Merge branch 'main' into py_test_tts

main/xiaozhi-server/core/providers/llm/openai/openai.py

@@ -1,3 +1,4 @@
+import httpx
 import openai
 from openai.types import CompletionUsage
 from config.logger import setup_logging
@@ -16,6 +17,9 @@ class LLMProvider(LLMProviderBase):
             self.base_url = config.get("base_url")
         else:
             self.base_url = config.get("url")
+        # 增加timeout的配置项，单位为秒
+        timeout = config.get("timeout", 300)
+        self.timeout = int(timeout) if timeout else 300
 
         param_defaults = {
             "max_tokens": (500, int),
@@ -42,7 +46,7 @@ class LLMProvider(LLMProviderBase):
         model_key_msg = check_model_key("LLM", self.api_key)
         if model_key_msg:
             logger.bind(tag=TAG).error(model_key_msg)
-        self.client = openai.OpenAI(api_key=self.api_key, base_url=self.base_url)
+        self.client = openai.OpenAI(api_key=self.api_key, base_url=self.base_url, timeout=httpx.Timeout(self.timeout))
 
     def response(self, session_id, dialogue, **kwargs):
         try:

main/xiaozhi-server/core/providers/tts/base.py

@@ -161,13 +161,6 @@ class TTSProviderBase(ABC):
             else:
                 sentence_id = str(uuid.uuid4()).replace("-", "")
                 conn.sentence_id = sentence_id
-        self.tts_text_queue.put(
-            TTSMessageDTO(
-                sentence_id=sentence_id,
-                sentence_type=SentenceType.FIRST,
-                content_type=ContentType.ACTION,
-            )
-        )
         # 对于单句的文本，进行分段处理
         segments = re.split(r"([。！？!?；;\n])", content_detail)
         for seg in segments:
@@ -180,13 +173,6 @@ class TTSProviderBase(ABC):
                     content_file=content_file,
                 )
             )
-        self.tts_text_queue.put(
-            TTSMessageDTO(
-                sentence_id=sentence_id,
-                sentence_type=SentenceType.LAST,
-                content_type=ContentType.ACTION,
-            )
-        )
 
     async def open_audio_channels(self, conn):
         self.conn = conn
@@ -209,6 +195,8 @@ class TTSProviderBase(ABC):
         while not self.conn.stop_event.is_set():
             try:
                 message = self.tts_text_queue.get(timeout=1)
+                if message.sentence_type == SentenceType.FIRST:
+                    self.conn.client_abort = False
                 if self.conn.client_abort:
                     logger.bind(tag=TAG).info("收到打断信息，终止TTS文本处理线程")
                     continue
@@ -362,10 +350,8 @@ class TTSProviderBase(ABC):
         return audio_datas
 
     def _process_before_stop_play_files(self):
-        for tts_file, text in self.before_stop_play_files:
-            if tts_file and os.path.exists(tts_file):
-                audio_datas = self._process_audio_file(tts_file)
-                self.tts_audio_queue.put((SentenceType.MIDDLE, audio_datas, text))
+        for audio_datas, text in self.before_stop_play_files:
+            self.tts_audio_queue.put((SentenceType.MIDDLE, audio_datas, text))
         self.before_stop_play_files.clear()
         self.tts_audio_queue.put((SentenceType.LAST, [], None))
 

main/xiaozhi-server/core/providers/tts/fishspeech.py

@@ -85,8 +85,12 @@ class TTSProvider(TTSProviderBase):
         self.reference_id = (
             None if not config.get("reference_id") else config.get("reference_id")
         )
-        self.reference_audio = parse_string_to_list(config.get("reference_audio"))
-        self.reference_text = parse_string_to_list(config.get("reference_text"))
+        self.reference_audio = parse_string_to_list(
+             config.get('ref_audio')if config.get('ref_audio') else config.get("reference_audio")
+        )
+        self.reference_text = parse_string_to_list(
+             config.get('ref_text')if config.get('ref_text') else config.get("reference_text")
+        )
         self.format = config.get("response_format", "wav")
         self.audio_file_type = config.get("response_format", "wav")
         self.api_key = config.get("api_key", "YOUR_API_KEY")

