update:python单模块部署声纹识别对接

main/xiaozhi-server/config.yaml

@@ -137,6 +137,15 @@ plugins:
       - ".wav"
       - ".p3"
     refresh_time: 300 # 刷新音乐列表的时间间隔，单位为秒
+  # 声纹识别配置
+  voiceprint:
+    # 声纹接口地址
+    url: 
+    # 说话人配置：speaker_id,名称,描述
+    speakers:
+      - "test1,张三,张三是一个程序员"
+      - "test2,李四,李四是一个产品经理"
+      - "test3,王五,王五是一个设计师"
 
 # #####################################################################################
 # ################################以下是角色模型配置######################################

main/xiaozhi-server/core/connection.py

@@ -595,8 +595,33 @@ class ConnectionHandler:
         self.logger.bind(tag=TAG).info(f"大模型收到用户消息: {query}")
         self.llm_finish_task = False
 
+        # 检查是否是JSON格式的消息（包含说话人信息）
+        enhanced_query = query
+        try:
+            if query.strip().startswith('{') and query.strip().endswith('}'):
+                data = json.loads(query)
+                if 'speaker' in data and 'content' in data:
+                    # 直接使用JSON格式，不重新格式化
+                    enhanced_query = query
+                    self.logger.bind(tag=TAG).info(f"识别到说话人: {data['speaker']}")
+                else:
+                    # 如果有说话人信息但不是JSON格式，按原逻辑处理
+                    if hasattr(self, 'current_speaker') and self.current_speaker:
+                        enhanced_query = f"[说话人: {self.current_speaker}] {query}"
+                        self.logger.bind(tag=TAG).info(f"识别到说话人: {self.current_speaker}")
+            else:
+                # 如果有说话人信息但不是JSON格式，按原逻辑处理
+                if hasattr(self, 'current_speaker') and self.current_speaker:
+                    enhanced_query = f"[说话人: {self.current_speaker}] {query}"
+                    self.logger.bind(tag=TAG).info(f"识别到说话人: {self.current_speaker}")
+        except json.JSONDecodeError:
+            # JSON解析失败，按原逻辑处理
+            if hasattr(self, 'current_speaker') and self.current_speaker:
+                enhanced_query = f"[说话人: {self.current_speaker}] {query}"
+                self.logger.bind(tag=TAG).info(f"识别到说话人: {self.current_speaker}")
+
         if not tool_call:
-            self.dialogue.put(Message(role="user", content=query))
+            self.dialogue.put(Message(role="user", content=enhanced_query))
 
         # Define intent functions
         functions = None
@@ -609,7 +634,7 @@ class ConnectionHandler:
             memory_str = None
             if self.memory is not None:
                 future = asyncio.run_coroutine_threadsafe(
-                    self.memory.query_memory(query), self.loop
+                    self.memory.query_memory(enhanced_query), self.loop
                 )
                 memory_str = future.result()
 

main/xiaozhi-server/core/handle/receiveAudioHandle.py

@@ -4,6 +4,7 @@ from core.utils.output_counter import check_device_output_limit
 from core.handle.abortHandle import handleAbortMessage
 import time
 import asyncio
+import json
 from core.handle.sendAudioHandle import SentenceType
 from core.utils.util import audio_to_data
 
@@ -38,6 +39,31 @@ async def resume_vad_detection(conn):
 
 
 async def startToChat(conn, text):
+    # 检查输入是否是JSON格式（包含说话人信息）
+    speaker_name = None
+    actual_text = text
+    
+    try:
+        # 尝试解析JSON格式的输入
+        if text.strip().startswith('{') and text.strip().endswith('}'):
+            data = json.loads(text)
+            if 'speaker' in data and 'content' in data:
+                speaker_name = data['speaker']
+                actual_text = data['content']
+                conn.logger.bind(tag=TAG).info(f"解析到说话人信息: {speaker_name}")
+                
+                # 直接使用JSON格式的文本，不解析
+                actual_text = text
+    except (json.JSONDecodeError, KeyError):
+        # 如果解析失败，继续使用原始文本
+        pass
+    
+    # 保存说话人信息到连接对象
+    if speaker_name:
+        conn.current_speaker = speaker_name
+    else:
+        conn.current_speaker = None
+
     if conn.need_bind:
         await check_bind_device(conn)
         return
@@ -52,16 +78,16 @@ async def startToChat(conn, text):
     if conn.client_is_speaking:
         await handleAbortMessage(conn)
 
