Merge pull request #2158 from xinnan-tech/future

Future

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

@@ -4,15 +4,15 @@ import random
 import asyncio
 from core.utils.dialogue import Message
 from core.utils.util import audio_to_data
+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 remove_punctuation_and_length, opus_datas_to_wav_bytes
-from core.providers.tts.dto.dto import ContentType, SentenceType
 from core.providers.tools.device_mcp import (
     MCPClient,
     send_mcp_initialize_message,
     send_mcp_tools_list_request,
 )
-from core.utils.wakeup_word import WakeupWordsConfig
 
 TAG = __name__
 
@@ -56,7 +56,16 @@ async def checkWakeupWords(conn, text):
         "enable_wakeup_words_response_cache"
     ]
 
-    if not enable_wakeup_words_response_cache or not conn.tts:
+    # 等待tts初始化，最多等待3秒
+    start_time = time.time()
+    while time.time() - start_time < 3:
+        if conn.tts:
+            break
+        await asyncio.sleep(0.1)
+    else:
+        return False
+
+    if not enable_wakeup_words_response_cache:
         return False
 
     _, filtered_text = remove_punctuation_and_length(text)
@@ -81,9 +90,10 @@ async def checkWakeupWords(conn, text):
             "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 = audio_to_data(response.get("file_path"))
     # 播放唤醒词回复
     conn.client_abort = False
-    opus_packets, _ = audio_to_data(response.get("file_path"))
 
     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"))

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

@@ -1,12 +1,12 @@
 import json
-import asyncio
 import uuid
-from core.handle.sendAudioHandle import send_stt_message
+import asyncio
-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
+from core.providers.tts.dto.dto import ContentType
+from core.handle.helloHandle import checkWakeupWords
 from plugins_func.register import Action, ActionResponse
+from core.handle.sendAudioHandle import send_stt_message
+from core.utils.util import remove_punctuation_and_length
 from core.providers.tts.dto.dto import TTSMessageDTO, SentenceType
 
 TAG = __name__
@@ -24,9 +24,10 @@ async def handle_user_intent(conn, text):
         pass
 
     # 检查是否有明确的退出命令
-    filtered_text = remove_punctuation_and_length(text)[1]
+    _, filtered_text = remove_punctuation_and_length(text)
     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

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

@@ -1,12 +1,11 @@
-from core.handle.sendAudioHandle import send_stt_message
+import time
+import json
+import asyncio
+from core.utils.util import audio_to_data
+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 send_stt_message, SentenceType
-import time
-import asyncio
-import json
-from core.handle.sendAudioHandle import SentenceType
-from core.utils.util import audio_to_data
 
 TAG = __name__
 
@@ -22,7 +21,6 @@ async def handleAudioMessage(conn, audio):
         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)
@@ -31,13 +29,11 @@ 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
@@ -118,10 +114,12 @@ async def no_voice_close_connect(conn, have_voice):
 
 
 async def max_out_size(conn):
+    # 播放超出最大输出字数的提示
+    conn.client_abort = False
     text = "不好意思，我现在有点事情要忙，明天这个时候我们再聊，约好了哦！明天不见不散，拜拜！"
     await send_stt_message(conn, text)
     file_path = "config/assets/max_output_size.wav"
-    opus_packets, _ = audio_to_data(file_path)
+    opus_packets = audio_to_data(file_path)
     conn.tts.tts_audio_queue.put((SentenceType.LAST, opus_packets, text))
     conn.close_after_chat = True
 
@@ -140,7 +138,7 @@ async def check_bind_device(conn):
 
         # 播放提示音
         music_path = "config/assets/bind_code.wav"
-        opus_packets, _ = audio_to_data(music_path)
+        opus_packets = audio_to_data(music_path)
         conn.tts.tts_audio_queue.put((SentenceType.FIRST, opus_packets, text))
 
         # 逐个播放数字
@@ -148,15 +146,17 @@ async def check_bind_device(conn):
             try:
                 digit = conn.bind_code[i]
                 num_path = f"config/assets/bind_code/{digit}.wav"
-                num_packets, _ = audio_to_data(num_path)
+                num_packets = audio_to_data(num_path)
                 conn.tts.tts_audio_queue.put((SentenceType.MIDDLE, num_packets, None))
             except Exception as e:
                 conn.logger.bind(tag=TAG).error(f"播放数字音频失败: {e}")
                 continue
         conn.tts.tts_audio_queue.put((SentenceType.LAST, [], None))
     else:
+        # 播放未绑定提示
+        conn.client_abort = False
         text = f"没有找到该设备的版本信息，请正确配置 OTA地址，然后重新编译固件。"
         await send_stt_message(conn, text)
         music_path = "config/assets/bind_not_found.wav"
-        opus_packets, _ = audio_to_data(music_path)
+        opus_packets = audio_to_data(music_path)
         conn.tts.tts_audio_queue.put((SentenceType.LAST, opus_packets, text))

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

@@ -1,25 +1,26 @@
 import json
-import asyncio
 import time
-from core.providers.tts.dto.dto import SentenceType
+import asyncio
 from core.utils import textUtils
+from core.utils.util import audio_to_data
+from core.providers.tts.dto.dto import SentenceType
 
 TAG = __name__
 
 
 async def sendAudioMessage(conn, sentenceType, audios, text):
-    # 发送句子开始消息
-    conn.logger.bind(tag=TAG).info(f"发送音频消息: {sentenceType}, {text}")
-
-    pre_buffer = False
     if conn.tts.tts_audio_first_sentence:
         conn.logger.bind(tag=TAG).info(f"发送第一段语音: {text}")
         conn.tts.tts_audio_first_sentence = False
-        pre_buffer = True
+        await send_tts_message(conn, "start", None)
 
-    await send_tts_message(conn, "sentence_start", text)
+    if sentenceType == SentenceType.FIRST:
+        await send_tts_message(conn, "sentence_start", text)
 
-    await sendAudio(conn, audios, pre_buffer)
+    await sendAudio(conn, audios)
+    # 发送句子开始消息
+    if sentenceType is not SentenceType.MIDDLE:
+        conn.logger.bind(tag=TAG).info(f"发送音频消息: {sentenceType}, {text}")
 
     # 发送结束消息（如果是最后一个文本）
     if conn.llm_finish_task and sentenceType == SentenceType.LAST:
@@ -30,45 +31,83 @@ async def sendAudioMessage(conn, sentenceType, audios, text):
 