main/xiaozhi-server/core/providers/tts/gpt_sovits_v2.py

@@ -12,8 +12,8 @@ class TTSProvider(TTSProviderBase):
         super().__init__(config, delete_audio_file)
         self.url = config.get("url")
         self.text_lang = config.get("text_lang", "zh")
-        self.ref_audio_path = config.get("ref_audio_path")
-        self.prompt_text = config.get("prompt_text")
+        self.ref_audio_path = config.get('ref_audio') if config.get('ref_audio') else config.get("ref_audio_path")
+        self.prompt_text = config.get('ref_text') if config.get('ref_text') else config.get("prompt_text")
         self.prompt_lang = config.get("prompt_lang", "zh")
 
         # 处理空字符串的情况

main/xiaozhi-server/core/providers/tts/gpt_sovits_v3.py

@@ -11,8 +11,8 @@ class TTSProvider(TTSProviderBase):
     def __init__(self, config, delete_audio_file):
         super().__init__(config, delete_audio_file)
         self.url = config.get("url")
-        self.refer_wav_path = config.get("refer_wav_path")
-        self.prompt_text = config.get("prompt_text")
+        self.refer_wav_path = config.get('ref_audio')if config.get('ref_audio') else config.get("refer_wav_path")
+        self.prompt_text = config.get('ref_text')if config.get('ref_text') else config.get("prompt_text")
         self.prompt_language = config.get("prompt_language")
         self.text_language = config.get("text_language", "audo")
 

main/xiaozhi-server/core/providers/tts/huoshan_double_stream.py

@@ -10,7 +10,6 @@ from config.logger import setup_logging
 from core.utils import opus_encoder_utils
 from core.utils.util import check_model_key
 from core.providers.tts.base import TTSProviderBase
-from core.handle.abortHandle import handleAbortMessage
 from core.providers.tts.dto.dto import SentenceType, ContentType, InterfaceType
 from asyncio import Task
 
@@ -175,6 +174,9 @@ class TTSProvider(TTSProviderBase):
     async def _ensure_connection(self):
         """建立新的WebSocket连接"""
         try:
+            if self.ws:
+                logger.bind(tag=TAG).info(f"使用已有链接...")
+                return self.ws
             logger.bind(tag=TAG).info("开始建立新连接...")
             ws_header = {
                 "X-Api-App-Key": self.appId,
@@ -200,6 +202,10 @@ class TTSProvider(TTSProviderBase):
                 logger.bind(tag=TAG).debug(
                     f"收到TTS任务｜{message.sentence_type.name} ｜ {message.content_type.name} | 会话ID: {self.conn.sentence_id}"
                 )
+
+                if message.sentence_type == SentenceType.FIRST:
+                    self.conn.client_abort = False
+
                 if self.conn.client_abort:
                     logger.bind(tag=TAG).info("收到打断信息，终止TTS文本处理线程")
                     continue
@@ -244,9 +250,12 @@ class TTSProvider(TTSProviderBase):
                     logger.bind(tag=TAG).info(
                         f"添加音频文件到待播放列表: {message.content_file}"
                     )
-                    self.before_stop_play_files.append(
-                        (message.content_file, message.content_detail)
-                    )
+                    if message.content_file and os.path.exists(message.content_file):
+                        # 先处理文件音频数据
+                        file_audio = self._process_audio_file(message.content_file)
+                        self.before_stop_play_files.append(
+                            (file_audio, message.content_detail)
+                        )
 
