.path.dirname(__file__)
src = os.path.join(test_path, 'audio_sample.ogg')
audio = FramedAudio.from_file(src, 2048, 1024, True)
audio2 = FramedAudio.from_file(src, config={'block-size':2048, 'hop-size':1024, 'centered':True})
self.assertEqual(audio.get_config(), audio2.get_config())
array, _ = audio_io.read(src)
self.assertTrue(np.array_equal(array, audio.array))
def test_matches_cfg(self):
vector = np.arange(128)
audio = FramedAudio(vector, 44100, 4, 2)
self.assertTrue(audio.matches_cfg({'fs':44100}))
self.assertFalse(audio.matches_cfg({'fs':48100}))
self.assertFalse(audio.matches_cfg({'no-key':44100}))
self.assertTrue(audio.matches_cfg(audio.get_config()))
def test_pitch_acf(self):
block_size = 4096
fs = 44100
f0Sweep = np.linspace(200, 800, 8 * block_size)
phase = np.add.accumulate(2. * np.pi * f0Sweep / fs)
x = np.sin(phase)
audio = FramedAudio(x, fs, block_size, block_size)
pitch = pitch_acf(audio, fs)
pitch_reference = f0Sweep[0:-1:block_size]
self.assertTrue(np.max(np.abs(pitch - pitch_reference)) < 50)
def test_set_get_config(self):
vector = np.arange(128)
audio = FramedAudio(vector, 44100, 4, 2)
cfg = audio.get_config()
self.assertEqual(cfg['fs'], 44100)
self.assertEqual(cfg['block-size'], 4)
self.assertEqual(cfg['hop-size'], 2)
vector = np.arange(128)
audio2 = FramedAudio(vector, 123, 2, 4)
audio2.set_config(cfg)
self.assertEqual(audio2.get_config(), cfg)
def test_trajectories(self):
vector = np.arange(128)
audio = FramedAudio(vector, 44100, 4, 2)
traj = np.arange(audio.get_num_frames())
audio.store_trajectory('traj', traj)
traj2 = audio.get_trajectory('traj')
self.assertTrue(np.array_equal(traj, traj2))
self.assertTrue( audio.get_trajectory('no-key') is None)
def test_get_time(self):
audio = FramedAudio(np.arange(512), 256, 256,256)
expected = np.array([0., 1.])
result = audio.get_time(False)
self.assertTrue(np.array_equal(expected, result))
expected = np.array([0.5, 1.5])
result = audio.get_time(True)
self.assertTrue(np.array_equal(expected, result))
audio = FramedAudio(np.arange(512), 256, 256, 256, centered=True)
expected = np.array([-0.5, 0.5])
result = audio.get_time(False)
self.assertTrue(np.array_equal(expected, result))
expected = np.array([0., 1.])
result = audio.get_time(True)
self.assertTrue(np.array_equal(expected, result))
# allowing growth on GPU
config = tf.compat.v1.ConfigProto()
config.gpu_options.allow_growth = True
session = tf.compat.v1.Session(config=config)
#%% Import Dataset
print('Import Dataset')
# get cached analysis result and source sample
num_overtones = 40
wav_file = dataset.get_sample('scales', 'slow_forte', 'a', 'f6')[0]
json_file = os.path.splitext(wav_file)[0] + '.json'
# import framed audio
audio = FramedAudio.from_json(json_file)
sample_rate = audio.fs
# run onset / offset detection
onset, offset = magnitude.get_onset_offset(audio)
onset += 100
offset -= 100
num_samples = offset - onset
#run overtone extraction
overtones = extract_overtones.extract_overtones_from_audio(audio, num_overtones = num_overtones)
overtonesDB = 20. * np.log10(np.abs(overtones[onset:offset, :]))
overtonesDB -= np.max(overtonesDB)
#extract pitch information
pitch = np.reshape(audio.get_trajectory('pitch')[onset:offset], [num_samples,1])
def test_get_num_frames(self):
audio = FramedAudio(np.zeros((1,128)), 44100, 256, 128)
self.assertEqual(audio.get_num_frames(), 1)
audio = FramedAudio(np.zeros((1,256)), 44100, 256, 128)
self.assertEqual(audio.get_num_frames(), 1)
audio = FramedAudio(np.zeros((1,384)), 44100, 256, 128)
self.assertEqual(audio.get_num_frames(), 2)
audio = FramedAudio(np.zeros((1,385)), 44100, 256, 128)
self.assertEqual(audio.get_num_frames(), 2)
audio = FramedAudio(np.zeros((0,0)),44100, 256, 128)
self.assertEqual(audio.get_num_frames(), 1)
def test_centered(self):
audio = FramedAudio(np.arange(512), 44100, 256,128, centered=True)
# frames expected as follows
# 1: [-128, 128[
# 2: [0, 256[
# 3: [128, 384[
# 4: [256, 512[
self.assertEqual(audio.get_num_frames(), 4)
self.assertEqual(audio.get_frame(0).size, 256)
self.assertEqual(audio.get_frame(1).size, 256)
self.assertEqual(audio.get_frame(2).size, 256)
self.assertEqual(audio.get_frame(3).size, 256)
self.assertEqual(audio.get_frame(0)[-1], 127)
self.assertEqual(audio.get_frame(1)[-1], 255)
self.assertEqual(audio.get_frame(2)[-1], 383)
self.assertEqual(audio.get_frame(3)[-1], 511)
audio = FramedAudio(np.arange(512), 44100, 256, 128, centered=False)
# frames expected as follows
# 1: [-128, 128[
# 2: [0, 256[
# 3: [128, 384[
# 4: [256, 512[
self.assertEqual(audio.get_num_frames(), 3)
self.assertEqual(audio.get_frame(0).size, 256)
self.assertEqual(audio.get_frame(1).size, 256)
self.assertEqual(audio.get_frame(2).size, 256)
self.assertEqual(audio.get_frame(0)[-1], 255)
self.assertEqual(audio.get_frame(1)[-1], 383)
self.assertEqual(audio.get_frame(2)[-1], 511)
audio = FramedAudio(np.arange(513), 44100, 256, 128, centered=True)
self.assertEqual(audio.get_num_frames(), 4)
def test_json(self):
vector = np.arange(128)
audio = FramedAudio(vector, 44100, 4, 2)
traj = np.arange(audio.get_num_frames())
audio.store_trajectory('traj', traj)
audio.save_json('temp.json')
# create another instance with same vector but different
audio2 = FramedAudio.from_json('temp.json')
self.assertTrue( np.array_equal(audio.get_trajectory('traj'), audio2.get_trajectory('traj')))
self.assertEqual( audio.get_config(), audio2.get_config())
test_path = os.path.dirname(__file__)
src = os.path.join(test_path, 'audio_sample.ogg')
audio = FramedAudio.from_file(src, 2048, 1024)
audio.save_json('temp.json')
audio2 = FramedAudio.from_json('temp.json')
self.assertEqual( audio.get_config(), audio2.get_config())
self.assertEqual( audio.src_file, audio2.src_file)
self.assertTrue( np.array_equal(audio.array, audio2.array))
audio3 = FramedAudio.from_json('non-existing.json')
self.assertTrue(audio3 is None)