 
 # 播放音频
-async def sendAudio(conn, audios, pre_buffer=True):
+async def sendAudio(conn, audios, frame_duration=60):
+    """
+    发送单个opus包，支持流控
+    Args:
+        conn: 连接对象
+        opus_packet: 单个opus数据包
+        pre_buffer: 快速发送音频
+        frame_duration: 帧时长（毫秒），匹配 Opus 编码
+    """
     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
 
-    # 仅当第一句话时执行预缓冲
+    if isinstance(audios, bytes):
-    if pre_buffer:
-        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:]
-    else:
-        remaining_audios = audios
-
-    # 播放剩余音频帧
-    for opus_packet in remaining_audios:
         if conn.client_abort:
-            break
+            return
 
-        # 重置没有声音的状态
         conn.last_activity_time = time.time() * 1000
 
-        # 计算预期发送时间
+        # 获取或初始化流控状态
-        expected_time = start_time + (play_position / 1000)
+        if not hasattr(conn, "audio_flow_control"):
+            conn.audio_flow_control = {
+                "last_send_time": 0,
+                "packet_count": 0,
+                "start_time": time.perf_counter(),
+            }
+
+        flow_control = conn.audio_flow_control
         current_time = time.perf_counter()
+        # 计算预期发送时间
+        expected_time = flow_control["start_time"] + (
+            flow_control["packet_count"] * frame_duration / 1000
+        )
         delay = expected_time - current_time
         if delay > 0:
             await asyncio.sleep(delay)
 
-        await conn.websocket.send(opus_packet)
+        # 发送数据包
+        await conn.websocket.send(audios)
 
-        play_position += frame_duration
+        # 更新流控状态
+        flow_control["packet_count"] += 1
+        flow_control["last_send_time"] = time.perf_counter()
+    else:
+        # 文件型音频走普通播放
+        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):
     """发送 TTS 状态消息"""
+    if text is None and state == "sentence_start":
+        return
     message = {"type": "tts", "state": state, "session_id": conn.session_id}
     if text is not None:
         message["text"] = textUtils.check_emoji(text)
@@ -81,7 +120,7 @@ async def send_tts_message(conn, state, text=None):
             stop_tts_notify_voice = conn.config.get(
                 "stop_tts_notify_voice", "config/assets/tts_notify.mp3"
             )
-            audios, _ = conn.tts.audio_to_opus_data(stop_tts_notify_voice)
+            audios = audio_to_data(stop_tts_notify_voice, is_opus=True)
             await sendAudio(conn, audios)
         # 清除服务端讲话状态
         conn.clearSpeakStatus()
@@ -91,18 +130,17 @@ async def send_tts_message(conn, state, text=None):
 
 
 async def send_stt_message(conn, text):
+    """发送 STT 状态消息"""
     end_prompt_str = conn.config.get("end_prompt", {}).get("prompt")
     if end_prompt_str and end_prompt_str == text:
         await send_tts_message(conn, "start")
         return
 
-    """发送 STT 状态消息"""
-    
     # 解析JSON格式，提取实际的用户说话内容
     display_text = text
     try:
         # 尝试解析JSON格式
-        if text.strip().startswith('{') and text.strip().endswith('}'):
+        if text.strip().startswith("{") and text.strip().endswith("}"):
             parsed_data = json.loads(text)
             if isinstance(parsed_data, dict) and "content" in parsed_data:
                 # 如果是包含说话人信息的JSON格式，只显示content部分

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

@@ -1,18 +1,18 @@
 import os
+import io
 import wave
 import uuid
+import json
+import time
 import queue
 import asyncio
 import traceback
 import threading
 import opuslib_next
-import json
-import io
-import time
 import concurrent.futures
 from abc import ABC, abstractmethod
 from config.logger import setup_logging
-from typing import Optional, Tuple, List, Dict, Any
+from typing import Optional, Tuple, List
 from core.handle.receiveAudioHandle import startToChat
 from core.handle.reportHandle import enqueue_asr_report
 from core.utils.util import remove_punctuation_and_length
@@ -87,11 +87,9 @@ class ASRProviderBase(ABC):
             
             # 预先准备WAV数据
             wav_data = None
-            # 使用连接的声纹识别提供者
             if conn.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()
@@ -132,8 +130,6 @@ class ASRProviderBase(ABC):
                     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)
                 
@@ -151,7 +147,7 @@ class ASRProviderBase(ABC):
             
             
             # 处理结果
-            raw_text, file_path = results.get("asr", ("", None))
+            raw_text, _ = results.get("asr", ("", None))
             speaker_name = results.get("voiceprint", None)
             
             # 记录识别结果

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

@@ -268,11 +268,7 @@ class TTSProvider(TTSProviderBase):
                     )
                     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._process_audio_file_stream(message.content_file, callback=lambda audio_data: self.handle_audio_file(audio_data, message.content_detail))
-                        self.before_stop_play_files.append(
-                            (file_audio, message.content_detail)
-                        )
-
                 if message.sentence_type == SentenceType.LAST:
                     try:
                         logger.bind(tag=TAG).info("开始结束TTS会话...")
@@ -422,9 +418,6 @@ class TTSProvider(TTSProviderBase):
 