                 if message.sentence_type == SentenceType.LAST:
                     try:
@@ -295,25 +304,15 @@ class TTSProvider(TTSProviderBase):
     async def start_session(self, session_id):
         logger.bind(tag=TAG).info(f"开始会话～～{session_id}")
         try:
-            task = self._monitor_task
+            # 会话开始时检测上个会话的监听状态
             if (
-                task is not None
-                and isinstance(task, Task)
-                and not task.done()
+                self._monitor_task is not None
+                and isinstance(self._monitor_task, Task)
+                and not self._monitor_task.done()
             ):
-                logger.bind(tag=TAG).info("等待上一个监听任务结束...")
-                if self.ws is not None:
-                    logger.bind(tag=TAG).info("强制关闭上一个WebSocket连接以唤醒监听任务...")
-                    try:
-                        await self.ws.close()
-                    except Exception as e:
-                        logger.bind(tag=TAG).warning(f"关闭上一个ws异常: {e}")
-                    self.ws = None
-                try:
-                    await asyncio.wait_for(task, timeout=8)
-                except Exception as e:
-                    logger.bind(tag=TAG).warning(f"等待监听任务异常: {e}")
-                self._monitor_task = None
+                logger.bind(tag=TAG).info("检测到未完成的上个会话，关闭监听任务和连接...")
+                await self.close()
+
             # 建立新连接
             await self._ensure_connection()
 
@@ -336,19 +335,7 @@ class TTSProvider(TTSProviderBase):
         except Exception as e:
             logger.bind(tag=TAG).error(f"启动会话失败: {str(e)}")
             # 确保清理资源
-            if hasattr(self, "_monitor_task"):
-                try:
-                    self._monitor_task.cancel()
-                    await self._monitor_task
-                except:
-                    pass
-                self._monitor_task = None
-            if self.ws:
-                try:
-                    await self.ws.close()
-                except:
-                    pass
-                self.ws = None
+            await self.close()
             raise
 
     async def finish_session(self, session_id):
@@ -368,7 +355,7 @@ class TTSProvider(TTSProviderBase):
                 logger.bind(tag=TAG).info("会话结束请求已发送")
 
                 # 等待监听任务完成
-                if hasattr(self, "_monitor_task"):
+                if self._monitor_task:
                     try:
                         await self._monitor_task
                     except Exception as e:
@@ -378,28 +365,25 @@ class TTSProvider(TTSProviderBase):
                     finally:
                         self._monitor_task = None
 
-                # 关闭连接
-                await self.close()
         except Exception as e:
             logger.bind(tag=TAG).error(f"关闭会话失败: {str(e)}")
             # 确保清理资源
-            if hasattr(self, "_monitor_task"):
-                try:
-                    self._monitor_task.cancel()
-                    await self._monitor_task
-                except:
-                    pass
-                self._monitor_task = None
-            if self.ws:
-                try:
-                    await self.ws.close()
-                except:
-                    pass
-                self.ws = None
+            await self.close()
             raise
 
     async def close(self):
         """资源清理方法"""
+        # 取消监听任务
+        if self._monitor_task:
+            try:
+                self._monitor_task.cancel()
+                await self._monitor_task
+            except asyncio.CancelledError:
+                pass
+            except Exception as e:
+                logger.bind(tag=TAG).warning(f"关闭时取消监听任务错误: {e}")
+            self._monitor_task = None
+
         if self.ws:
             try:
                 await self.ws.close()
@@ -413,6 +397,7 @@ class TTSProvider(TTSProviderBase):
         is_first_sentence = True
         first_sentence_segment_count = 0  # 添加计数器
         try:
+            session_finished = False  # 标记会话是否正常结束
             while not self.conn.stop_event.is_set():
                 try:
                     # 确保 `recv()` 运行在同一个 event loop
@@ -466,6 +451,7 @@ class TTSProvider(TTSProviderBase):
                     elif res.optional.event == EVENT_SessionFinished:
                         logger.bind(tag=TAG).debug(f"会话结束～～")
                         self._process_before_stop_play_files()
+                        session_finished = True
                         break
                 except websockets.ConnectionClosed:
                     logger.bind(tag=TAG).warning("WebSocket连接已关闭")
@@ -476,15 +462,15 @@ class TTSProvider(TTSProviderBase):
                     )
                     traceback.print_exc()
                     break
-        finally:
-            # 确保清理资源
-            if self.ws:
+            # 仅在连接异常时才关闭
+            if not session_finished and self.ws:
                 try:
                     await self.ws.close()
                 except:
                     pass
                 self.ws = None
             # 监听任务退出时清理引用
+        finally:
             self._monitor_task = None
 
