feat: ngram语言模型流式解码实现

asrserver_http.py

@@ -31,7 +31,7 @@ from flask import Flask, Response, request
 from speech_model import ModelSpeech
 from speech_model_zoo import SpeechModel251BN
 from speech_features import Spectrogram
-from LanguageModel2 import ModelLanguage
+from language_model3 import ModelLanguage
 from utils.ops import decode_wav_bytes
 
 API_STATUS_CODE_OK = 200000 # OK
@@ -62,7 +62,7 @@ ms = ModelSpeech(sm251bn, feat, max_label_length=64)
 ms.load_model('save_models/' + sm251bn.get_model_name() + '.model.h5')
 
 ml = ModelLanguage('model_language')
-ml.LoadModel()
+ml.load_model()
 
 
 class AsrtApiResponse:
@@ -131,7 +131,7 @@ def recognition_post(level):
 
             seq_pinyin = request_data['sequence_pinyin']
 
-            result = ml.SpeechToText(seq_pinyin)
+            result = ml.pinyin_to_text(seq_pinyin)
 
             json_data = AsrtApiResponse(API_STATUS_CODE_OK, 'language level')
             json_data.result = result
@@ -150,7 +150,7 @@ def recognition_post(level):
             wavdata = decode_wav_bytes(samples_data=wavdata_bytes,
                                         channels=channels, byte_width=byte_width)
             result_speech = ms.recognize_speech(wavdata, sample_rate)
-            result = ml.SpeechToText(result_speech)
+            result = ml.pinyin_to_text(result_speech)
 
             json_data = AsrtApiResponse(API_STATUS_CODE_OK, 'all level')
             json_data.result = result

language_model3.py

@@ -0,0 +1,121 @@
+#!/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)

predict_speech_file.py

@@ -28,7 +28,7 @@ import os
 from speech_model import ModelSpeech
 from speech_model_zoo import SpeechModel251BN
 from speech_features import Spectrogram
-from LanguageModel2 import ModelLanguage
+from language_model3 import ModelLanguage
 
 os.environ["CUDA_VISIBLE_DEVICES"] = "0"
 
@@ -49,7 +49,7 @@ res = ms.recognize_speech_from_file('filename.wav')
 print('*[提示] 声学模型语音识别结果：\n', res)
 
 ml = ModelLanguage('model_language')
-ml.LoadModel()
+ml.load_model()
 str_pinyin = res
-res = ml.SpeechToText(str_pinyin)
+res = ml.pinyin_to_text(str_pinyin)
 print('语音识别最终结果：\n',res)