     async def _start_monitor_tts_response(self):
         """监听TTS响应"""
-        opus_datas_cache = []
-        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():
@@ -445,28 +438,16 @@ class TTSProvider(TTSProviderBase):
                                 self.tts_audio_queue.put(
                                     (SentenceType.FIRST, [], None)
                                 )
-                            elif event_name == "SentenceBegin":
-                                opus_datas_cache = []
                             elif event_name == "SentenceEnd":
-                                if (
+                                # 发送缓存的数据
-                                    not is_first_sentence
+                                if self.conn.tts_MessageText:
-                                    or first_sentence_segment_count > 10
+                                    logger.bind(tag=TAG).info(
-                                ):
+                                        f"句子语音生成成功： {self.conn.tts_MessageText}"
-                                    # 发送缓存的数据
+                                    )
-                                    if self.conn.tts_MessageText:
+                                    self.tts_audio_queue.put(
-                                        logger.bind(tag=TAG).info(
+                                        (SentenceType.FIRST, [], self.conn.tts_MessageText)
-                                            f"句子语音生成成功： {self.conn.tts_MessageText}"
+                                    )
-                                        )
+                                    self.conn.tts_MessageText = None
-                                        self.tts_audio_queue.put(
-                                            (SentenceType.MIDDLE, opus_datas_cache, self.conn.tts_MessageText)
-                                        )
-                                        self.conn.tts_MessageText = None
-                                    else:
-                                        self.tts_audio_queue.put(
-                                            (SentenceType.MIDDLE, opus_datas_cache, None)
-                                        )
-                                # 第一句话结束后，将标志设置为False
-                                is_first_sentence = False
                             elif event_name == "SynthesisCompleted":
                                 logger.bind(tag=TAG).debug(f"会话结束～～")
                                 self._process_before_stop_play_files()
@@ -477,22 +458,7 @@ class TTSProvider(TTSProviderBase):
                     # 二进制消息（音频数据）
                     elif isinstance(msg, (bytes, bytearray)):
                         logger.bind(tag=TAG).debug(f"推送数据到队列里面～～")
-                        opus_datas = self.opus_encoder.encode_pcm_to_opus(msg, False)
+                        self.opus_encoder.encode_pcm_to_opus_stream(msg, False, self.handle_opus)
-                        logger.bind(tag=TAG).debug(
-                            f"推送数据到队列里面帧数～～{len(opus_datas)}"
-                        )
-                        if is_first_sentence:
-                            first_sentence_segment_count += 1
-                            if first_sentence_segment_count <= 6:
-                                self.tts_audio_queue.put(
-                                    (SentenceType.MIDDLE, opus_datas, None)
-                                )
-                            else:
-                                opus_datas_cache.extend(opus_datas)
-                        else:
-                            # 后续句子缓存
-                            opus_datas_cache.extend(opus_datas)
-
                 except websockets.ConnectionClosed:
                     logger.bind(tag=TAG).warning("WebSocket连接已关闭")
                     break
@@ -615,11 +581,11 @@ class TTSProvider(TTSProviderBase):
                     while not synthesis_completed:
                         msg = await ws.recv()
                         if isinstance(msg, (bytes, bytearray)):
-                            # 编码为Opus并收集
+                            self.opus_encoder.encode_pcm_to_opus_stream(
-                            opus_frames = self.opus_encoder.encode_pcm_to_opus(
+                                msg,
-                                msg, False
+                                end_of_stream=False,
+                                callback=lambda opus: audio_data.append(opus)
                             )
-                            audio_data.extend(opus_frames)
                         elif isinstance(msg, str):
                             data = json.loads(msg)
                             header = data.get("header", {})

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

@@ -1,19 +1,22 @@
 import os
 import re
-import queue
+import time
 import uuid
+import queue
 import asyncio
 import threading
+import traceback
 from core.utils import p3
 from datetime import datetime
 from core.utils import textUtils
+from typing import Callable, Any
 from abc import ABC, abstractmethod
 from config.logger import setup_logging
-from core.utils.util import audio_to_data, audio_bytes_to_data
 from core.utils.tts import MarkdownCleaner
 from core.utils.output_counter import add_device_output
 from core.handle.reportHandle import enqueue_tts_report
 from core.handle.sendAudioHandle import sendAudioMessage
+from core.utils.util import audio_bytes_to_data_stream, audio_to_data_stream
 from core.providers.tts.dto.dto import (
     TTSMessageDTO,
     SentenceType,
@@ -21,8 +24,6 @@ from core.providers.tts.dto.dto import (
     InterfaceType,
 )
 
-import traceback
-
 TAG = __name__
 logger = setup_logging()
 
@@ -31,7 +32,6 @@ class TTSProviderBase(ABC):
     def __init__(self, config, delete_audio_file):
         self.interface_type = InterfaceType.NON_STREAM
         self.conn = None
-        self.tts_timeout = 10
         self.delete_audio_file = delete_audio_file
         self.audio_file_type = "wav"
         self.output_file = config.get("output_dir", "tmp/")
@@ -50,11 +50,9 @@ class TTSProviderBase(ABC):
             "；",
             ";",
             "：",
-            "~",
         )
         self.first_sentence_punctuations = (
             "，",
-            "～",
             "~",
             "、",
             ",",
@@ -77,6 +75,75 @@ class TTSProviderBase(ABC):
             f"tts-{datetime.now().date()}@{uuid.uuid4().hex}{extension}",
         )
 
+    def handle_opus(self, opus_data: bytes):
+        logger.bind(tag=TAG).debug(f"推送数据到队列里面帧数～～ {len(opus_data)}")
+        self.tts_audio_queue.put((SentenceType.MIDDLE, opus_data, None))
+
+    def handle_audio_file(self, file_audio: bytes, text):
+        self.before_stop_play_files.append((file_audio, text))
+
+    def to_tts_stream(self, text, opus_handler: Callable[[bytes], None] = None) -> None:
+        text = MarkdownCleaner.clean_markdown(text)
+        max_repeat_time = 5
+        if self.delete_audio_file:
+            # 需要删除文件的直接转为音频数据
+            while max_repeat_time > 0:
+                try:
+                    audio_bytes = asyncio.run(self.text_to_speak(text, None))
+                    if audio_bytes:
+                        self.tts_audio_queue.put((SentenceType.FIRST, None, text))
+                        audio_bytes_to_data_stream(
+                            audio_bytes,
+                            file_type=self.audio_file_type,
+                            is_opus=True,
+                            callback=opus_handler,
+                        )
+                        break
+                    else:
+                        max_repeat_time -= 1
+                except Exception as e:
+                    logger.bind(tag=TAG).warning(
+                        f"语音生成失败{5 - max_repeat_time + 1}次: {text}，错误: {e}"
+                    )
+                    max_repeat_time -= 1
+            if max_repeat_time > 0:
+                logger.bind(tag=TAG).info(
+                    f"语音生成成功: {text}，重试{5 - max_repeat_time}次"
+                )
+            else:
+                logger.bind(tag=TAG).error(
+                    f"语音生成失败: {text}，请检查网络或服务是否正常"
+                )
+            return None
+        else:
+            tmp_file = self.generate_filename()
+            try:
+                while not os.path.exists(tmp_file) and max_repeat_time > 0:
+                    try:
+                        asyncio.run(self.text_to_speak(text, tmp_file))
+                    except Exception as e:
+                        logger.bind(tag=TAG).warning(
+                            f"语音生成失败{5 - max_repeat_time + 1}次: {text}，错误: {e}"
+                        )
+                        # 未执行成功，删除文件
+                        if os.path.exists(tmp_file):
+                            os.remove(tmp_file)
+                        max_repeat_time -= 1
+
+                if max_repeat_time > 0:
+                    logger.bind(tag=TAG).info(
+                        f"语音生成成功: {text}:{tmp_file}，重试{5 - max_repeat_time}次"
+                    )
+                else:
+                    logger.bind(tag=TAG).error(
+                        f"语音生成失败: {text}，请检查网络或服务是否正常"
+                    )
+                    self.tts_audio_queue.put((SentenceType.FIRST, None, text))
+                self._process_audio_file_stream(tmp_file, callback=opus_handler)
+            except Exception as e:
+                logger.bind(tag=TAG).error(f"Failed to generate TTS file: {e}")
+                return None
+    
     def to_tts(self, text):
         text = MarkdownCleaner.clean_markdown(text)
         max_repeat_time = 5
@@ -86,8 +153,12 @@ class TTSProviderBase(ABC):
                 try:
                     audio_bytes = asyncio.run(self.text_to_speak(text, None))
                     if audio_bytes:
-                        audio_datas, _ = audio_bytes_to_data(
+                        audio_datas = []
-                            audio_bytes, file_type=self.audio_file_type, is_opus=True
+                        audio_bytes_to_data_stream(
+                            audio_bytes,
+                            file_type=self.audio_file_type,
+                            is_opus=True,
+                            callback=lambda data: audio_datas.append(data)
                         )
                         return audio_datas
                     else:
@@ -139,13 +210,17 @@ class TTSProviderBase(ABC):
     async def text_to_speak(self, text, output_file):
         pass
 
