修复：流式LLM时如果结尾没有中文标点符号会丢句子的bug

Ikaros-521 · Oct 2, 2024 · 89011a2 · 89011a2
2 parents ea21bdc + c02512d
commit 89011a2
Show file tree

Hide file tree

Showing 8 changed files with 281 additions and 185 deletions.
diff --git a/README.md b/README.md
@@ -1,16 +1,27 @@
-# ✨ Luna AI ✨
+<div align="center">
+  <a href="https://ikaros-521.github.io/Luna-Docs/site/">
+    <img src="https://raw.githubusercontent.com/Ikaros-521/AI-Vtuber/refs/heads/main/ui/icon.png" width="240" height="240" alt="点我跳转文档">
+  </a>
+</div>
+
+<div align="center">
+
+# ✨ Luna AI  ✨
+
 
 
-[![][FOSSA-Status]][FOSSA-Status]
 [![][python]][python]
 [![][github-release-shield]][github-release-link]
 [![][github-stars-shield]][github-stars-link]
 [![][github-forks-shield]][github-forks-link]
+[![][github-issues-shield]][github-issues-link]  
 [![][github-contributors-shield]][github-contributors-link]
-[![][github-issues-shield]][github-issues-link]
 [![][github-license-shield]][github-license-link]
+[![][FOSSA-Status]][FOSSA-Status]
 
 
+</div>
+
 `Luna AI` 是一款结合了最先进技术的虚拟AI主播。它的核心是一系列高效的人工智能模型和平台，包括 `ChatterBot、GPT、Claude、langchain、chatglm、text-generation-webui、讯飞星火、智谱AI、谷歌Bard、通义星尘、阿里云百炼（通义千问、百川、月之暗面、零一万物、MiniMax）、千帆大模型（文心一言）、Gemini、Kimi Chat、QAnything、koboldcpp、FastGPT、Ollama、One-API、AnythingLLM、LLM_TPU、Dify、火山引擎（豆包）`。这些模型既可以在本地运行，也可以通过云端服务提供支持。当然，为了让对话照进现实，还结合了多模态模型，包括 `Gemini、glm-4v` 的图像识别能力，获取电脑画面进行分析讲解。  
 
 `Luna AI` 的外观由 `Live2D、Vtube Studio、xuniren、UE5 结合 Audio2Face、EasyAIVtuber、数字人视频播放器（Easy-Wav2Lip、Sadtalker、GeneFace++、MuseTalk、AniTalker、本地视频）、metahuman-stream（ernerf、musetalk、wav2lip）、DH_live、live2d-TTS-LLM-GPT-SoVITS-Vtuber` 技术打造，为用户提供了一个生动、互动的虚拟形象。这使得 `Luna AI` 能够在各大直播平台，如 `Bilibili、抖音、快手、微信视频号、拼多多、1688、斗鱼、YouTube、Twitch 和 TikTok`，进行实时互动直播。当然，它也可以在本地环境中与您进行个性化对话。

diff --git a/config.json b/config.json
@@ -17,6 +17,15 @@
     "enable": false,
     "copywriting": "{username}说：{comment}"
   },
+  "reply_template": {
+    "enable": false,
+    "username_max_len": 10,
+    "copywriting": [
+      "回复{username}{data}",
+      "{username}{data}"
+    ],
+    "username_max_le": 10
+  },
   "comment_log_type": "回答",
   "visual_body": "其他",
   "xuniren": {
@@ -73,21 +82,6 @@
     "proxy_server": "127.0.0.1",
     "proxy_port": "10809"
   },
-  "read_username": {
-    "enable": false,
-    "username_max_len": 10,
-    "voice_change": false,
-    "reply_before": [
-      "回复{username}",
-      "{username}",
-      "回{username}"
-    ],
-    "reply_after": [
-      "以上内容回复{username}",
-      "回的{username}",
-      "以上回复{username}"
-    ]
-  },
   "read_comment": {
     "enable": true,
     "read_username_enable": true,
@@ -263,6 +257,11 @@
   },
   "search_online": {
     "enable": false,
+    "keyword_enable": true,
+    "before_keyword": [
+      "联网",
+      "在线"
+    ],
     "engine": "baidu",
     "engine_id": 1,
     "count": 1,
@@ -570,7 +569,7 @@
   },
   "local_qa": {
     "periodic_trigger": {
-      "enable": true,
+      "enable": false,
       "periodic_time_min": 10,
       "periodic_time_max": 30,
       "trigger_num_min": 0,

