update: 恢复唤醒播放机制，待优化和改造（目前只有linkerai拥有正常保存唤醒音频to_tts）

main/xiaozhi-server/config.yaml

@@ -61,6 +61,10 @@ delete_audio: true
 close_connection_no_voice_time: 120
 # TTS请求超时时间(秒)
 tts_timeout: 10
+# 开启唤醒词加速
+enable_wakeup_words_response_cache: true
+# 开场是否回复唤醒词
+enable_greeting: true
 # 说完话是否开启提示音
 enable_stop_tts_notify: false
 # 说完话是否开启提示音，音效地址

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

@@ -1,5 +1,16 @@
+import time
 import json
+import random
 import asyncio
+from core.utils.dialogue import Message
+from core.providers.tts.dto.dto import SentenceType
+from core.utils.wakeup_word import WakeupWordsConfig
+from core.handle.sendAudioHandle import sendAudioMessage, send_stt_message
+from core.utils.util import (
+    audio_to_data_stream,
+    remove_punctuation_and_length, 
+    opus_datas_to_wav_bytes
+)
 from core.providers.tools.device_mcp import (
     MCPClient,
     send_mcp_initialize_message,
@@ -8,6 +19,17 @@ from core.providers.tools.device_mcp import (
 
 TAG = __name__
 
+WAKEUP_CONFIG = {
+    "refresh_time": 5,
+    "words": ["你好", "你好啊", "嘿，你好", "嗨"],
+}
+
+# 创建全局的唤醒词配置管理器
+wakeup_words_config = WakeupWordsConfig()
+
+# 用于防止并发调用wakeupWordsResponse的锁
+_wakeup_response_lock = asyncio.Lock()
+
 
 async def handleHelloMessage(conn, msg_json):
     """处理hello消息"""
@@ -30,3 +52,99 @@ async def handleHelloMessage(conn, msg_json):
             asyncio.create_task(send_mcp_tools_list_request(conn))
 
     await conn.websocket.send(json.dumps(conn.welcome_msg))
+
+
+async def checkWakeupWords(conn, text):
+    enable_wakeup_words_response_cache = conn.config[
+        "enable_wakeup_words_response_cache"
+    ]
+
+    if not enable_wakeup_words_response_cache or not conn.tts:
+        return False
+
+    _, filtered_text = remove_punctuation_and_length(text)
+    if filtered_text not in conn.config.get("wakeup_words"):
+        return False
+
+    conn.just_woken_up = True
+    await send_stt_message(conn, text)
+
+    # 获取当前音色
+    voice = getattr(conn.tts, "voice", "default")
+    if not voice:
+        voice = "default"
+
+    # 获取唤醒词回复配置
+    response = wakeup_words_config.get_wakeup_response(voice)
+    if not response or not response.get("file_path"):
+        response = {
+            "voice": "default",
+            "file_path": "config/assets/wakeup_words.wav",
+            "time": 0,
+            "text": "哈啰啊，我是小智啦，声音好听的台湾女孩一枚，超开心认识你耶，最近在忙啥，别忘了给我来点有趣的料哦，我超爱听八卦的啦",
+        }
+    
+    # 获取音频数据
+    opus_packets = []
+    def handle_audio_frame(frame_data):
+        opus_packets.append(frame_data)
+    
+    audio_to_data_stream(response.get("file_path"), is_opus=True, callback=handle_audio_frame)
+
+    # 播放唤醒词回复
+    conn.client_abort = False
+
+    conn.logger.bind(tag=TAG).info(f"播放唤醒词回复: {response.get('text')}")
+    await sendAudioMessage(conn, SentenceType.FIRST, opus_packets, response.get("text"))
+    await sendAudioMessage(conn, SentenceType.LAST, [], None)
+
+    # 补充对话
+    conn.dialogue.put(Message(role="assistant", content=response.get("text")))
+
+    # 检查是否需要更新唤醒词回复
+    if time.time() - response.get("time", 0) > WAKEUP_CONFIG["refresh_time"]:
+        if not _wakeup_response_lock.locked():
+            asyncio.create_task(wakeupWordsResponse(conn))
+    return True
+
+
+async def wakeupWordsResponse(conn):
+    if not conn.tts or not conn.llm or not conn.llm.response_no_stream:
+        return
+
+    try:
+        # 尝试获取锁，如果获取不到就返回
+        if not await _wakeup_response_lock.acquire():
+            return
+
+        # 生成唤醒词回复
+        wakeup_word = random.choice(WAKEUP_CONFIG["words"])
+        question = (
+            "此刻用户正在和你说```"
+            + wakeup_word
+            + "```。\n请你根据以上用户的内容进行20-30字回复。要符合系统设置的角色情感和态度，不要像机器人一样说话。\n"
+            + "请勿对这条内容本身进行任何解释和回应，请勿返回表情符号，仅返回对用户的内容的回复。"
+        )
+
+        result = conn.llm.response_no_stream(conn.config["prompt"], question)
+        if not result or len(result) == 0:
+            return
+
+        # 生成TTS音频
+        tts_result = await asyncio.to_thread(conn.tts.to_tts, result)
+        if not tts_result:
+            return
+
+        # 获取当前音色
+        voice = getattr(conn.tts, "voice", "default")
+
+        wav_bytes = opus_datas_to_wav_bytes(tts_result, sample_rate=16000)
+        file_path = wakeup_words_config.generate_file_path(voice)
+        with open(file_path, "wb") as f:
+            f.write(wav_bytes)
+        # 更新配置
+        wakeup_words_config.update_wakeup_response(voice, file_path, result)
+    finally:
+        # 确保在任何情况下都释放锁
+        if _wakeup_response_lock.locked():
+            _wakeup_response_lock.release()