-    def audio_to_pcm_data(self, audio_file_path):
+    def audio_to_pcm_data_stream(
+        self, audio_file_path, callback: Callable[[Any], Any] = None
+    ):
         """音频文件转换为PCM编码"""
-        return audio_to_data(audio_file_path, is_opus=False)
+        return audio_to_data_stream(audio_file_path, is_opus=False, callback=callback)
 
-    def audio_to_opus_data(self, audio_file_path):
+    def audio_to_opus_data_stream(
+        self, audio_file_path, callback: Callable[[Any], Any] = None
+    ):
         """音频文件转换为Opus编码"""
-        return audio_to_data(audio_file_path, is_opus=True)
+        return audio_to_data_stream(audio_file_path, is_opus=True, callback=callback)
 
     def tts_one_sentence(
         self,
@@ -177,7 +252,6 @@ class TTSProviderBase(ABC):
 
     async def open_audio_channels(self, conn):
         self.conn = conn
-        self.tts_timeout = conn.config.get("tts_timeout", 10)
         # tts 消化线程
         self.tts_priority_thread = threading.Thread(
             target=self.tts_text_priority_thread, daemon=True
@@ -212,30 +286,16 @@ class TTSProviderBase(ABC):
                     self.tts_text_buff.append(message.content_detail)
                     segment_text = self._get_segment_text()
                     if segment_text:
-                        if self.delete_audio_file:
+                        self.to_tts_stream(segment_text, opus_handler=self.handle_opus)
-                            audio_datas = self.to_tts(segment_text)
-                            if audio_datas:
-                                self.tts_audio_queue.put(
-                                    (message.sentence_type, audio_datas, segment_text)
-                                )
-                        else:
-                            tts_file = self.to_tts(segment_text)
-                            if tts_file:
-                                audio_datas = self._process_audio_file(tts_file)
-                                self.tts_audio_queue.put(
-                                    (message.sentence_type, audio_datas, segment_text)
-                                )
                 elif ContentType.FILE == message.content_type:
-                    self._process_remaining_text()
+                    self._process_remaining_text_stream(opus_handler=self.handle_opus)
                     tts_file = message.content_file
                     if tts_file and os.path.exists(tts_file):
-                        audio_datas = self._process_audio_file(tts_file)
+                        self._process_audio_file_stream(
-                        self.tts_audio_queue.put(
+                            tts_file, callback=self.handle_opus
-                            (message.sentence_type, audio_datas, message.content_detail)
                         )
-
                 if message.sentence_type == SentenceType.LAST:
-                    self._process_remaining_text()
+                    self._process_remaining_text_stream(opus_handler=self.handle_opus)
                     self.tts_audio_queue.put(
                         (message.sentence_type, [], message.content_detail)
                     )
@@ -249,29 +309,59 @@ class TTSProviderBase(ABC):
                 continue
 
     def _audio_play_priority_thread(self):
+        # 需要上报的文本和音频列表
+        enqueue_text = None
+        enqueue_audio = None
         while not self.conn.stop_event.is_set():
             text = None
             try:
                 try:
                     sentence_type, audio_datas, text = self.tts_audio_queue.get(
-                        timeout=1
+                        timeout=0.1
                     )
                 except queue.Empty:
                     if self.conn.stop_event.is_set():
                         break
                     continue
+
+                if self.conn.client_abort:
+                    logger.bind(tag=TAG).debug("收到打断信号，跳过当前音频数据")
+                    enqueue_text, enqueue_audio = None, []
+                    continue
+
+                # 收到下一个文本开始或会话结束时进行上报
+                if sentence_type is not SentenceType.MIDDLE:
+                    # 重置音频流控状态（新句子开始或者结束）
+                    if hasattr(self.conn, 'audio_flow_control'):
+                        self.conn.audio_flow_control = {
+                            'last_send_time': 0,
+                            'packet_count': 0,
+                            'start_time': time.perf_counter()
+                        }
+                    
+                    # 上报TTS数据
+                    if enqueue_text is not None and enqueue_audio is not None:
+                        enqueue_tts_report(self.conn, enqueue_text, enqueue_audio)
+                    enqueue_audio = []
+                    enqueue_text = text
+
+                # 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
                 future = asyncio.run_coroutine_threadsafe(
                     sendAudioMessage(self.conn, sentence_type, audio_datas, text),
                     self.conn.loop,
                 )
                 future.result()
+
+                # 记录输出和报告
                 if self.conn.max_output_size > 0 and text:
                     add_device_output(self.conn.headers.get("device-id"), len(text))
-                enqueue_tts_report(self.conn, text, audio_datas)
+
             except Exception as e:
-                logger.bind(tag=TAG).error(
+                logger.bind(tag=TAG).error(f"audio_play_priority_thread: {text} {e}")
-                    f"audio_play_priority priority_thread: {text} {e}"
-                )
 
     async def start_session(self, session_id):
         pass
@@ -323,22 +413,21 @@ class TTSProviderBase(ABC):
         else:
             return None
 
-    def _process_audio_file(self, tts_file):
+    def _process_audio_file_stream(
+        self, tts_file, callback: Callable[[Any], Any]
+    ) -> None:
         """处理音频文件并转换为指定格式
 
