def read_data(self, DataListName, FeatureName,shape):
'''
Load original data and the feature that you choose to use is needed
You can choose different output shape.
:DataListName: The log.txt path
:FeatureName: The name of the feature that you choose to use
:shape: The output shape that you prefer
'''
Feature = FeatureName.capitalize()
Data = np.zeros((1,shape))
Label = []
processer = PreProcessing(512, 128)
wav_list, frame_list, energy_list, zcr_list, endpoint_list, Label = processer.process(DataListName)
if Feature[0] == 'E':
for i in range(len(zcr_list)):
temp = processer.effective_feature(energy_list[i], endpoint_list[i])
temp = processer.reshape(temp, shape)
Data=np.concatenate((Data,temp),axis = 0)
Data = Data[1:]
return Data, Label
elif Feature[0] == 'Z':
for i in range(len(zcr_list)):
temp = processer.effective_feature(zcr_list[i], endpoint_list[i])
temp = processer.reshape(temp, shape)
Data=np.concatenate((Data,temp),axis = 0)
Data = Data[1:]
return Data, Label
else:
print("please choose correct feature, and we will return ZCR by default")
for i in range(len(zcr_list)):
temp = processer.effective_feature(zcr_list[i], endpoint_list[i])
temp = processer.reshape(temp, shape)
Data=np.concatenate((Data,temp),axis = 0)
Data = Data[1:]
return Data, Label
def setUp(self):
self.DATA_DIR = './TEST_DIR'
self.LOG = './TEST_DIR/data_log.txt'
builder = VoiceDataSetBuilder(
dst_path=self.DATA_DIR,
log_file=self.LOG,
rate=44100)
builder.build()
self.pre_process = PreProcessing(frame_size=512, overlap=128)
plt.ion()
plt.figure(1)
def load_target(self, ModelListName):
'''
Load model data
this is the model data for classification
:ModelListName: The Model_log.txt path
'''
eff_label_list = []
eff_mfcc = []
processer = PreProcessing(512, 128)
wav_list, frame_list, mfcc_list, energy_list, zcr_list, endpoint_list, label_list = processer.process(
ModelListName)
for i in range(len(mfcc_list)):
temp = processer.effective_feature(mfcc_list[i], endpoint_list[i])
if endpoint_list[i][1] - endpoint_list[i][0] != 0:
eff_label_list.append(label_list[i])
eff_mfcc.append(mfcc_list[i])
else:
continue
return eff_mfcc, eff_label_list
def read_data(self, DataListName):
'''
Load original data
You can choose different output shape.
:DataListName: The log.txt path
'''
eff_label_list = []
eff_mfcc = []
processer = PreProcessing(512, 128)
wav_list, frame_list, mfcc_list, energy_list, zcr_list, endpoint_list, label_list = processer.process(
DataListName)
for i in range(len(mfcc_list)):
temp = processer.effective_feature(mfcc_list[i], endpoint_list[i])
if endpoint_list[i][1] - endpoint_list[i][0] != 0:
eff_label_list.append(label_list[i])
eff_mfcc.append(mfcc_list[i])
else:
continue
return eff_mfcc, eff_label_list
class PreProcessingTests(unittest.TestCase):
def setUp(self):
self.DATA_DIR = './TEST_DIR'
self.LOG = './TEST_DIR/data_log.txt'
builder = VoiceDataSetBuilder(
dst_path=self.DATA_DIR,
log_file=self.LOG,
rate=44100)
builder.build()
self.pre_process = PreProcessing(frame_size=512, overlap=128)
plt.ion()
plt.figure(1)
def test_all(self):
wav_list, frame_list, energy_list, zcr_list, endpoint_list = \
self.pre_process.process(self.LOG)
for i in range(len(frame_list)):
wav_data = wav_list[i]
print(np.max(np.abs(wav_data)))
frames = frame_list[i]
energys = energy_list[i]
zcrs = zcr_list[i]
endpoints = endpoint_list[i]
effective_es = PreProcessing.effective_feature(zcrs, endpoints)
plt.figure(i + 1)
plt.subplot(221)
plt.plot(wav_data)
print(endpoints)
for ep in endpoints:
plt.axvline(ep * (self.pre_process.frame_size - self.pre_process.overlap), color='r')
plt.subplot(222)
plt.plot(energys)
for ep in endpoints:
plt.axvline(ep, color='r')
plt.subplot(223)
longest_e = []
for e in effective_es:
if len(e) > len(longest_e):
longest_e = e
plt.plot(longest_e)
plt.subplot(224)
plt.plot(zcrs)
for ep in endpoints:
plt.axvline(ep, color='r')
plt.show()
plt.waitforbuttonpress()
def tearDown(self):
for file in os.listdir(self.DATA_DIR):
os.remove(os.path.join(self.DATA_DIR, file))
if os.path.exists(self.DATA_DIR):
os.removedirs(self.DATA_DIR)
if os.path.exists(self.LOG):
os.remove(self.LOG)
class FeatureExtractorsTests(unittest.TestCase):
def setUp(self):
self.WAVE_NUM = 10