-    # 首先进行意图分析
+    # 首先进行意图分析，使用实际文本内容
-    intent_handled = await handle_user_intent(conn, text)
+    intent_handled = await handle_user_intent(conn, actual_text)
 
     if intent_handled:
         # 如果意图已被处理，不再进行聊天
         return
 
-    # 意图未被处理，继续常规聊天流程
+    # 意图未被处理，继续常规聊天流程，使用实际文本内容
-    await send_stt_message(conn, text)
+    await send_stt_message(conn, actual_text)
-    conn.executor.submit(conn.chat, text)
+    conn.executor.submit(conn.chat, actual_text)
 
 
 async def no_voice_close_connect(conn, have_voice):

main/xiaozhi-server/core/handle/sendAudioHandle.py

@@ -76,7 +76,6 @@ async def sendAudio(conn, audios, pre_buffer=True):
     frame_duration = 60  # 帧时长（毫秒），匹配 Opus 编码
     start_time = time.perf_counter()
     play_position = 0
-    last_reset_time = time.perf_counter()  # 记录最后的重置时间
 
     # 仅当第一句话时执行预缓冲
     if pre_buffer:

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

@@ -1,15 +1,20 @@
 import os
 import wave
-import copy
 import uuid
 import queue
 import asyncio
 import traceback
 import threading
 import opuslib_next
+import json
+import io
+import aiohttp
+import time
+import concurrent.futures
 from abc import ABC, abstractmethod
 from config.logger import setup_logging
-from typing import Optional, Tuple, List
+from urllib.parse import urlparse, parse_qs
+from typing import Optional, Tuple, List, Dict, Any
 from core.handle.receiveAudioHandle import startToChat
 from core.handle.reportHandle import enqueue_asr_report
 from core.utils.util import remove_punctuation_and_length
@@ -18,16 +23,145 @@ from core.handle.receiveAudioHandle import handleAudioMessage
 TAG = __name__
 logger = setup_logging()
 