         Args:
             tts_file: 音频文件路径
-            content_detail: 内容详情
+            callback: 文件处理函数
-
-        Returns:
-            tuple: (sentence_type, audio_datas, content_detail)
         """
         if tts_file.endswith(".p3"):
-            audio_datas, _ = p3.decode_opus_from_file(tts_file)
+            p3.decode_opus_from_file_stream(tts_file, callback=callback)
         elif self.conn.audio_format == "pcm":
-            audio_datas, _ = self.audio_to_pcm_data(tts_file)
+            self.audio_to_pcm_data_stream(tts_file, callback=callback)
         else:
-            audio_datas, _ = self.audio_to_opus_data(tts_file)
+            self.audio_to_opus_data_stream(tts_file, callback=callback)
 
         if (
             self.delete_audio_file
@@ -347,7 +436,6 @@ class TTSProviderBase(ABC):
             and tts_file.startswith(self.output_file)
         ):
             os.remove(tts_file)
-        return audio_datas
 
     def _process_before_stop_play_files(self):
         for audio_datas, text in self.before_stop_play_files:
@@ -355,7 +443,9 @@ class TTSProviderBase(ABC):
         self.before_stop_play_files.clear()
         self.tts_audio_queue.put((SentenceType.LAST, [], None))
 
-    def _process_remaining_text(self):
+    def _process_remaining_text_stream(
+        self, opus_handler: Callable[[bytes], None] = None
+    ):
         """处理剩余的文本并生成语音
 
         Returns:
@@ -366,18 +456,7 @@ class TTSProviderBase(ABC):
         if remaining_text:
             segment_text = textUtils.get_string_no_punctuation_or_emoji(remaining_text)
             if segment_text:
-                if self.delete_audio_file:
+                self.to_tts_stream(segment_text, opus_handler=opus_handler)
-                    audio_datas = self.to_tts(segment_text)
-                    if audio_datas:
-                        self.tts_audio_queue.put(
-                            (SentenceType.MIDDLE, audio_datas, segment_text)
-                        )
-                else:
-                    tts_file = self.to_tts(segment_text)
-                    audio_datas = self._process_audio_file(tts_file)
-                    self.tts_audio_queue.put(
-                        (SentenceType.MIDDLE, audio_datas, segment_text)
-                    )
                 self.processed_chars += len(full_text)
                 return True
         return False

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

@@ -4,6 +4,7 @@ import json
 import queue
 import asyncio
 import traceback
+from typing import Callable, Any
 import websockets
 from core.utils.tts import MarkdownCleaner
 from config.logger import setup_logging
@@ -266,11 +267,7 @@ class TTSProvider(TTSProviderBase):
                     )
                     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._process_audio_file_stream(message.content_file, callback=lambda audio_data: self.handle_audio_file(audio_data, message.content_detail))
-                        self.before_stop_play_files.append(
-                            (file_audio, message.content_detail)
-                        )
-
                 if message.sentence_type == SentenceType.LAST:
                     try:
                         logger.bind(tag=TAG).info("开始结束TTS会话...")
@@ -428,9 +425,6 @@ class TTSProvider(TTSProviderBase):
 
     async def _start_monitor_tts_response(self):
         """监听TTS响应"""
-        opus_datas_cache = []
-        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():
@@ -451,37 +445,14 @@ class TTSProvider(TTSProviderBase):
                         self.tts_audio_queue.put(
                             (SentenceType.FIRST, [], self.tts_text)
                         )
-                        opus_datas_cache = []
-                        first_sentence_segment_count = 0  # 重置计数器
                     elif (
                         res.optional.event == EVENT_TTSResponse
                         and res.header.message_type == AUDIO_ONLY_RESPONSE
                     ):
                         logger.bind(tag=TAG).debug(f"推送数据到队列里面～～")
-                        opus_datas = self.wav_to_opus_data_audio_raw(res.payload)
+                        self.wav_to_opus_data_audio_raw_stream(res.payload, callback=self.handle_opus)
-                        logger.bind(tag=TAG).debug(
-                            f"推送数据到队列里面帧数～～{len(opus_datas)}"
-                        )
-                        if is_first_sentence:
-                            first_sentence_segment_count += 1
-                            if first_sentence_segment_count <= 6:
-                                self.tts_audio_queue.put(
-                                    (SentenceType.MIDDLE, opus_datas, None)
-                                )
-                            else:
-                                opus_datas_cache.extend(opus_datas)
-                        else:
-                            # 后续句子缓存
-                            opus_datas_cache.extend(opus_datas)
                     elif res.optional.event == EVENT_TTSSentenceEnd:
                         logger.bind(tag=TAG).info(f"句子语音生成成功：{self.tts_text}")
-                        if not is_first_sentence or first_sentence_segment_count > 10:
-                            # 发送缓存的数据
-                            self.tts_audio_queue.put(
-                                (SentenceType.MIDDLE, opus_datas_cache, None)
-                            )
-                        # 第一句话结束后，将标志设置为False
-                        is_first_sentence = False
                     elif res.optional.event == EVENT_SessionFinished:
                         logger.bind(tag=TAG).debug(f"会话结束～～")
                         self._process_before_stop_play_files()
@@ -655,16 +626,13 @@ class TTSProvider(TTSProviderBase):
             )
         )
 
-    def wav_to_opus_data_audio_raw(self, raw_data_var, is_end=False):
+    def wav_to_opus_data_audio_raw_stream(self, raw_data_var, is_end=False, callback: Callable[[Any], Any]=None):
-        opus_datas = self.opus_encoder.encode_pcm_to_opus(raw_data_var, is_end)
+        return self.opus_encoder.encode_pcm_to_opus_stream(raw_data_var, is_end, callback=callback)
-        return opus_datas
 
     def to_tts(self, text: str) -> list:
         """非流式生成音频数据，用于生成音频及测试场景
-
         Args:
             text: 要转换的文本
-
         Returns:
             list: 音频数据列表
         """
@@ -741,8 +709,7 @@ class TTSProvider(TTSProviderBase):
                             res.optional.event == EVENT_TTSResponse
                             and res.header.message_type == AUDIO_ONLY_RESPONSE
                         ):
-                            opus_datas = self.wav_to_opus_data_audio_raw(res.payload)
+                            self.wav_to_opus_data_audio_raw_stream(res.payload, callback=lambda opus_frame: audio_data.append(opus_frame))
-                            audio_data.extend(opus_datas)
                         elif res.optional.event == EVENT_SessionFinished:
                             break
 