diff --git a/config.json.bak b/config.json.bak
@@ -17,6 +17,15 @@
     "enable": false,
     "copywriting": "{username}说：{comment}"
   },
+  "reply_template": {
+    "enable": false,
+    "username_max_len": 10,
+    "copywriting": [
+      "回复{username}{data}",
+      "{username}{data}"
+    ],
+    "username_max_le": 10
+  },
   "comment_log_type": "回答",
   "visual_body": "其他",
   "xuniren": {
@@ -73,21 +82,6 @@
     "proxy_server": "127.0.0.1",
     "proxy_port": "10809"
   },
-  "read_username": {
-    "enable": false,
-    "username_max_len": 10,
-    "voice_change": false,
-    "reply_before": [
-      "回复{username}",
-      "{username}",
-      "回{username}"
-    ],
-    "reply_after": [
-      "以上内容回复{username}",
-      "回的{username}",
-      "以上回复{username}"
-    ]
-  },
   "read_comment": {
     "enable": true,
     "read_username_enable": true,
@@ -263,6 +257,11 @@
   },
   "search_online": {
     "enable": false,
+    "keyword_enable": true,
+    "before_keyword": [
+      "联网",
+      "在线"
+    ],
     "engine": "baidu",
     "engine_id": 1,
     "count": 1,
@@ -570,7 +569,7 @@
   },
   "local_qa": {
     "periodic_trigger": {
-      "enable": true,
+      "enable": false,
       "periodic_time_min": 10,
       "periodic_time_max": 30,
       "trigger_num_min": 0,

diff --git a/main.py b/main.py
@@ -182,7 +182,7 @@ async def send(msg: SendMessage):
 
                 try:
                     tmp_json = msg.dict()
-                    logger.info(f"API收到数据：{tmp_json}")
+                    logger.info(f"内部HTTP API send接口收到数据：{tmp_json}")
                     data_json = tmp_json["data"]
                     if "type" not in data_json:
                         data_json["type"] = tmp_json["type"]
@@ -228,7 +228,7 @@ async def callback(msg: CallbackMessage):
 
                 try:
                     data_json = msg.dict()
-                    logger.info(f"API收到数据：{data_json}")
+                    logger.info(f"内部HTTP API callback接口收到数据：{data_json}")
 
                     # 音频播放完成
                     if data_json["type"] in ["audio_playback_completed"]:

diff --git a/utils/audio.py b/utils/audio.py
@@ -670,22 +670,6 @@ def audio_synthesis(self, message):
                 if "insert_index" in data_json:
                     data_json["insert_index"] = message["insert_index"]
 
-                # 回复时是否念用户名字
-                if self.config.get("read_username", "enable"):
-                    # 由于线程是独立的，所以回复音频的合成会慢于本地音频直接播放，所以以倒述的形式回复
-                    tmp_message = deepcopy(message)
-                    tmp_message['type'] = "reply"
-                    tmp_message['content'] = random.choice(self.config.get("read_username", "reply_after"))
-                    if "{username}" in tmp_message['content']:
-                        tmp_message['content'] = tmp_message['content'].format(username=message['username'][:self.config.get("read_username", "username_max_len")])
-
-                    logger.info(f"tmp_message={tmp_message}")
-
-                    self.data_priority_insert("等待合成消息", tmp_message)
-                # else:
-                #     logger.info(f"message={message}")
-                #     self.data_priority_insert("等待合成消息", message)
-
                 # 是否开启了音频播放
                 if self.config.get("play_audio", "enable"):
                     self.data_priority_insert("等待合成消息", data_json)
@@ -708,18 +692,6 @@ def audio_synthesis(self, message):
                     self.data_priority_insert("等待合成消息", data_json)
                 return
 