+# 创建全局线程池执行器用于CPU密集型操作
+executor = concurrent.futures.ThreadPoolExecutor(max_workers=4)
+
+class VoiceprintProvider:
+    """声纹识别服务提供者"""
+    
+    def __init__(self, config: dict):
+        self.original_url = config.get("url", "")
+        self.speakers = config.get("speakers", [])
+        self.speaker_map = self._parse_speakers()
+        
+        # 解析API地址和密钥
+        self.api_url = None
+        self.api_key = None
+        self.speaker_ids = []
+        
+        if not self.original_url:
+            logger.bind(tag=TAG).warning("声纹识别URL未配置，声纹识别将被禁用")
+            self.enabled = False
+        else:
+            # 解析URL和key
+            parsed_url = urlparse(self.original_url)
+            base_url = f"{parsed_url.scheme}://{parsed_url.netloc}"
+            
+            # 从查询参数中提取key
+            query_params = parse_qs(parsed_url.query)
+            self.api_key = query_params.get('key', [''])[0]
+            
+            if not self.api_key:
+                logger.bind(tag=TAG).error("URL中未找到key参数，声纹识别将被禁用")
+                self.enabled = False
+            else:
+                # 构造identify接口地址
+                self.api_url = f"{base_url}/voiceprint/identify"
+                
+                # 提取speaker_ids
+                for speaker_str in self.speakers:
+                    try:
+                        parts = speaker_str.split(",", 2)
+                        if len(parts) >= 1:
+                            speaker_id = parts[0].strip()
+                            self.speaker_ids.append(speaker_id)
+                    except Exception:
+                        continue
+                
+                # 检查是否有有效的说话人配置
+                if not self.speaker_ids:
+                    logger.bind(tag=TAG).warning("未配置有效的说话人，声纹识别将被禁用")
+                    self.enabled = False
+                else:
+                    self.enabled = True
+                    logger.bind(tag=TAG).info(f"声纹识别已配置: API={self.api_url}, 说话人={len(self.speaker_ids)}个")
+    
+    def _parse_speakers(self) -> Dict[str, Dict[str, str]]:
+        """解析说话人配置"""
+        speaker_map = {}
+        for speaker_str in self.speakers:
+            try:
+                parts = speaker_str.split(",", 2)
+                if len(parts) >= 3:
+                    speaker_id, name, description = parts[0].strip(), parts[1].strip(), parts[2].strip()
+                    speaker_map[speaker_id] = {
+                        "name": name,
+                        "description": description
+                    }
+            except Exception as e:
+                logger.bind(tag=TAG).warning(f"解析说话人配置失败: {speaker_str}, 错误: {e}")
+        return speaker_map
+    
+    async def identify_speaker(self, audio_data: bytes, session_id: str) -> Optional[str]:
+        """识别说话人"""
+        if not self.enabled or not self.api_url or not self.api_key:
+            logger.bind(tag=TAG).debug("声纹识别功能已禁用或未配置，跳过识别")
+            return None
+            
+        try:
+            api_start_time = time.monotonic()
+            
+            # 准备请求头
+            headers = {
+                'Authorization': f'Bearer {self.api_key}',
+                'Accept': 'application/json'
+            }
+            
+            # 准备multipart/form-data数据
+            data = aiohttp.FormData()
+            data.add_field('speaker_ids', ','.join(self.speaker_ids))
+            data.add_field('file', audio_data, filename='audio.wav', content_type='audio/wav')
+            
+            timeout = aiohttp.ClientTimeout(total=10)
+            
+            # 网络请求
+            async with aiohttp.ClientSession(timeout=timeout) as session:
+                async with session.post(self.api_url, headers=headers, data=data) as response:
+                    
+                    if response.status == 200:
+                        result = await response.json()
+                        speaker_id = result.get("speaker_id")
+                        score = result.get("score", 0)
+                        total_elapsed_time = time.monotonic() - api_start_time
+                        
+                        logger.bind(tag=TAG).info(f"声纹识别耗时: {total_elapsed_time:.3f}s")
+                        
+                        # 置信度检查
+                        if score < 0.5:
+                            logger.bind(tag=TAG).warning(f"声纹识别置信度较低: {score:.3f}")
+                        
+                        if speaker_id and speaker_id in self.speaker_map:
+                            result_name = self.speaker_map[speaker_id]["name"]
+                            return result_name
+                        else:
+                            logger.bind(tag=TAG).warning(f"未识别的说话人ID: {speaker_id}")
+                            return "未知说话人"
+                    else:
+                        logger.bind(tag=TAG).error(f"声纹识别API错误: HTTP {response.status}")
+                        return None
+                        
+        except asyncio.TimeoutError:
+            elapsed = time.monotonic() - api_start_time
+            logger.bind(tag=TAG).error(f"声纹识别超时: {elapsed:.3f}s")
+            return None
+        except Exception as e:
+            elapsed = time.monotonic() - api_start_time
+            logger.bind(tag=TAG).error(f"声纹识别失败: {e}")
+            return None
+
 
 class ASRProviderBase(ABC):
     def __init__(self):
-        pass
+        self.voiceprint_provider = None
+
+    def init_voiceprint(self, voiceprint_config: dict):
+        """初始化声纹识别"""
+        if voiceprint_config:
+            self.voiceprint_provider = VoiceprintProvider(voiceprint_config)
+            logger.bind(tag=TAG).info("声纹识别模块已初始化")
 
