def test_load_labels_from_txt(self):
filename_data, filename_labels = get_labeled_txt()
df = AudioDataFile().load(filename_data, formatter=AIFFormatter())
df.load_labels(filename_labels,
labels_formatter=TXTLabelsFormatter(),
label="whale")
assert [0, 1] in df.get_labeled_data().labels.unique()
assert "whale" in df.name_label
def test_concatenate_data_files(self):
fns = get_5_file_names()
dfs = []
for d in fns:
dfs.append(AudioDataFile().load(d, formatter=AIFFormatter()))
big_df = AudioDataFile().concatenate(dfs)
assert big_df.duration.seconds >= sum(
[d.duration.seconds for d in dfs])
def test_copy_dataframe(self):
filename_data, filename_labels = get_labeled_txt()
df = AudioDataFile().load(filename_data, formatter=AIFFormatter())
df.load_labels(filename_labels,
labels_formatter=TXTLabelsFormatter(),
label="whale")
new_df = AudioDataFile(df)
np.testing.assert_equal(new_df.metadata["labels"],
df.metadata["labels"])
def test_add_remove_datafile(self):
filename = get_file_name()
df = AudioDataFile()
df.load(filename,
formatter=AIFFormatter())
ds = OneDataFileOut()
ds.add_data_file(df)
assert len(ds.datafiles) == 1
ds.remove_data_file(df)
assert len(ds.datafiles) == 0
def test_add_windows(self):
filename_data, filename_labels = get_labeled_txt()
df = AudioDataFile().load(filename_data, formatter=AIFFormatter())
df.load_labels(filename_labels,
labels_formatter=TXTLabelsFormatter(),
label="whale")
st1 = df.data.index[0]
en1 = df.data.index[5]
st2 = df.metadata["labels"][0][0]
en2 = st2 + 4
# No windows added yet
assert df.parameters["number_of_windows"] == 0
d = df.get_window(0)
assert all(d == df.data)
# Add windows
# First an incorrect one
with pytest.raises(AttributeError):
df.add_window(st1 + 100, en1 - 1)
df.add_window(st1, en1) # Length 6
df.add_window(st2, en2) # Length 4
assert df.parameters["number_of_windows"] == 2
assert str(df) == "AudioDataFile (2 windows)"
def test_sliding_windows():
filename = get_file_name()
p = SlidingWindows()
p.parameters["sliding_window_width"] = "13s"
p.parameters["overlap"] = 0.12
df = AudioDataFile()
df.load(filename, formatter=AIFFormatter())
p.parameters["data"] = df
new_df = p.transform()
new_df.get_windows_data_frame()
assert p.description != ""
def test_load(self):
filename = get_file_name()
df = AudioDataFile()
# No attribute data
assert str(df) == "AudioDataFile"
with pytest.raises(RuntimeError):
df.duration
with pytest.raises(RuntimeError):
df.start_time
with pytest.raises(RuntimeError):
df.end_time
df.load(filename, formatter=AIFFormatter())
assert df.duration.seconds > 0
assert str(df) == "AudioDataFile (0 days 00:14:59.999500)"
def get_validation(self):
if not self.parameters["validation"]:
return None
for i in list(range(len(self.datafiles) - 1)) + [-1]:
curr_datafiles = self.datafiles.copy()
validation = curr_datafiles.pop(i + 1)
yield AudioDataFile().__class__().concatenate([validation])
def get_testing(self):
if not self.parameters["testing"]:
return None
for i in range(len(self.datafiles)):
curr_datafiles = self.datafiles.copy()
testing = curr_datafiles.pop(i)
yield AudioDataFile().__class__().concatenate([testing])
def test_scale():
file_name = get_file_name()
df = AudioDataFile().load(file_name, formatter=AIFFormatter())
p = Scale()
p.parameters["data"] = df
result = p.transform()
assert p.description != ""
assert type(result) is df.__class__
def test_save(self):
filename = get_file_name()
df = AudioDataFile()
df.load(filename, formatter=AIFFormatter())
df.save("tmp.h5", formatter=HDF5Formatter())
df2 = AudioDataFile().load("tmp.h5", formatter=HDF5Formatter())
assert df2.duration.seconds > 0
def test_identity():
file_name = get_file_name()
df = AudioDataFile().load(file_name, formatter=AIFFormatter())
f = Identity()
f.parameters["data"] = df
t = f.transform()
np.testing.assert_allclose(df.data.values.ravel(), t.data.values.ravel())
assert f.description != ""
assert t.data.values.ndim == 2
def get_training(self):
if not self.parameters["training"]:
return None
for i in list(range(len(self.datafiles) - 1)) + [-1]:
curr_datafiles = self.datafiles.copy()
if self.parameters["validation"]:
curr_datafiles.pop(i + 1) # i + 1 is validation
if self.parameters["testing"]:
curr_datafiles.pop(i) # i is testing
yield AudioDataFile().concatenate(curr_datafiles)
def test_skewness():
file_name = get_file_name()
df = AudioDataFile().load(file_name, formatter=AIFFormatter())
df.data -= df.data.mean()
f = Skewness()
f.parameters["data"] = df
t = f.transform()
assert t.data.values.shape[1] == 1
assert f.description != ""
assert t.data.values.ndim == 2