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

@@ -2,6 +2,7 @@ import json
 import asyncio
 import uuid
 from core.handle.sendAudioHandle import send_stt_message
+from core.handle.helloHandle import checkWakeupWords
 from core.utils.util import remove_punctuation_and_length
 from core.providers.tts.dto.dto import ContentType
 from core.utils.dialogue import Message
@@ -26,6 +27,10 @@ async def handle_user_intent(conn, text):
     filtered_text = remove_punctuation_and_length(text)[1]
     if await check_direct_exit(conn, filtered_text):
         return True
+    
+    # 检查是否是唤醒词
+    if await checkWakeupWords(conn, filtered_text):
+        return True
 
     if conn.intent_type == "function_call":
         # 使用支持function calling的聊天方法,不再进行意图分析

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

@@ -1,11 +1,11 @@
 import time
 import json
-from core.handle.sendAudioHandle import send_stt_message
+import asyncio
+from core.utils.util import audio_to_data_stream
+from core.handle.abortHandle import handleAbortMessage
 from core.handle.intentHandler import handle_user_intent
 from core.utils.output_counter import check_device_output_limit
-from core.handle.abortHandle import handleAbortMessage
-from core.handle.sendAudioHandle import SentenceType
-from core.utils.util import audio_to_data_stream
+from core.handle.sendAudioHandle import send_stt_message, SentenceType
 
 TAG = __name__
 
@@ -13,7 +13,14 @@ TAG = __name__
 async def handleAudioMessage(conn, audio):
     # 当前片段是否有人说话
     have_voice = conn.vad.is_vad(conn, audio)
-
+    # 如果设备刚刚被唤醒，短暂忽略VAD检测
+    if have_voice and hasattr(conn, "just_woken_up") and conn.just_woken_up:
+        have_voice = False
+        # 设置一个短暂延迟后恢复VAD检测
+        conn.asr_audio.clear()
+        if not hasattr(conn, "vad_resume_task") or conn.vad_resume_task.done():
+            conn.vad_resume_task = asyncio.create_task(resume_vad_detection(conn))
+        return
     if have_voice:
         if conn.client_is_speaking:
             await handleAbortMessage(conn)
@@ -22,6 +29,13 @@ async def handleAudioMessage(conn, audio):
     # 接收音频
     await conn.asr.receive_audio(conn, audio, have_voice)
 