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

@@ -1,10 +1,10 @@
 import os
-import queue
-import asyncio
-import traceback
-import aiohttp
-import requests
 import time
+import queue
+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
@@ -27,15 +27,13 @@ class TTSProvider(TTSProviderBase):
         self.api_url = config.get("api_url", "http://8.138.114.124:11996/tts")
         self.audio_format = "pcm"
         self.before_stop_play_files = []
-        self.segment_count = 0
 
         # 创建Opus编码器 需注意接口返回的采样率为24000
         self.opus_encoder = opus_encoder_utils.OpusEncoderUtils(
             sample_rate=24000, channels=1, frame_size_ms=60
         )
 
-        # 文本缓冲区和PCM缓冲区
+        # PCM缓冲区
-        self.text_buffer = ""
         self.pcm_buffer = bytearray()
 
     def tts_text_priority_thread(self):
@@ -48,7 +46,6 @@ class TTSProvider(TTSProviderBase):
                     self.tts_stop_request = False
                     self.processed_chars = 0
                     self.tts_text_buff = []
-                    self.segment_count = 0
                     self.before_stop_play_files.clear()
                 elif ContentType.TEXT == message.content_type:
                     self.tts_text_buff.append(message.content_detail)
@@ -62,14 +59,11 @@ class TTSProvider(TTSProviderBase):
                     )
                     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._process_audio_file_stream(message.content_file, callback=lambda audio_data: self.handle_audio_file(audio_data, message.content_detail))
-                        self.before_stop_play_files.append(
-                            (file_audio, message.content_detail)
-                        )
 
                 if message.sentence_type == SentenceType.LAST:
                     # 处理剩余的文本
-                    self._process_remaining_text(True)
+                    self._process_remaining_text_stream(True)
 
             except queue.Empty:
                 continue
@@ -78,7 +72,7 @@ class TTSProvider(TTSProviderBase):
                     f"处理TTS文本失败: {str(e)}, 类型: {type(e).__name__}, 堆栈: {traceback.format_exc()}"
                 )
 
-    def _process_remaining_text(self, is_last=False):
+    def _process_remaining_text_stream(self, is_last=False):
         """处理剩余的文本并生成语音
         Returns:
             bool: 是否成功处理了文本
@@ -143,8 +137,6 @@ class TTSProvider(TTSProviderBase):
                         return
 
                     self.pcm_buffer.clear()
-                    opus_datas_cache = []
-
                     self.tts_audio_queue.put((SentenceType.FIRST, [], text))
 
                     # 处理音频流数据
@@ -158,41 +150,22 @@ class TTSProvider(TTSProviderBase):
                         while len(self.pcm_buffer) >= frame_bytes:
                             frame = bytes(self.pcm_buffer[:frame_bytes])
                             del self.pcm_buffer[:frame_bytes]
-                            opus = self.opus_encoder.encode_pcm_to_opus(
+
-                                frame, end_of_stream=False
+                            self.opus_encoder.encode_pcm_to_opus_stream(
+                                frame,
+                                end_of_stream=False,
+                                callback=self.handle_opus
                             )
-                            if opus:
-                                if self.segment_count < 10:  # 前10个片段直接发送
-                                    self.tts_audio_queue.put(
-                                        (SentenceType.MIDDLE, opus, None)
-                                    )
-                                    self.segment_count += 1
-                                else:
-                                    opus_datas_cache.extend(opus)
 
                     # flush 剩余不足一帧的数据
                     if self.pcm_buffer:
-                        opus = self.opus_encoder.encode_pcm_to_opus(
+                        self.opus_encoder.encode_pcm_to_opus_stream(
-                            bytes(self.pcm_buffer), end_of_stream=True
+                            bytes(self.pcm_buffer),
+                            end_of_stream=True,
+                            callback=self.handle_opus
                         )
-                        if opus:
-                            if self.segment_count < 10:  # 前10个片段直接发送
-                                # 直接发送
-                                self.tts_audio_queue.put(
-                                    (SentenceType.MIDDLE, opus, None)
-                                )
-                                self.segment_count += 1
-                            else:
-                                # 后续片段缓存
-                                opus_datas_cache.extend(opus)
                         self.pcm_buffer.clear()
 
-                    # 如果不是前10个片段，发送缓存的数据
-                    if self.segment_count >= 10 and opus_datas_cache:
-                        self.tts_audio_queue.put(
-                            (SentenceType.MIDDLE, opus_datas_cache, None)
-                        )
-
                     # 如果是最后一段，输出音频获取完毕
                     if is_last:
                         self._process_before_stop_play_files()
@@ -209,10 +182,8 @@ class TTSProvider(TTSProviderBase):
 
     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
         """
@@ -251,11 +222,11 @@ class TTSProvider(TTSProviderBase):
                         # 最后一帧可能不足，用0填充
                         frame = frame + b"\x00" * (frame_bytes - len(frame))
 
-                    opus = self.opus_encoder.encode_pcm_to_opus(
+                    self.opus_encoder.encode_pcm_to_opus_stream(
-                        frame, end_of_stream=(i + frame_bytes >= len(pcm_data))
+                        frame,
+                        end_of_stream=(i + frame_bytes >= len(pcm_data)),
+                        callback=lambda opus: opus_datas.append(opus)
                     )
-                    if opus:
-                        opus_datas.extend(opus)
 
                 return opus_datas
 

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

@@ -1,10 +1,10 @@
 import os
-import queue
-import asyncio
-import traceback
-import aiohttp
-import requests
 import time
+import queue
+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
@@ -24,23 +24,15 @@ class TTSProvider(TTSProviderBase):
         self.api_url = config.get("api_url")
         self.audio_format = "pcm"
         self.before_stop_play_files = []
-        self.segment_count = 0  # 添加片段计数器
 
         # 创建Opus编码器
         self.opus_encoder = opus_encoder_utils.OpusEncoderUtils(
             sample_rate=16000, channels=1, frame_size_ms=60
         )
 
-        # 添加文本缓冲区
-        self.text_buffer = ""
-
         # PCM缓冲区
         self.pcm_buffer = bytearray()
 