     # 打开音频通道
-    # 这里默认是非流式的处理方式
-    # 流式处理方式请在子类中重写
     async def open_audio_channels(self, conn):
-        # tts 消化线程
         conn.asr_priority_thread = threading.Thread(
             target=self.asr_text_priority_thread, args=(conn,), daemon=True
         )
@@ -52,41 +186,173 @@ class ASRProviderBase(ABC):
                 continue
 
     # 接收音频
-    # 这里默认是非流式的处理方式
-    # 流式处理方式请在子类中重写
     async def receive_audio(self, conn, audio, audio_have_voice):
         if conn.client_listen_mode == "auto" or conn.client_listen_mode == "realtime":
             have_voice = audio_have_voice
         else:
             have_voice = conn.client_have_voice
-        # 如果本次没有声音，本段也没声音，就把声音丢弃了
+        
         conn.asr_audio.append(audio)
-        if have_voice == False and conn.client_have_voice == False:
+        if not have_voice and not conn.client_have_voice:
             conn.asr_audio = conn.asr_audio[-10:]
             return
 
-        # 如果本段有声音，且已经停止了
         if conn.client_voice_stop:
-            asr_audio_task = copy.deepcopy(conn.asr_audio)
+            asr_audio_task = conn.asr_audio.copy()
             conn.asr_audio.clear()
-
-            # 音频太短了，无法识别
             conn.reset_vad_states()
+
             if len(asr_audio_task) > 15:
                 await self.handle_voice_stop(conn, asr_audio_task)
 