+
+async def resume_vad_detection(conn):
+    # 等待2秒后恢复VAD检测
+    await asyncio.sleep(1)
+    conn.just_woken_up = False
+
+
 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
@@ -146,7 +160,6 @@ async def check_bind_device(conn):
         play_audio_frames(conn, music_path)
         conn.tts.tts_audio_queue.put((SentenceType.LAST, [], None))
 
-
 def play_audio_frames(conn, file_path):
     """播放音频文件并处理发送帧数据"""
     def handle_audio_frame(frame_data):

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

@@ -1,8 +1,8 @@
 import json
-import asyncio
 import time
-from core.providers.tts.dto.dto import SentenceType
+import asyncio
 from core.utils import textUtils
+from core.providers.tts.dto.dto import SentenceType
 
 TAG = __name__
 
@@ -30,7 +30,7 @@ async def sendAudioMessage(conn, sentenceType, audios, text):
 
 
 # 播放音频
-async def sendAudio(conn, audios, pre_buffer=False):
+async def sendAudio(conn, audios, pre_buffer=False, frame_duration=60):
     """
     发送单个opus包，支持流控
     Args:
@@ -50,12 +50,8 @@ async def sendAudio(conn, audios, pre_buffer=False):
             await conn.websocket.send(audios)
             return
 
-        # 重置没有声音的状态
         conn.last_activity_time = time.time() * 1000
 
-        # 流控逻辑：确保按60ms的帧时长间隔发送
-        frame_duration = 60  # 毫秒
-
         # 获取或初始化流控状态
         if not hasattr(conn, "audio_flow_control"):
             conn.audio_flow_control = {
@@ -66,13 +62,10 @@ async def sendAudio(conn, audios, pre_buffer=False):
 
         flow_control = conn.audio_flow_control
         current_time = time.perf_counter()
-
-        # 计算期望的发送时间
+        # 计算预期发送时间
         expected_time = flow_control["start_time"] + (
             flow_control["packet_count"] * frame_duration / 1000
         )
-
-        # 流控延迟
         delay = expected_time - current_time
         if delay > 0:
             await asyncio.sleep(delay)
@@ -83,6 +76,38 @@ async def sendAudio(conn, audios, pre_buffer=False):
         # 更新流控状态
         flow_control["packet_count"] += 1
         flow_control["last_send_time"] = time.perf_counter()
+    else:
+        if audios is None or len(audios) == 0:
+            return
+        # 流控参数优化
+        frame_duration = 60  # 帧时长（毫秒），匹配 Opus 编码
+        start_time = time.perf_counter()
+        play_position = 0
+
+        # 执行预缓冲
+        pre_buffer_frames = min(3, len(audios))
+        for i in range(pre_buffer_frames):
+            await conn.websocket.send(audios[i])
+        remaining_audios = audios[pre_buffer_frames:]
+
+        # 播放剩余音频帧
+        for opus_packet in remaining_audios:
+            if conn.client_abort:
+                break
+
+            # 重置没有声音的状态
+            conn.last_activity_time = time.time() * 1000
+
+            # 计算预期发送时间
+            expected_time = start_time + (play_position / 1000)
+            current_time = time.perf_counter()
+            delay = expected_time - current_time
+            if delay > 0:
+                await asyncio.sleep(delay)
+
+            await conn.websocket.send(opus_packet)
+
+            play_position += frame_duration
 
 
 async def send_tts_message(conn, state, text=None):

