def test_learn():
FeatureExtractor(store=True, overwrite=True).extract(
audio_file=os.path.join('material', 'test.wav'),
extractor_name='mfcc',
extractor_params={'mfcc': {
'n_mfcc': 10
}},
storage_paths={'mfcc': os.path.join('material', 'test.mfcc.cpickle')})
feature_container = FeatureContainer(
filename=os.path.join('material', 'test.mfcc.cpickle'))
data = {
'file1.wav': feature_container,
'file2.wav': feature_container,
}
annotations = {
'file1.wav':
MetaDataContainer([{
'file': 'file1.wav',
'scene_label': 'scene1',
'event_onset': 0.0,
'event_offset': 1.0,
'event_label': 'event1',
'identifier': 'a',
}, {
'file': 'file1.wav',
'scene_label': 'scene1',
'event_onset': 1.0,
'event_offset': 2.0,
'event_label': 'event2',
'identifier': 'a',
}, {
'file': 'file1.wav',
'scene_label': 'scene1',
'event_onset': 2.0,
'event_offset': 3.0,
'event_label': 'event2',
'identifier': 'a',
}, {
'file': 'file1.wav',
'scene_label': 'scene1',
'event_onset': 4.0,
'event_offset': 5.0,
'event_label': 'event1',
'identifier': 'a',
}, {
'file': 'file1.wav',
'scene_label': 'scene1',
'event_onset': 1.0,
'event_offset': 2.0,
'event_label': 'event1',
'identifier': 'a',
}]),
'file2.wav':
MetaDataContainer([{
'file': 'file2.wav',
'scene_label': 'scene1',
'event_onset': 0.0,
'event_offset': 1.0,
'event_label': 'event2',
'identifier': 'b',
}, {
'file': 'file2.wav',
'scene_label': 'scene1',
'event_onset': 1.0,
'event_offset': 2.0,
'event_label': 'event1',
'identifier': 'b',
}, {
'file': 'file2.wav',
'scene_label': 'scene1',
'event_onset': 2.0,
'event_offset': 3.0,
'event_label': 'event2',
'identifier': 'b',
}, {
'file': 'file2.wav',
'scene_label': 'scene1',
'event_onset': 3.0,
'event_offset': 4.0,
'event_label': 'event2',
'identifier': 'b',
}])
}
ed = EventDetectorGMM(
method='gmm',
class_labels=['event1', 'event2'],
params={
'hop_length_seconds': 0.02,
'parameters': {
'n_components': 6,
'covariance_type': 'diag',
'tol': 0.001,
'reg_covar': 0,
'max_iter': 40,
'n_init': 1,
'init_params': 'kmeans',
'random_state': 0,
}
},
filename=os.path.join('material', 'test.model.cpickle'),
disable_progress_bar=True,
)
ed.learn(data=data, annotations=annotations)
# Test model count
nose.tools.eq_(len(ed.model), 2)
# Test model dimensions
nose.tools.eq_(ed.model['event1']['positive'].means_.shape[0], 6)
def test_formats():
delimiters = [',', ';', '\t']
for delimiter in delimiters:
tmp = tempfile.NamedTemporaryFile('r+',
suffix='.txt',
dir='/tmp',
delete=False)
try:
tmp.write('0.5' + delimiter + '0.7\n')
tmp.write('2.5' + delimiter + '2.7\n')
tmp.close()
item = MetaDataContainer().load(filename=tmp.name)[0]
nose.tools.eq_(item.onset, 0.5)
nose.tools.eq_(item.offset, 0.7)
finally:
os.unlink(tmp.name)
tmp = tempfile.NamedTemporaryFile('r+',
suffix='.txt',
dir='/tmp',
delete=False)
try:
tmp.write('0.5' + delimiter + '0.7' + delimiter + 'event\n')
tmp.write('2.5' + delimiter + '2.7' + delimiter + 'event\n')
tmp.close()
item = MetaDataContainer().load(filename=tmp.name)[0]
nose.tools.eq_(item.onset, 0.5)
nose.tools.eq_(item.offset, 0.7)
nose.tools.eq_(item.event_label, 'event')
finally:
os.unlink(tmp.name)
tmp = tempfile.NamedTemporaryFile('r+',
suffix='.txt',
dir='/tmp',
delete=False)
try:
tmp.write('file.wav' + delimiter + 'scene' + delimiter + '0.5' +
delimiter + '0.7' + delimiter + 'event\n')
tmp.write('file.wav' + delimiter + 'scene' + delimiter + '0.5' +
delimiter + '0.7' + delimiter + 'event\n')
tmp.close()
item = MetaDataContainer().load(filename=tmp.name)[0]
nose.tools.eq_(item.onset, 0.5)
nose.tools.eq_(item.offset, 0.7)
nose.tools.eq_(item.event_label, 'event')
nose.tools.eq_(item.file, 'file.wav')
nose.tools.eq_(item.scene_label, 'scene')
finally:
os.unlink(tmp.name)
tmp = tempfile.NamedTemporaryFile('r+',
suffix='.txt',
dir='/tmp',
delete=False)
try:
tmp.write('file.wav' + delimiter + 'scene' + delimiter + '0.5' +
delimiter + '0.7' + delimiter + 'event' + delimiter +
'm' + delimiter + 'a1\n')
tmp.write('file.wav' + delimiter + 'scene' + delimiter + '0.5' +
delimiter + '0.7' + delimiter + 'event' + delimiter +
