def recover(result, label):
'''
:param result:
:return:
'''
key2id, id2key = index_utils.get_mapper(
index_file="../IndexFiles/en_char.csv")
# print("key2id:\n",key2id)
recoverd = []
print("recoverd:\n")
for i in range(result.shape[0]):
s = ""
for j in range(result.shape[1]):
if int(result[i][j]) == 0:
s += " "
else:
s += id2key[int(result[i][j])]
print("s:", s)
print("label:\n")
for i in range(label.shape[0]):
s = ""
for j in range(label.shape[1]):
if int(label[i][j]) == 0:
s += " "
else:
s += id2key[int(label[i][j])]
print("s:", s)
def get_features(audioFile,
labelFile,
max_frame_size=MAX_FRAME_SIZE,
max_label_size=MAX_LABEL_SIZE):
'''
从音频文件和标签文件得到提取出来的结果
:param audioFile:音频文件
:param labelFile:标签文件
:return:features，是一个列表，里面的元素分别为[mfcc,mfcc真实长度,labels结果，labels真实长度]
'''
features = []
key2id, id2key = index_utils.get_mapper(
index_file="../IndexFiles/en_char.csv")
# print("key2id:",key2id)
# print("id2key:",id2key)
# 提取音频特征
audio, sample_rate = librosa.core.load(path=audioFile, sr=None)
# print("audio:",audio)
# print("sample_rate:",sample_rate)
mfcc = dst.MFCC_Delta2(audio=audio, sample_rate=sample_rate)
# print("mfcc\n", mfcc.shape)
# padding label index list
mfcc_padded = np.zeros(shape=(max_frame_size, mfcc.shape[1]),
dtype=mfcc.dtype)
mfcc_padded[:mfcc.shape[0]] = mfcc[:]
#print("mfcc_padded:\n", mfcc_padded)
features.append(mfcc_padded) # 记录mfcc结果
features.append(mfcc.shape[0]) # 记录mfcc真实帧长度
# 得到标签字母/音素特征并且转化为id
label_in = open(file=labelFile)
lines = label_in.readlines()
index_list = []
str = ""
for line in lines:
line_list = line.strip().split(sep=" ")
# print("line_list:", line_list)
for char in line_list[-1]:
str += char
index_list.append(key2id[char])
label_in.close()
#print("str:",str)
#print("index_list:",index_list)
#padding label index list
index_list_padded = np.zeros(shape=(max_label_size, ),
dtype=type(index_list[0]))
index_list_padded[:len(index_list)] = index_list[:]
#print("index_list_padded:", index_list_padded)
features.append(index_list_padded) # 记录padding之后的索引列表结果
features.append(len(index_list)) # 记录索列表真实长度
return features
def getMaxSize(file_list):
'''
这个是一个辅助函数，用来获取信息
从根目录得到提取特征之后的最大特征长度和label长度
:param file_list:
:return:
'''
max_frame_size = 0
max_label_size = 0
#get mapper
key2id, id2key = index_utils.get_mapper(
index_file="../IndexFiles/en_char.csv")
# print("key2id:", key2id)
for file in file_list:
with open(file=file + ".WRD", encoding="utf-8",
errors="ignore") as file_label:
str = ""
lines = file_label.readlines()
for line in lines:
line_list = line.strip().split(sep=" ")
for char in line_list[-1]:
str += char
# print("str:",str)
# labels = file_label.readlines()[0].strip()
# index = indexLabel(labels=labels, label_map=char2id)
# # print("index:",index,len(index))
if len(str) > max_label_size:
max_label_size = len(str)
# get audios
audio, rate = librosa.core.load(file + ".WAV", sr=None)
# print("audio:\n", audio)
# print("rate:\n", rate)
features = dst.MFCC_Delta2(audio=audio, sample_rate=rate)
size = features.shape[0]
# print("size:",size)
if size > max_frame_size:
max_frame_size = size
print("max_frame_size:", max_frame_size)
print("max_label_size:", max_label_size)