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

@@ -1,13 +1,14 @@
 import json
 import time
+import asyncio
 from core.handle.abortHandle import handleAbortMessage
 from core.handle.helloHandle import handleHelloMessage
-from core.providers.tools.device_mcp import handle_mcp_message
-from core.utils.util import remove_punctuation_and_length, filter_sensitive_info
-from core.handle.receiveAudioHandle import startToChat, handleAudioMessage
-from core.providers.tools.device_iot import handleIotDescriptors, handleIotStatus
 from core.handle.reportHandle import enqueue_asr_report
-import asyncio
+from core.providers.tools.device_mcp import handle_mcp_message
+from core.handle.receiveAudioHandle import startToChat, handleAudioMessage
+from core.handle.sendAudioHandle import send_stt_message, send_tts_message
+from core.utils.util import remove_punctuation_and_length, filter_sensitive_info
+from core.providers.tools.device_iot import handleIotDescriptors, handleIotStatus
 
 TAG = __name__
 
@@ -50,9 +51,23 @@ async def handleTextMessage(conn, message):
                     filtered_len, filtered_text = remove_punctuation_and_length(
                         original_text
                     )
+
                     # 识别是否是唤醒词
                     is_wakeup_words = filtered_text in conn.config.get("wakeup_words")
-                    if not is_wakeup_words:
+                    # 是否开启唤醒词回复
+                    enable_greeting = conn.config.get("enable_greeting", True)
+
+                    if is_wakeup_words and not enable_greeting:
+                        # 如果是唤醒词，且关闭了唤醒词回复，就不用回答
+                        await send_stt_message(conn, original_text)
+                        await send_tts_message(conn, "stop", None)
+                        conn.client_is_speaking = False
+                    elif is_wakeup_words:
+                        conn.just_woken_up = True
+                        # 上报纯文字数据（复用ASR上报功能，但不提供音频数据）
+                        enqueue_asr_report(conn, "嘿，你好呀", [])
+                        await startToChat(conn, "嘿，你好呀")
+                    else:
                         # 上报纯文字数据（复用ASR上报功能，但不提供音频数据）
                         enqueue_asr_report(conn, original_text, [])
                         # 否则需要LLM对文字内容进行答复

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

@@ -286,8 +286,8 @@ class TTSProviderBase(ABC):
                     enqueue_audio = []
                     enqueue_text = text
 
-                # 计算音频数据的帧数
-                if isinstance(audio_datas, bytes):
+                # 收集上报音频数据
+                if isinstance(audio_datas, bytes) and enqueue_audio is not None:
                     enqueue_audio.append(audio_datas)
 
                 # 发送音频

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

@@ -1,8 +1,10 @@
 import os
+import time
 import queue
-import asyncio
-import traceback
 import aiohttp
+import asyncio
+import requests
+import traceback
 from config.logger import setup_logging
 from core.utils.tts import MarkdownCleaner
 from core.providers.tts.base import TTSProviderBase
@@ -199,3 +201,71 @@ class TTSProvider(TTSProviderBase):
         except Exception as e:
             logger.bind(tag=TAG).error(f"TTS请求异常: {e}")
             self.tts_audio_queue.put((SentenceType.LAST, [], None))
+
+    def to_tts(self, text: str) -> list:
+        """非流式TTS处理，用于测试及保存音频文件的场景
+        Args:
+            text: 要转换的文本
+        Returns:
+            list: 返回opus编码后的音频数据列表
+        """
+        start_time = time.time()
+        text = MarkdownCleaner.clean_markdown(text)
+
+        params = {
+            "tts_text": text,
+            "spk_id": self.voice,
+            "frame_duration": 60,
+            "stream": False,
+            "target_sr": 16000,
+            "audio_format": self.audio_format,
+            "instruct_text": "请生成一段自然流畅的语音",
+        }
+        headers = {
+            "Authorization": f"Bearer {self.access_token}",
+            "Content-Type": "application/json",
+        }
+
+        try:
+            with requests.get(
+                self.api_url, params=params, headers=headers, timeout=5
+            ) as response:
+                if response.status_code != 200:
+                    logger.bind(tag=TAG).error(
+                        f"TTS请求失败: {response.status_code}, {response.text}"
+                    )
+                    return []
+
+                logger.info(f"TTS请求成功: {text}, 耗时: {time.time() - start_time}秒")
+
+                # 使用opus编码器处理PCM数据
+                opus_datas = []
+                pcm_data = response.content
+
+                # 计算每帧的字节数
+                frame_bytes = int(
+                    self.opus_encoder.sample_rate
+                    * self.opus_encoder.channels
+                    * self.opus_encoder.frame_size_ms
+                    / 1000
+                    * 2
+                )
+
+                # 分帧处理PCM数据
+                for i in range(0, len(pcm_data), frame_bytes):
+                    frame = pcm_data[i : i + frame_bytes]
+                    if len(frame) < frame_bytes:
+                        # 最后一帧可能不足，用0填充
+                        frame = frame + b"\x00" * (frame_bytes - len(frame))
+
+                    self.opus_encoder.encode_pcm_to_opus_stream(
+                        frame,
+                        end_of_stream=(i + frame_bytes >= len(pcm_data)),
+                        callback=lambda opus: opus_datas.append(opus)
+                    )
+
+                return opus_datas
+
+        except Exception as e:
+            logger.bind(tag=TAG).error(f"TTS请求异常: {e}")
+            return []