     # 处理语音停止
-    async def handle_voice_stop(self, conn, asr_audio_task):
+    async def handle_voice_stop(self, conn, asr_audio_task: List[bytes]):
-        raw_text, _ = await self.speech_to_text(
+        """并行处理ASR和声纹识别"""
-            asr_audio_task, conn.session_id, conn.audio_format
+        try:
-        )  # 确保ASR模块返回原始文本
+            total_start_time = time.monotonic()
-        conn.logger.bind(tag=TAG).info(f"识别文本: {raw_text}")
+            
-        text_len, _ = remove_punctuation_and_length(raw_text)
+            # 准备音频数据
-        self.stop_ws_connection()
+            if conn.audio_format == "pcm":
-        if text_len > 0:
+                pcm_data = asr_audio_task
-            # 使用自定义模块进行上报
+            else:
-            await startToChat(conn, raw_text)
+                pcm_data = self.decode_opus(asr_audio_task)
-            enqueue_asr_report(conn, raw_text, asr_audio_task)
+            
+            combined_pcm_data = b"".join(pcm_data)
+            
+            # 预先准备WAV数据
+            wav_data = None
+            if self.voiceprint_provider and combined_pcm_data:
+                wav_data = self._pcm_to_wav(combined_pcm_data)
+            
+            
+            # 定义ASR任务
+            def run_asr():
+                start_time = time.monotonic()
+                try:
+                    import asyncio
+                    loop = asyncio.new_event_loop()
+                    asyncio.set_event_loop(loop)
+                    try:
+                        result = loop.run_until_complete(
+                            self.speech_to_text(asr_audio_task, conn.session_id, conn.audio_format)
+                        )
+                        end_time = time.monotonic()
+                        logger.bind(tag=TAG).info(f"ASR耗时: {end_time - start_time:.3f}s")
+                        return result
+                    finally:
+                        loop.close()
+                except Exception as e:
+                    end_time = time.monotonic()
+                    logger.bind(tag=TAG).error(f"ASR失败: {e}")
+                    return ("", None)
+            
+            # 定义声纹识别任务
+            def run_voiceprint():
+                if not wav_data:
+                    return None
+                start_time = time.monotonic()
+                try:
+                    import asyncio
+                    loop = asyncio.new_event_loop()
+                    asyncio.set_event_loop(loop)
+                    try:
+                        result = loop.run_until_complete(
+                            self.voiceprint_provider.identify_speaker(wav_data, conn.session_id)
+                        )
+                        return result
+                    finally:
+                        loop.close()
+                except Exception as e:
+                    logger.bind(tag=TAG).error(f"声纹识别失败: {e}")
+                    return None
+            
+            # 使用线程池执行器并行运行
+            parallel_start_time = time.monotonic()
+            
+            with concurrent.futures.ThreadPoolExecutor(max_workers=2) as thread_executor:
+                asr_future = thread_executor.submit(run_asr)
+                
+                if self.voiceprint_provider and wav_data:
+                    voiceprint_future = thread_executor.submit(run_voiceprint)
+                    
+                    # 等待两个线程都完成
+                    asr_result = asr_future.result(timeout=15)
+                    voiceprint_result = voiceprint_future.result(timeout=15)
+                    
+                    results = {"asr": asr_result, "voiceprint": voiceprint_result}
+                else:
+                    asr_result = asr_future.result(timeout=15)
+                    results = {"asr": asr_result, "voiceprint": None}
+            
+            parallel_execution_time = time.monotonic() - parallel_start_time
+            
+            # 处理结果
+            raw_text, file_path = results.get("asr", ("", None))
+            speaker_name = results.get("voiceprint", None)
+            
+            # 记录识别结果
+            if raw_text:
+                logger.bind(tag=TAG).info(f"识别文本: {raw_text}")
+            if speaker_name:
+                logger.bind(tag=TAG).info(f"识别说话人: {speaker_name}")
+            
+            # 性能监控
+            total_time = time.monotonic() - total_start_time
+            logger.bind(tag=TAG).info(f"总处理耗时: {total_time:.3f}s")
+            
+            # 检查文本长度
+            text_len, _ = remove_punctuation_and_length(raw_text)
+            self.stop_ws_connection()
+            
+            if text_len > 0:
+                # 构建包含说话人信息的JSON字符串
+                enhanced_text = self._build_enhanced_text(raw_text, speaker_name)
+                
+                # 使用自定义模块进行上报
+                await startToChat(conn, enhanced_text)
+                enqueue_asr_report(conn, enhanced_text, asr_audio_task)
+                
+        except Exception as e:
+            logger.bind(tag=TAG).error(f"处理语音停止失败: {e}")
+            import traceback
+            logger.bind(tag=TAG).debug(f"异常详情: {traceback.format_exc()}")
+
+    def _build_enhanced_text(self, text: str, speaker_name: Optional[str]) -> str:
+        """构建包含说话人信息的文本"""
+        if speaker_name:
+            return json.dumps({
+                "speaker": speaker_name,
+                "content": text
+            }, ensure_ascii=False)
+        else:
+            return text
+
+    def _pcm_to_wav(self, pcm_data: bytes) -> bytes:
+        """将PCM数据转换为WAV格式"""
+        if len(pcm_data) == 0:
+            logger.bind(tag=TAG).warning("PCM数据为空，无法转换WAV")
+            return b""
+        
+        # 确保数据长度是偶数（16位音频）
+        if len(pcm_data) % 2 != 0:
+            pcm_data = pcm_data[:-1]
+        
+        # 创建WAV文件头
+        wav_buffer = io.BytesIO()
+        try:
+            with wave.open(wav_buffer, 'wb') as wav_file:
+                wav_file.setnchannels(1)      # 单声道
+                wav_file.setsampwidth(2)      # 16位
+                wav_file.setframerate(16000)  # 16kHz采样率
+                wav_file.writeframes(pcm_data)
+            
+            wav_buffer.seek(0)
+            wav_data = wav_buffer.read()
+            
+            return wav_data
+        except Exception as e:
+            logger.bind(tag=TAG).error(f"WAV转换失败: {e}")
+            return b""
 
     def stop_ws_connection(self):
         pass
@@ -99,7 +365,7 @@ class ASRProviderBase(ABC):
 
         with wave.open(file_path, "wb") as wf:
             wf.setnchannels(1)
-            wf.setsampwidth(2)  # 2 bytes = 16-bit
+            wf.setsampwidth(2)
             wf.setframerate(16000)
             wf.writeframes(b"".join(pcm_data))
 