-            # 只有信息类型是 弹幕，才会进行念用户名
-            elif message['type'] == "comment":
-                # 回复时是否念用户名字
-                if self.config.get("read_username", "enable"):
-                    tmp_message = deepcopy(message)
-                    tmp_message['type'] = "reply"
-                    tmp_message['content'] = random.choice(self.config.get("read_username", "reply_before"))
-                    if "{username}" in tmp_message['content']:
-                        # 将用户名中特殊字符替换为空
-                        message['username'] = self.common.replace_special_characters(message['username'], "！!@#￥$%^&*_-+/——=()（）【】}|{:;<>~`\\")
-                        tmp_message['content'] = tmp_message['content'].format(username=message['username'][:self.config.get("read_username", "username_max_len")])
-                    self.data_priority_insert("等待合成消息", tmp_message)
             # 闲时任务
             elif message['type'] == "idle_time_task":
                 if message['content_type'] in ["comment", "reread"]:
@@ -1238,13 +1210,13 @@ async def voice_change_and_put_to_queue(message, voice_tmp_path):
                 data_json["insert_index"] = message["insert_index"]
 
             # 区分消息类型是否是 回复xxx 并且 关闭了变声
-            if message["type"] == "reply" and False == self.config.get("read_username", "voice_change"):
+            if message["type"] == "reply":
                 # 是否开启了音频播放，如果没开，则不会传文件路径给播放队列
                 if self.config.get("play_audio", "enable"):
                     self.data_priority_insert("待播放音频列表", data_json)
                     return True
             # 区分消息类型是否是 念弹幕 并且 关闭了变声
-            elif message["type"] == "read_comment" and False == self.config.get("read_comment", "voice_change"):
+            elif message["type"] == "read_comment" and not self.config.get("read_comment", "voice_change"):
                 # 是否开启了音频播放，如果没开，则不会传文件路径给播放队列
                 if self.config.get("play_audio", "enable"):
                     self.data_priority_insert("待播放音频列表", data_json)
@@ -1274,7 +1246,7 @@ async def voice_change_and_put_to_queue(message, voice_tmp_path):
 
             return False
 
-        logger.info(f"{message['tts_type']}合成成功，合成内容：【{message['content']}】，输出到={voice_tmp_path}")
+        logger.info(f"[{message['tts_type']}]合成成功，合成内容：【{message['content']}】，音频存储在 {voice_tmp_path}")
 
         await voice_change_and_put_to_queue(message, voice_tmp_path)  
 

diff --git a/utils/common.py b/utils/common.py
@@ -73,6 +73,30 @@ def is_valid_port(self, port):
         except ValueError:
             return False
 
+    # 判断传入的字符串是否是文件夹路径或文件路径，且此文件夹路径或文件路径是否存在，返回bool
+    def is_dir_or_file(self, path: str, type: str="all"):
+        """判断传入的字符串是否是文件夹路径或文件路径，且此文件夹路径或文件路径是否存在，返回bool
+
+        Args:
+            path (str): 文件夹路径或文件路径
+            type (str, optional): 检测类型. Defaults to "all".
+
+        Returns:
+            bool: 结果
+        """
+        if type == "dir":
+            if os.path.isdir(path):
+                return True
+            return False
+        elif type == "file":
+            if os.path.isfile(path):
+                return True
+            return False
+        else:
+            if os.path.isdir(path) or os.path.isfile(path):
+                return True
+            return False
+
     # 识别操作系统
     def detect_os(self):
         """
@@ -764,6 +788,23 @@ def get_random_str_in_list_and_format(self, ori_content: str = None, ori_list: l
 
         return {"ret": 0, "content": content}
 
+    def get_list_random_or_default(self, strings: list, default_value):
+        """
+
+        从列表中随机选择一个字符串，如果列表为空，则返回默认值。
+
+        参数:
+            strings (list of str): 字符串列表。
+            default_value (str): 默认值。
+
+        返回:
+            str: 随机选择的字符串或默认值。
+        """
+        if not strings:  # 如果列表是空的
+            return default_value
+        else:
+            return random.choice(strings)
+
     """
     
             .@@@             @@@        @@^ =@@@@@@@@    /@@ /@@              =@@@@@*,@@\]]]]  ,@@@@@@@@@@@@*                      .@@@         @@/.\]`@@@       =@@\]]]]]]]   =@@..@@@@@@@@@   =@@\   /@@^           
@@ -946,6 +987,7 @@ def get_live2d_model_name(self, path):
         else:
             return None
 
+
 
 
     """
@@ -995,7 +1037,7 @@ def write_content_to_file(self, file_path, content, write_log=True):
             with open(file_path, 'w', encoding='utf-8') as file:
                 file.write(content)
 
-            if write_log == True:
+            if write_log:
                 logger.info(f"写入文件:{file_path}，内容：【{content}】")
 
             return True