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

@@ -1,16 +1,17 @@
-import json
-import socket
-import subprocess
 import re
 import os
-from io import BytesIO
-from typing import Callable, Any
-from core.utils import p3
-import numpy as np
-import requests
-import opuslib_next
-from pydub import AudioSegment
+import json
+import wave
 import copy
+import socket
+import requests
+import subprocess
+import numpy as np
+import opuslib_next
+from io import BytesIO
+from core.utils import p3
+from pydub import AudioSegment
+from typing import Callable, Any
 
 TAG = __name__
 emoji_map = {
@@ -274,6 +275,33 @@ def pcm_to_data_stream(raw_data, is_opus=True, callback: Callable[[Any], Any] =
             callback(frame_data)
 
 
+def opus_datas_to_wav_bytes(opus_datas, sample_rate=16000, channels=1):
+    """
+    将opus帧列表解码为wav字节流
+    """
+    decoder = opuslib_next.Decoder(sample_rate, channels)
+    pcm_datas = []
+
+    frame_duration = 60  # ms
+    frame_size = int(sample_rate * frame_duration / 1000)  # 960
+
+    for opus_frame in opus_datas:
+        # 解码为PCM（返回bytes，2字节/采样点）
+        pcm = decoder.decode(opus_frame, frame_size)
+        pcm_datas.append(pcm)
+
+    pcm_bytes = b"".join(pcm_datas)
+
+    # 写入wav字节流
+    wav_buffer = BytesIO()
+    with wave.open(wav_buffer, "wb") as wf:
+        wf.setnchannels(channels)
+        wf.setsampwidth(2)  # 16bit
+        wf.setframerate(sample_rate)
+        wf.writeframes(pcm_bytes)
+    return wav_buffer.getvalue()
+
+
 def check_vad_update(before_config, new_config):
     if (
         new_config.get("selected_module") is None