def test_spectral_frames():
file_name = get_file_name()
df = AudioDataFile().load(file_name, formatter=AIFFormatter())
df.data -= df.data.mean()
f = SpectralFrames()
f.parameters["sampling_rate"] = df.sampling_rate
f.parameters["data"] = df
t = f.transform()
assert t.data.values.shape[0] == 1
assert f.description != ""
assert t.data.values.ndim == 2
def test_mfcc():
file_name = get_file_name()
df = AudioDataFile().load(file_name, formatter=AIFFormatter())
df.data -= df.data.mean()
f = MFCC()
f.parameters["sampling_rate"] = df.sampling_rate
f.parameters["n_components"] = 25
f.parameters["data"] = df
t = f.transform()
assert t.data.values.shape[1] == 25
assert f.description != ""
assert t.data.values.ndim == 2
def test_get_training_testing_validation_set(self):
ds = OneDataFileOut()
file_names = get_5_file_names()
for filename in file_names:
ds.add_data_file(AudioDataFile().load(filename,
formatter=AIFFormatter()))
assert len(ds.datafiles) == 5
training = ds.get_training()
testing = ds.get_testing()
validation = ds.get_validation()
expected_iterations = ds.iterations
i = 0
for tr, te, val in zip(training, testing, validation):
i += 1
assert expected_iterations == i
def test_get_training_testing_set(self):
ds = OneDataFileOut()
ds.parameters["validation"] = False # Disable validation set generation
file_names = get_5_file_names()
for filename in file_names:
ds.add_data_file(AudioDataFile().load(filename,
formatter=AIFFormatter()))
assert len(ds.datafiles) == 5
training = ds.get_training()
testing = ds.get_testing()
expected_iterations = ds.iterations
i = 0
for tr, te in zip(training, testing):
i += 1
assert expected_iterations == i
def test_segments_repr(self):
filename_data, filename_labels = get_labeled_txt()
df = AudioDataFile().load(filename_data, formatter=AIFFormatter())
df.load_labels(filename_labels,
labels_formatter=TXTLabelsFormatter(),
label="whale")
# Make 5 windows
st = df.start_time
step = df.duration / 5
out = AudioSegments()
assert str(out) == "AudioSegments"
for i in range(5):
df.add_window(st + i * step, st + (i + 1) * step)
out.add_segment(*df.get_window(i))
assert str(out) == "AudioSegments (5 segments)"
def build_data_file(self, params: dict):
""""""
available_data_files = getters.get_available_data_files()
available_formatters = getters.get_available_formatters()
# Load every small input data file and concatenate all into the big data file
dfs = []
for elem in params["input_data"]:
self.logger.info(f"Loading and appending file {elem['file_name']}")
file_name = elem["file_name"]
data_file_name = elem["data_file"]
formatter_name = elem["formatter"]
df = available_data_files[data_file_name]()
fmt = available_formatters[formatter_name]()
df.load(file_name=file_name, formatter=fmt)
dfs.append(df)
big_df = AudioDataFile().concatenate(dfs)
return {"input_data": big_df}
def load_feature_datafile(self):
filename_data, filename_labels = get_labeled_txt()
df = AudioDataFile().load(filename_data, formatter=AIFFormatter())
df.load_labels(filename_labels,
labels_formatter=TXTLabelsFormatter(),
label="whale")
# Make 5 windows
st = df.start_time
step = df.duration / 5
out = AudioSegments()
for i in range(5):
df.add_window(st + i * step, st + (i + 1) * step)
out.add_segment(*df.get_window(i))
df = out
# Get features
f1 = Energy()
f1.parameters["data"] = df
f2 = Kurtosis()
f2.parameters["data"] = df
f1 = f1.transform()
f2 = f2.transform()
return FeatureDataFile().concatenate([f1, f2])
def test_sampling_rate(self):
filename = get_file_name()
df = AudioDataFile()
assert df.sampling_rate is None
df.load(filename, formatter=AIFFormatter())
assert df.sampling_rate > 0
def test_get_window(self):
filename_data, filename_labels = get_labeled_txt()
df = AudioDataFile().load(filename_data, formatter=AIFFormatter())
df.load_labels(filename_labels,
labels_formatter=TXTLabelsFormatter(),
label="whale")
st1 = df.data.index[0]
en1 = df.data.index[5]
st2 = df.metadata["labels"][0][0]
en2 = st2 + 4
df.add_window(st1, en1) # Length 6
df.add_window(st2, en2) # Length 4
wdf = df.get_windows_data_frame()
assert len(wdf) == 2
w1, l1 = df.get_window(0)
w2, l2 = df.get_window(1)
assert len(w1) == 6
assert len(w2) == 4
# Out of range index
assert df.get_window(100) is None
# Other label treatments
df.parameters["labels_treatment"] = "mode"
df.get_window(0)
df.parameters["labels_treatment"] = "mean"
_, l1mn = df.get_window(0)
assert type(l1mn) is float
with pytest.raises(ValueError):
df.parameters["labels_treatment"] = "fourier"
# There is actually no labels treatment called fourier
df.get_window(0) # Raises value error