self.WAVE_DURANCE = 1000
self.wave_list = np.random.randn(self.WAVE_DURANCE, self.WAVE_NUM)
self.base = PreProcessing(100, 10)
self.frames = np.array([self.base.Enframe(wave) for wave in self.wave_list])
def test_enhance_frame(self):
for i in range(self.WAVE_NUM):
self.assertTrue((
FeatureExtractors.enhance_frame(
wav_data=self.wave_list[:, i],
frame_size=100,
overlap=10,
windowing_method='Hamming'
) ==
self.base.Enframe(self.wave_list[:, i])
).all())
def test_zero_crossing_rate(self):
for i in range(self.WAVE_NUM):
self.assertTrue((
FeatureExtractors.zero_crossing_rate(
frames=self.frames
) ==
self.base.ZCR(self.frames)
).any())
def test_energy(self):
for i in range(self.WAVE_NUM):
self.assertTrue((
FeatureExtractors.energy(self.frames) ==
self.base.energy(self.frames)
).any())
def call_extraction(args):
if len(args) == 3:
input_file = args[0]
template_name = args[1]
output_file = args[2]
else:
raise IncorrectArgumentNumberException(3, args)
xml_tree = ET.ElementTree()
with open(constants.base_filepath + input_file) as file:
data = file.read()
try:
xml_tree = ET.fromstring(data)
except (ET.XMLSyntaxError, ValueError) as e:
data = re.sub('<\\?.*?\\?>', '', data)
xml_tree = ET.fromstring(data)
template = XMLUtil.Template(template_name)
pre_processed_xml_tree = PreProcessing.apply_all(
xml_tree, template.pre_process_queue)
extracted_xml = XMLExtractor.extract_template_data_from_xml(
template.get_template(), pre_processed_xml_tree)
extracted_xml = PostProcessing.apply_all(extracted_xml,
template.post_process_queue)
out = XMLUtil.xml_to_string(extracted_xml)
if output_file != 'None':
with open(constants.base_filepath + output_file, 'w') as file:
file.write(out)
return out
def read_data(self, DataListName, FeatureName, shape):
'''
Load original data and the feature that you choose to use is needed
You can choose different output shape.
:DataListName: The log.txt path
:FeatureName: The name of the feature that you choose to use
:shape: The output shape that you prefer
'''
Feature = FeatureName.capitalize()
Data = np.zeros((1, shape))
zcrdata = np.zeros((1, shape))
energydata = np.zeros((1, shape))
eff_label_list = []
processer = PreProcessing(512, 128)
wav_list, frame_list, mfcc_list, energy_list, zcr_list, endpoint_list, label_list = processer.process(
DataListName)
if Feature[0] == 'E':
for i in range(len(energy_list)):
temp = processer.effective_feature(energy_list[i],
endpoint_list[i])
temp = processer.reshape(temp, shape)
if len(temp) != 0:
eff_label_list.append(label_list[i])
else:
continue
Data = np.concatenate((Data, temp), axis=0)
Data = Data[1:]
return Data, eff_label_list
elif Feature[0] == 'Z':
for i in range(len(zcr_list)):
temp = processer.effective_feature(zcr_list[i],
endpoint_list[i])
temp = processer.reshape(temp, shape)
print(np.shape(temp))
if len(temp) != 0:
eff_label_list.append(label_list[i])
else:
continue
Data = np.concatenate((Data, temp), axis=0)
Data = Data[1:]
return Data, eff_label_list
elif Feature[0] == 'W':
for i in range(len(zcr_list)):
temp = processer.effective_feature(zcr_list[i],
endpoint_list[i])
temp = processer.reshape(temp, shape)
if len(temp) != 0:
eff_label_list.append(label_list[i])
else:
continue
zcrdata = np.concatenate((zcrdata, temp), axis=0)
zcrdata = zcrdata[1:]
for i in range(len(zcr_list)):
temp = processer.effective_feature(energy_list[i],
endpoint_list[i])
temp = processer.reshape(temp, shape)
if len(temp) == 0:
continue
energydata = np.concatenate((energydata, temp), axis=0)
energydata = energydata[1:]
data = energydata * zcrdata
return data, eff_label_list
else:
print(
"please choose correct feature, and we will return ZCR by default"
)
for i in range(len(zcr_list)):
temp = processer.effective_feature(zcr_list[i],
endpoint_list[i])
temp = processer.reshape(temp, shape)
if len(temp) != 0:
eff_label_list.append(label_list[i])
else:
continue
Data = np.concatenate((Data, temp), axis=0)
Data = Data[1:]
return Data, eff_label_list
def setUp(self):
self.WAVE_NUM = 10
self.WAVE_DURANCE = 1000
self.wave_list = np.random.randn(self.WAVE_DURANCE, self.WAVE_NUM)
self.base = PreProcessing(100, 10)
self.frames = np.array([self.base.Enframe(wave) for wave in self.wave_list])