     async def send_event(

main/xiaozhi-server/core/providers/tts/linkerai.py

@@ -1,3 +1,4 @@
+import os
 import queue
 import asyncio
 import traceback
@@ -63,9 +64,12 @@ class TTSProvider(TTSProviderBase):
                     logger.bind(tag=TAG).info(
                         f"添加音频文件到待播放列表: {message.content_file}"
                     )
-                    self.before_stop_play_files.append(
-                        (message.content_file, message.content_detail)
-                    )
+                    if message.content_file and os.path.exists(message.content_file):
+                        # 先处理文件音频数据
+                        file_audio = self._process_audio_file(message.content_file)
+                        self.before_stop_play_files.append(
+                            (file_audio, message.content_detail)
+                        )
 
                 if message.sentence_type == SentenceType.LAST:
                     # 处理剩余的文本

main/xiaozhi-server/core/providers/vad/silero.py

@@ -23,22 +23,24 @@ class VADProvider(VADProviderBase):
 
         # 处理空字符串的情况
         threshold = config.get("threshold", "0.5")
+        threshold_low = config.get("threshold_low", "0.2")
         min_silence_duration_ms = config.get("min_silence_duration_ms", "1000")
 
         self.vad_threshold = float(threshold) if threshold else 0.5
+        self.vad_threshold_low = float(threshold_low) if threshold_low else 0.2
+
         self.silence_threshold_ms = (
             int(min_silence_duration_ms) if min_silence_duration_ms else 1000
         )
 
+        # 至少要多少帧才算有语音
+        self.frame_window_threshold = 3
+
     def is_vad(self, conn, opus_packet):
         try:
             pcm_frame = self.decoder.decode(opus_packet, 960)
             conn.client_audio_buffer.extend(pcm_frame)  # 将新数据加入缓冲区
 
-            # 初始化帧计数器
-            if not hasattr(conn, "client_voice_frame_count"):
-                conn.client_voice_frame_count = 0
-
             # 处理缓冲区中的完整帧（每次处理512采样点）
             client_have_voice = False
             while len(conn.client_audio_buffer) >= 512 * 2:
@@ -54,26 +56,30 @@ class VADProvider(VADProviderBase):
                 # 检测语音活动
                 with torch.no_grad():
                     speech_prob = self.model(audio_tensor, 16000).item()
-                is_voice = speech_prob >= self.vad_threshold
 
-                if is_voice:
-                    conn.client_voice_frame_count += 1
+                # 双阈值判断
+                if speech_prob >= self.vad_threshold:
+                    is_voice = True
+                elif speech_prob <= self.vad_threshold_low:
+                    is_voice = False
                 else:
-                    conn.client_voice_frame_count = 0
+                    is_voice = conn.last_is_voice
 
-                # 只有连续4帧检测到语音才认为有语音
-                client_have_voice = conn.client_voice_frame_count >= 4
+                # 声音没低于最低值则延续前一个状态，判断为有声音
+                conn.last_is_voice = is_voice
+
+                # 更新滑动窗口
+                conn.client_voice_window.append(is_voice)
+                client_have_voice = (conn.client_voice_window.count(True) >= self.frame_window_threshold)
 
                 # 如果之前有声音，但本次没有声音，且与上次有声音的时间差已经超过了静默阈值，则认为已经说完一句话
                 if conn.client_have_voice and not client_have_voice:
-                    stop_duration = (
-                        time.time() * 1000 - conn.client_have_voice_last_time
-                    )
+                    stop_duration = time.time() * 1000 - conn.last_activity_time
                     if stop_duration >= self.silence_threshold_ms:
                         conn.client_voice_stop = True
                 if client_have_voice:
                     conn.client_have_voice = True
-                    conn.client_have_voice_last_time = time.time() * 1000
+                    conn.last_activity_time = time.time() * 1000
 
             return client_have_voice
         except opuslib_next.OpusError as e:

main/xiaozhi-server/core/providers/vllm/openai.py

@@ -40,6 +40,7 @@ class VLLMProvider(VLLMProviderBase):
         self.client = openai.OpenAI(api_key=self.api_key, base_url=self.base_url)
 
     def response(self, question, base64_image):
+        question = question + "(请使用中文回复)"
         try:
             messages = [
                 {