-    ###################################################################################
-    # linkerai单流式TTS重写父类的方法--开始
-    ###################################################################################
-
     def tts_text_priority_thread(self):
         """流式文本处理线程"""
         while not self.conn.stop_event.is_set():
@@ -51,7 +43,6 @@ class TTSProvider(TTSProviderBase):
                     self.tts_stop_request = False
                     self.processed_chars = 0
                     self.tts_text_buff = []
-                    self.segment_count = 0
                     self.before_stop_play_files.clear()
                 elif ContentType.TEXT == message.content_type:
                     self.tts_text_buff.append(message.content_detail)
@@ -65,14 +56,10 @@ class TTSProvider(TTSProviderBase):
                     )
                     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._process_audio_file_stream(message.content_file, callback=lambda audio_data: self.handle_audio_file(audio_data, message.content_detail))
-                        self.before_stop_play_files.append(
-                            (file_audio, message.content_detail)
-                        )
-
                 if message.sentence_type == SentenceType.LAST:
                     # 处理剩余的文本
-                    self._process_remaining_text(True)
+                    self._process_remaining_text_stream(True)
 
             except queue.Empty:
                 continue
@@ -81,7 +68,7 @@ class TTSProvider(TTSProviderBase):
                     f"处理TTS文本失败: {str(e)}, 类型: {type(e).__name__}, 堆栈: {traceback.format_exc()}"
                 )
 
-    def _process_remaining_text(self, is_last=False):
+    def _process_remaining_text_stream(self, is_last=False):
         """处理剩余的文本并生成语音
 
         Returns:
@@ -124,10 +111,6 @@ class TTSProvider(TTSProviderBase):
         finally:
             return None
 
-    ###################################################################################
-    # linkerai单流式TTS重写父类的方法--结束
-    ###################################################################################
-
     async def text_to_speak(self, text, is_last):
         """流式处理TTS音频，每句只推送一次音频列表"""
         await self._tts_request(text, is_last)
@@ -176,8 +159,6 @@ class TTSProvider(TTSProviderBase):
                         return
 
                     self.pcm_buffer.clear()
-                    opus_datas_cache = []
-
                     self.tts_audio_queue.put((SentenceType.FIRST, [], text))
 
                     # 兼容 iter_chunked / iter_chunks / iter_any
@@ -194,41 +175,21 @@ class TTSProvider(TTSProviderBase):
                             frame = bytes(self.pcm_buffer[:frame_bytes])
                             del self.pcm_buffer[:frame_bytes]
 
-                            opus = self.opus_encoder.encode_pcm_to_opus(
+                            self.opus_encoder.encode_pcm_to_opus_stream(
-                                frame, end_of_stream=False
+                                frame,
+                                end_of_stream=False,
+                                callback=self.handle_opus
                             )
-                            if opus:
-                                if self.segment_count < 10:  # 前10个片段直接发送
-                                    self.tts_audio_queue.put(
-                                        (SentenceType.MIDDLE, opus, None)
-                                    )
-                                    self.segment_count += 1
-                                else:
-                                    opus_datas_cache.extend(opus)
 
                     # flush 剩余不足一帧的数据
                     if self.pcm_buffer:
-                        opus = self.opus_encoder.encode_pcm_to_opus(
+                        self.opus_encoder.encode_pcm_to_opus_stream(
-                            bytes(self.pcm_buffer), end_of_stream=True
+                            bytes(self.pcm_buffer),
+                            end_of_stream=True,
+                            callback=self.handle_opus
                         )
-                        if opus:
-                            if self.segment_count < 10:  # 前10个片段直接发送
-                                # 直接发送
-                                self.tts_audio_queue.put(
-                                    (SentenceType.MIDDLE, opus, None)
-                                )
-                                self.segment_count += 1
-                            else:
-                                # 后续片段缓存
-                                opus_datas_cache.extend(opus)
                         self.pcm_buffer.clear()
 
-                    # 如果不是前10个片段，发送缓存的数据
-                    if self.segment_count >= 10 and opus_datas_cache:
-                        self.tts_audio_queue.put(
-                            (SentenceType.MIDDLE, opus_datas_cache, None)
-                        )
-
                     # 如果是最后一段，输出音频获取完毕
                     if is_last:
                         self._process_before_stop_play_files()
@@ -239,10 +200,8 @@ class TTSProvider(TTSProviderBase):
 
     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
         """
@@ -295,11 +254,11 @@ class TTSProvider(TTSProviderBase):
                         # 最后一帧可能不足，用0填充
                         frame = frame + b"\x00" * (frame_bytes - len(frame))
 
-                    opus = self.opus_encoder.encode_pcm_to_opus(
+                    self.opus_encoder.encode_pcm_to_opus_stream(
-                        frame, end_of_stream=(i + frame_bytes >= len(pcm_data))
+                        frame,
+                        end_of_stream=(i + frame_bytes >= len(pcm_data)),
+                        callback=lambda opus: opus_datas.append(opus)
                     )
-                    if opus:
-                        opus_datas.extend(opus)
 
                 return opus_datas
 

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

@@ -1,13 +1,15 @@
-import asyncio
-import json
-import base64
-import aiohttp
-import numpy as np
 import io
 import wave
+import json
+import base64
+import asyncio
 import websockets
-from core.providers.tts.base import TTSProviderBase
+import numpy as np
+from datetime import datetime
 from config.logger import setup_logging
+from core.providers.tts.base import TTSProviderBase
+
+
 
 TAG = __name__
 logger = setup_logging()
@@ -18,6 +20,7 @@ class TTSProvider(TTSProviderBase):
         super().__init__(config, delete_audio_file)
         self.url = config.get("url", "ws://192.168.1.10:8092/paddlespeech/tts/streaming")
         self.protocol = config.get("protocol", "websocket")
+        
         if config.get("private_voice"):
             self.spk_id = int(config.get("private_voice"))
         else:
@@ -32,7 +35,21 @@ class TTSProvider(TTSProviderBase):
         volume = config.get("volume", 1.0)
         self.volume = float(volume) if volume else 1.0
         
-        self.save_path = config.get("save_path", "./streaming_tts.wav")
+        self.delete_audio_file = config.get("delete_audio", True)
+        if not self.delete_audio_file:
+            timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
+            save_path = config.get("save_path")
+            if save_path:
+                if not save_path.endswith('.wav'):
+                    save_path = f"{save_path}_{timestamp}.wav"
+                else:
+                    other_path = save_path[:-4]
+                    save_path = f"{other_path}_{timestamp}.wav"
+                self.save_path = save_path
+            else:
+                self.save_path = f"./streaming_tts_{timestamp}.wav"
+        else:
+            self.save_path = None
 
     async def pcm_to_wav(self, pcm_data: bytes, sample_rate: int = 24000, num_channels: int = 1,
                          bits_per_sample: int = 16) -> bytes:
@@ -58,43 +75,9 @@ class TTSProvider(TTSProviderBase):
     async def text_to_speak(self, text, output_file):
         if self.protocol == "websocket":
             return await self.text_streaming(text, output_file)
-        elif self.protocol == "http":
-            return await self.text(text, output_file)
         else:
             raise ValueError("Unsupported protocol. Please use 'websocket' or 'http'.")
 
