/
language_model3.py
123 lines (105 loc) · 4.52 KB
/
language_model3.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
# !/usr/bin/env python3
# -*- coding: utf-8 -*-
#
# Copyright 2016-2099 Ailemon.net
#
# This file is part of ASRT Speech Recognition Tool.
#
# ASRT is free software: you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
# ASRT is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with ASRT. If not, see <https://www.gnu.org/licenses/>.
# ============================================================================
"""
@author: nl8590687
ASRT语音识别的语言模型
基于N-Gram的语言模型
"""
import os
from utils.ops import get_symbol_dict, get_language_model
class ModelLanguage:
"""
ASRT专用N-Gram语言模型
"""
def __init__(self, model_path: str):
self.model_path = model_path
self.dict_pinyin = dict()
self.model1 = dict()
self.model2 = dict()
def load_model(self):
"""
加载N-Gram语言模型到内存
"""
self.dict_pinyin = get_symbol_dict('dict.txt')
self.model1 = get_language_model(os.path.join(self.model_path, 'language_model1.txt'))
self.model2 = get_language_model(os.path.join(self.model_path, 'language_model2.txt'))
model = (self.dict_pinyin, self.model1, self.model2)
return model
def pinyin_to_text(self, list_pinyin: list, beam_size: int = 100) -> str:
"""
拼音转文本，一次性取得全部结果
"""
result = list()
tmp_result_last = list()
for item_pinyin in list_pinyin:
tmp_result = self.pinyin_stream_decode(tmp_result_last, item_pinyin, beam_size)
if len(tmp_result) == 0 and len(tmp_result_last) > 0:
result.append(tmp_result_last[0][0])
tmp_result = self.pinyin_stream_decode([], item_pinyin, beam_size)
if len(tmp_result) > 0:
result.append(tmp_result[0][0])
tmp_result = []
tmp_result_last = tmp_result
if len(tmp_result_last) > 0:
result.append(tmp_result_last[0][0])
return ''.join(result)
def pinyin_stream_decode(self, temple_result: list,
item_pinyin: str,
beam_size: int = 100) -> list:
"""
拼音流式解码，逐字转换，每次返回中间结果
"""
# 如果这个拼音不在汉语拼音字典里的话，直接返回空列表，不做decode
if item_pinyin not in self.dict_pinyin:
return []
# 获取拼音下属的字的列表，cur_words包含了该拼音对应的所有的字
cur_words = self.dict_pinyin[item_pinyin]
# 第一个字做初始处理
if len(temple_result) == 0:
lst_result = list()
for word in cur_words:
# 添加该字到可能的句子列表，设置初始概率为1.0
lst_result.append([word, 1.0])
return lst_result
# 开始处理已经至少有一个字的中间结果情况
new_result = list()
for sequence in temple_result:
for cur_word in cur_words:
# 得到2-gram的汉字子序列
tuple2_word = sequence[0][-1] + cur_word
if tuple2_word not in self.model2:
# 如果2-gram子序列不存在
continue
# 计算状态转移概率
prob_origin = sequence[1] # 原始概率
count_two_word = float(self.model2[tuple2_word]) # 二字频数
count_one_word = float(self.model1[tuple2_word[-2]]) # 单字频数
cur_probility = prob_origin * count_two_word / count_one_word
new_result.append([sequence[0] + cur_word, cur_probility])
new_result = sorted(new_result, key=lambda x: x[1], reverse=True)
if len(new_result) > beam_size:
return new_result[0:beam_size]
return new_result
if __name__ == '__main__':
ml = ModelLanguage('model_language')
ml.load_model()
_str_pinyin = ['zhe4', 'zhen1', 'shi4', 'ji2', 'hao3', 'de5']
_RESULT = ml.pinyin_to_text(_str_pinyin)
print('语音转文字结果:\n', _RESULT)