'm' + delimiter + 'a2\n')
tmp.close()
item = MetaDataContainer().load(filename=tmp.name)[0]
nose.tools.eq_(item.onset, 0.5)
nose.tools.eq_(item.offset, 0.7)
nose.tools.eq_(item.event_label, 'event')
nose.tools.eq_(item.file, 'file.wav')
nose.tools.eq_(item.scene_label, 'scene')
nose.tools.eq_(item.identifier, 'a1')
nose.tools.eq_(item.source_label, 'm')
finally:
os.unlink(tmp.name)
def test_predict():
FeatureExtractor(store=True, overwrite=True).extract(
audio_file=os.path.join('material', 'test.wav'),
extractor_name='mfcc',
extractor_params={'mfcc': {
'n_mfcc': 10
}},
storage_paths={'mfcc': os.path.join('material', 'test.mfcc.cpickle')})
feature_container = FeatureContainer(
filename=os.path.join('material', 'test.mfcc.cpickle'))
data = {
'file1.wav': feature_container,
'file2.wav': feature_container,
}
annotations = {
'file1.wav':
MetaDataContainer([{
'file': 'file1.wav',
'scene_label': 'scene1',
'event_onset': 0.0,
'event_offset': 1.0,
'event_label': 'event1',
'identifier': 'a',
}, {
'file': 'file1.wav',
'scene_label': 'scene1',
'event_onset': 1.0,
'event_offset': 2.0,
'event_label': 'event2',
'identifier': 'a',
}, {
'file': 'file1.wav',
'scene_label': 'scene1',
'event_onset': 2.0,
'event_offset': 3.0,
'event_label': 'event2',
'identifier': 'a',
}, {
'file': 'file1.wav',
'scene_label': 'scene1',
'event_onset': 4.0,
'event_offset': 5.0,
'event_label': 'event1',
'identifier': 'a',
}, {
'file': 'file1.wav',
'scene_label': 'scene1',
'event_onset': 1.0,
'event_offset': 2.0,
'event_label': 'event1',
'identifier': 'a',
}]),
'file2.wav':
MetaDataContainer([{
'file': 'file2.wav',
'scene_label': 'scene1',
'event_onset': 0.0,
'event_offset': 1.0,
'event_label': 'event2',
'identifier': 'b',
}, {
'file': 'file2.wav',
'scene_label': 'scene1',
'event_onset': 1.0,
'event_offset': 2.0,
'event_label': 'event1',
'identifier': 'b',
}, {
'file': 'file2.wav',
'scene_label': 'scene1',
'event_onset': 2.0,
'event_offset': 3.0,
'event_label': 'event2',
'identifier': 'b',
}, {
'file': 'file2.wav',
'scene_label': 'scene1',
'event_onset': 3.0,
'event_offset': 4.0,
'event_label': 'event2',
'identifier': 'b',
}])
}
ed = EventDetectorGMM(
method='gmm',
class_labels=['event1', 'event2'],
params={
'hop_length_seconds': 0.02,
'parameters': {
'n_components': 6,
'covariance_type': 'diag',
'tol': 0.001,
'reg_covar': 0,
'max_iter': 40,
'n_init': 1,
'init_params': 'kmeans',
'random_state': 0,
}
},
filename=os.path.join('material', 'test.model.cpickle'),
disable_progress_bar=True,
)
ed.learn(data=data, annotations=annotations)
recognizer_params = {
'frame_accumulation': {
'enable': False,
'type': 'sliding_sum',
'window_length_frames': 2,
},
'frame_binarization': {
'enable': True,
'type': 'global_threshold',
'threshold': 10,
}
}
# Frame probabilities
frame_probabilities = ed.predict(feature_data=feature_container)
# Event recognizer
result = EventRecognizer(
hop_length_seconds=0.02,
params=recognizer_params,
class_labels=['event1', 'event2'],
).process(frame_probabilities=frame_probabilities)
# Test result
nose.tools.eq_(len(result) > 5, True)
def test_unique_scene_labels():
scenes = MetaDataContainer(content).unique_scene_labels
nose.tools.eq_(len(scenes), 2)
nose.tools.eq_(scenes[0], 'meeting')
nose.tools.eq_(scenes[1], 'office')
def test_max_event_offset():
nose.tools.eq_(MetaDataContainer(content).max_offset, 10)
#embed()
def test_unique_event_labels():
events = MetaDataContainer(content).unique_event_labels
nose.tools.eq_(len(events), 3)
nose.tools.eq_(events[0], 'mouse clicking')
nose.tools.eq_(events[1], 'printer')
nose.tools.eq_(events[2], 'speech')
def test_event_label_count():
nose.tools.eq_(MetaDataContainer(content).event_label_count, 3)
def test_scene_label_count():
nose.tools.eq_(MetaDataContainer(content).scene_label_count, 2)
def test_event_count():
nose.tools.eq_(MetaDataContainer(content).event_count, len(content))
def test_file_list():
files = MetaDataContainer(content).file_list
nose.tools.eq_(len(files), 2)
nose.tools.eq_(files[0], 'audio_001.wav')
nose.tools.eq_(files[1], 'audio_002.wav')