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

@@ -0,0 +1,140 @@
+import os
+import re
+import yaml
+import time
+import hashlib
+import portalocker
+from typing import Dict
+
+
+class FileLock:
+    def __init__(self, file, timeout=5):
+        self.file = file
+        self.timeout = timeout
+        self.start_time = None
+
+    def __enter__(self):
+        self.start_time = time.time()
+        while True:
+            try:
+                portalocker.lock(self.file, portalocker.LOCK_EX | portalocker.LOCK_NB)
+                return self.file
+            except portalocker.LockException:
+                if time.time() - self.start_time > self.timeout:
+                    raise TimeoutError("获取文件锁超时")
+                time.sleep(0.1)
+
+    def __exit__(self, exc_type, exc_val, exc_tb):
+        portalocker.unlock(self.file)
+
+
+class WakeupWordsConfig:
+    def __init__(self):
+        self.config_file = "data/.wakeup_words.yaml"
+        self.assets_dir = "config/assets/wakeup_words"
+        self._ensure_directories()
+        self._config_cache = None
+        self._last_load_time = 0
+        self._cache_ttl = 1  # 缓存有效期（秒）
+        self._lock_timeout = 5  # 文件锁超时时间（秒）
+
+    def _ensure_directories(self):
+        """确保必要的目录存在"""
+        os.makedirs(os.path.dirname(self.config_file), exist_ok=True)
+        os.makedirs(self.assets_dir, exist_ok=True)
+
+    def _load_config(self) -> Dict:
+        """加载配置文件，使用缓存机制"""
+        current_time = time.time()
+
+        # 如果缓存有效，直接返回缓存
+        if (
+            self._config_cache is not None
+            and current_time - self._last_load_time < self._cache_ttl
+        ):
+            return self._config_cache
+
+        try:
+            with open(self.config_file, "a+") as f:
+                with FileLock(f, timeout=self._lock_timeout):
+                    f.seek(0)
+                    content = f.read()
+                    config = yaml.safe_load(content) if content else {}
+                    self._config_cache = config
+                    self._last_load_time = current_time
+                    return config
+        except (TimeoutError, IOError) as e:
+            print(f"加载配置文件失败: {e}")
+            return {}
+        except Exception as e:
+            print(f"加载配置文件时发生未知错误: {e}")
+            return {}
+
+    def _save_config(self, config: Dict):
+        """保存配置到文件，使用文件锁保护"""
+        try:
+            with open(self.config_file, "w") as f:
+                with FileLock(f, timeout=self._lock_timeout):
+                    yaml.dump(config, f, allow_unicode=True)
+                    self._config_cache = config
+                    self._last_load_time = time.time()
+        except (TimeoutError, IOError) as e:
+            print(f"保存配置文件失败: {e}")
+            raise
+        except Exception as e:
+            print(f"保存配置文件时发生未知错误: {e}")
+            raise
+
+    def get_wakeup_response(self, voice: str) -> Dict:
+        voice = hashlib.md5(voice.encode()).hexdigest()
+        """获取唤醒词回复配置"""
+        config = self._load_config()
+
+        if not config or voice not in config:
+            return None
+
+        # 检查文件大小
+        file_path = config[voice]["file_path"]
+        if not os.path.exists(file_path) or os.stat(file_path).st_size < (15 * 1024):
+            return None
+
+        return config[voice]
+
+    def update_wakeup_response(self, voice: str, file_path: str, text: str):
+        """更新唤醒词回复配置"""
+        try:
+            # 过滤表情符号
+            filtered_text = re.sub(r'[\U0001F600-\U0001F64F\U0001F900-\U0001F9FF]', '', text)
+
+            config = self._load_config()
+            voice_hash = hashlib.md5(voice.encode()).hexdigest()
+            config[voice_hash] = {
+                "voice": voice,
+                "file_path": file_path,
+                "time": time.time(),
+                "text": filtered_text,
+            }
+            self._save_config(config)
+        except Exception as e:
+            print(f"更新唤醒词回复配置失败: {e}")
+            raise
+
+    def generate_file_path(self, voice: str) -> str:
+        """生成音频文件路径，使用voice的哈希值作为文件名"""
+        try:
+            # 生成voice的哈希值
+            voice_hash = hashlib.md5(voice.encode()).hexdigest()
+            file_path = os.path.join(self.assets_dir, f"{voice_hash}.wav")
+
+            # 如果文件已存在，先删除
+            if os.path.exists(file_path):
+                try:
+                    os.remove(file_path)
+                except Exception as e:
+                    print(f"删除已存在的音频文件失败: {e}")
+                    raise
+
+            return file_path
+        except Exception as e:
+            print(f"生成音频文件路径失败: {e}")
+            raise