def is_in_dict(x, dic, tol=0.01, by='max_err'):
"""
用来判断一个向量是否包含于一个字典所描述的子空间中
Example:
>>> W1 = np.mat([[0.5, 0, 0], [0.5, 1, 0], [0, 0, 0], [0, 0, 1]])
>>> W2 = np.mat([[1, 0, 0], [0, 1, 0], [0, 0, 1], [0, 0, 0]], dtype=np.double)
>>> v = W1[:, 0]
>>> is_in_dict(v, W2)
True
:param x: 待测试向量
:param dic: 描述子空间字典
:param tol: 可接受的阈值
:param by: 判断方式，max_err,mean_err,act
"""
if x is not np.matrixlib.defmatrix.matrix:
vec = np.mat(x).T
else:
vec = x
if vec.shape[1] != 1:
vec = vec.T
h = decompose_with_dict(vec, dic)
re_ = np.dot(dic, h)
err = np.square(vec - re_).reshape(-1, 1)
if by == 'max_err':
return np.max(err) < tol
elif by == 'mean_err':
return np.mean(err) < tol
elif by == 'act':
return np.any(h <= tol)
else:
raise Exception('get wrong param by')
def __getWight(self):
if os.access('Mixwight.txt', os.F_OK):
wight = loadValue('Mixwight.txt')
else:
wight = decompose_with_dict(np.abs(self.__signalMixSTFT),
self.__dictMixVector)
dumpValue(wight, 'Mixwight.txt')
return wight
def Initialization():
flielist = [
r'dataspeech\track1.wav', r'dataspeech\track10.wav',
r'dataspeech\across.wav'
]
signalA = readSignal(flielist[0])
signalB = readSignal(flielist[1])
mixSignal = readSignal(flielist[2])
IndividVector, W_siga, W_sigb, sigADict, sigBdict = MixVectorAndIndividVector(
signalA, signalB).process()
specSignal = get_spec(mixSignal)
W_mix = decompose_with_dict(np.abs(specSignal), IndividVector)
return IndividVector, W_mix, sigADict, W_siga, sigBdict, W_sigb
def get_child(x, dic, tol=0.001):
"""
获取一个向量所包含的字典中的成分
Example:
>>> W1 = np.mat([[0.5, 0, 0], [0.5, 1, 0], [0, 0, 0], [0, 0, 1]])
>>> W2 = np.mat([[1, 0, 0], [0, 1, 0], [0, 0, 1], [0, 0, 0]], dtype=np.double)
>>> x = W1[:, 0]
>>> get_child(x, W2)
[0, 1]
"""
if x is not np.matrixlib.defmatrix.matrix:
vec = np.mat(x).T
else:
vec = x
if vec.shape[1] != 1:
vec = vec.T
h = decompose_with_dict(vec, dic)
return list(filter(lambda x: h[x] > tol, range(len(h))))
def __getRata(self):
if os.access('CommonRata.txt', os.F_OK):
rate = loadValue('CommonRata.txt')
else:
value = None
m = self.__mixVector.shape[1]
k = self.__dictMixVector.shape[1]
for i in range(0, m):
V = np.transpose(self.__mixVector[:, i]).reshape(-1, 1)
dic = self.__dictMixVector[:, k - m - 2:k]
wigth = decompose_with_dict(V, dic)
if value is None:
value = wigth
else:
value = np.column_stack([value, wigth])
COUNT = 10**(-10)
rate = (np.sum(value[:, 0:self.__sigMixLen], axis=1) +
COUNT) / (np.sum(value, axis=1) + 2 * COUNT)
dumpValue(rate, 'CommonRata.txt')
return rate
def fit_noise(self, noise_spec):
"""
对已知的字典，使用字典分解，来学习公共字典对特征字典的对应关系
Example:
>>> import librosa
>>> from utils import get_spec, norm_signal
>>> noise, _ = librosa.load('data/noise/factory1.wav', sr=SAMPLE_RATE)
>>> speech, _ = librosa.load('data/train/sa1.wav', sr=SAMPLE_RATE)
>>> speech_spec = get_spec(norm_signal(speech))
>>> noise_spec = get_spec(norm_signal(noise))
>>> speech_dic, _ = decompose(np.abs(speech_spec), k=60)
>>> noise_dic, _ = decompose(np.abs(noise_spec), k=20)
>>> cs = CommonSpace(speech_dic, noise_dic)
>>> cs.fit_noise(np.abs(noise_spec))
"""
H = decompose_with_dict(noise_spec, self._noise_dic)
for i in range(H.shape[1]):
data = H[:, i]
x = np.delete(data, self._noise_list)
y = data[self._noise_list]
self._learning_table.append((x, y))