@@ -113,27 +379,29 @@ class ASRProviderBase(ABC):
         pass
 
     @staticmethod
-    def decode_opus(opus_data: List[bytes]) -> bytes:
+    def decode_opus(opus_data: List[bytes]) -> List[bytes]:
         """将Opus音频数据解码为PCM数据"""
         try:
-            decoder = opuslib_next.Decoder(16000, 1)  # 16kHz, 单声道
+            decoder = opuslib_next.Decoder(16000, 1)
             pcm_data = []
-            buffer_size = 960  # 每次处理960个采样点
+            buffer_size = 960  # 每次处理960个采样点 (60ms at 16kHz)
-
+            
-            for opus_packet in opus_data:
+            for i, opus_packet in enumerate(opus_data):
                 try:
-                    # 使用较小的缓冲区大小进行处理
+                    if not opus_packet or len(opus_packet) == 0:
+                        continue
+                    
                     pcm_frame = decoder.decode(opus_packet, buffer_size)
-                    if pcm_frame:
+                    if pcm_frame and len(pcm_frame) > 0:
                         pcm_data.append(pcm_frame)
+                        
                 except opuslib_next.OpusError as e:
-                    logger.bind(tag=TAG).warning(f"Opus解码错误，跳过当前数据包: {e}")
+                    logger.bind(tag=TAG).warning(f"Opus解码错误，跳过数据包 {i}: {e}")
-                    continue
                 except Exception as e:
-                    logger.bind(tag=TAG).error(f"音频处理错误: {e}", exc_info=True)
+                    logger.bind(tag=TAG).error(f"音频处理错误，数据包 {i}: {e}")
-                    continue
+            
-
             return pcm_data
+            
         except Exception as e:
-            logger.bind(tag=TAG).error(f"音频解码过程发生错误: {e}", exc_info=True)
+            logger.bind(tag=TAG).error(f"音频解码过程发生错误: {e}")
             return []

main/xiaozhi-server/core/utils/dialogue.py

@@ -74,8 +74,14 @@ class Dialogue:
         )
 
         if system_message:
+            # 构建增强的系统提示，包含说话人处理指导
+            speaker_guidance = "\n\n[说话人识别功能说明]\n" \
+                             "当用户消息包含 [说话人: 姓名] 前缀时，表示系统已识别出说话人身份。\n" \
+                             "请根据说话人的身份特征（如果之前有相关信息）来调整回应风格和内容。\n" \
+                             "你可以称呼说话人的名字，并参考他们的特点进行个性化回应。"
+            
             enhanced_system_prompt = (
-                f"{system_message.content}\n\n"
+                f"{system_message.content}{speaker_guidance}\n\n"
                 f"以下是用户的历史记忆：\n```\n{memory_str}\n```"
             )
             dialogue.append({"role": "system", "content": enhanced_system_prompt})

main/xiaozhi-server/core/utils/modules_initialize.py

@@ -125,4 +125,13 @@ def initialize_asr(config):
         config["ASR"][select_asr_module],
         str(config.get("delete_audio", True)).lower() in ("true", "1", "yes"),
     )
+    
+    # 初始化声纹识别功能
+    voiceprint_config = config.get("plugins", {}).get("voiceprint")
+    if voiceprint_config and voiceprint_config.get("url") and voiceprint_config.get("speakers"):
+        new_asr.init_voiceprint(voiceprint_config)
+        logger.bind(tag=TAG).info("ASR模块声纹识别功能已启用")
+    else:
+        logger.bind(tag=TAG).info("ASR模块声纹识别功能已禁用")
+    
     return new_asr