-    async def text(self, text, output_file):
-        request_json = {
-            "text": text,
-            "spk_id": self.spk_id,
-            "speed": self.speed,
-            "volume": self.volume,
-            "sample_rate": self.sample_rate,
-            "save_path": self.save_path
-        }
-
-        try:
-            async with aiohttp.ClientSession() as session:
-                async with session.post(self.url, json=request_json) as resp:
-                    if resp.status == 200:
-                        resp_json = await resp.json()
-                        if resp_json.get("success"):
-                            data = resp_json["result"]
-                            audio_bytes = base64.b64decode(data["audio"])
-                            if output_file:
-                                with open(output_file, "wb") as file_to_save:
-                                    file_to_save.write(audio_bytes)
-                            else:
-                                return audio_bytes
-                        else:
-                            raise Exception(
-                                f"Error: {resp_json.get('message', 'Unknown error')} while processing text: {text}")
-                    else:
-                        raise Exception(
-                            f"HTTP Error: {resp.status} - {await resp.text()} while processing text: {text}")
-        except Exception as e:
-            raise Exception(f"Error during TTS HTTP request: {e} while processing text: {text}")
-
     async def text_streaming(self, text, output_file):
         try:
             # 使用 websockets 异步连接到 WebSocket 服务器
@@ -151,6 +134,12 @@ class TTSProvider(TTSProviderBase):
                 # 接收结束响应避免服务抛出异常
                 await ws.recv()
 
+                # 根据配置决定是否保存文件
+                if not self.delete_audio_file and self.save_path:
+                    with open(self.save_path, "wb") as f:
+                        f.write(wav_data)
+                    logger.bind(tag=TAG).info(f"音频文件已保存到: {self.save_path}")
+                
                 # 返回或保存音频数据
                 if output_file:
                     with open(output_file, "wb") as file_to_save:

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

@@ -34,7 +34,7 @@ class VADProvider(VADProviderBase):
         )
 
         # 至少要多少帧才算有语音
-        self.frame_window_threshold = 1
+        self.frame_window_threshold = 3
 
     def is_vad(self, conn, opus_packet):
         try:
@@ -70,9 +70,7 @@ class VADProvider(VADProviderBase):
 
                 # 更新滑动窗口
                 conn.client_voice_window.append(is_voice)
-                client_have_voice = (
+                client_have_voice = (conn.client_voice_window.count(True) >= self.frame_window_threshold)
-                    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:

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

@@ -1,16 +1,17 @@
-import json
-import socket
-import subprocess
 import re
 import os
+import json
+import copy
 import wave
+import socket
+import requests
+import subprocess
+import numpy as np
+import opuslib_next
 from io import BytesIO
 from core.utils import p3
-import numpy as np
-import requests
-import opuslib_next
 from pydub import AudioSegment
-import copy
+from typing import Callable, Any
 
 TAG = __name__
 emoji_map = {
@@ -211,7 +212,7 @@ def extract_json_from_string(input_string):
     return None
 
 
-def audio_to_data(audio_file_path, is_opus=True):
+def audio_to_data_stream(audio_file_path, is_opus=True, callback: Callable[[Any], Any]=None) -> None:
     # 获取文件后缀名
     file_type = os.path.splitext(audio_file_path)[1]
     if file_type:
@@ -224,33 +225,32 @@ def audio_to_data(audio_file_path, is_opus=True):
     # 转换为单声道/16kHz采样率/16位小端编码（确保与编码器匹配）
     audio = audio.set_channels(1).set_frame_rate(16000).set_sample_width(2)
 
-    # 音频时长(秒)
+    # 获取原始PCM数据（16位小端）
-    duration = len(audio) / 1000.0
+    raw_data = audio.raw_data
+    pcm_to_data_stream(raw_data, is_opus, callback)
+
+def audio_to_data(audio_file_path: str, is_opus: bool = True) -> list[bytes]:
+    """
+    将音频文件转换为Opus/PCM编码的帧列表
+    Args:
+        audio_file_path: 音频文件路径
+        is_opus: 是否进行Opus编码
+    """
+    # 获取文件后缀名
+    file_type = os.path.splitext(audio_file_path)[1]
+    if file_type:
+        file_type = file_type.lstrip(".")
+    # 读取音频文件，-nostdin 参数：不要从标准输入读取数据，否则FFmpeg会阻塞
+    audio = AudioSegment.from_file(
+        audio_file_path, format=file_type, parameters=["-nostdin"]
+    )
+
+    # 转换为单声道/16kHz采样率/16位小端编码（确保与编码器匹配）
+    audio = audio.set_channels(1).set_frame_rate(16000).set_sample_width(2)
 
     # 获取原始PCM数据（16位小端）
     raw_data = audio.raw_data
-    return pcm_to_data(raw_data, is_opus), duration
 
-
-def audio_bytes_to_data(audio_bytes, file_type, is_opus=True):
-    """
-    直接用音频二进制数据转为opus/pcm数据，支持wav、mp3、p3
-    """
-    if file_type == "p3":
-        # 直接用p3解码
-        return p3.decode_opus_from_bytes(audio_bytes)
-    else:
-        # 其他格式用pydub
-        audio = AudioSegment.from_file(
-            BytesIO(audio_bytes), format=file_type, parameters=["-nostdin"]
-        )
-        audio = audio.set_channels(1).set_frame_rate(16000).set_sample_width(2)
-        duration = len(audio) / 1000.0
-        raw_data = audio.raw_data
-        return pcm_to_data(raw_data, is_opus), duration
-
-
-def pcm_to_data(raw_data, is_opus=True):
     # 初始化Opus编码器
     encoder = opuslib_next.Encoder(16000, 1, opuslib_next.APPLICATION_AUDIO)
 
@@ -280,6 +280,49 @@ def pcm_to_data(raw_data, is_opus=True):
 
     return datas
 
+def audio_bytes_to_data_stream(audio_bytes, file_type, is_opus, callback: Callable[[Any], Any]) -> None:
+    """
+    直接用音频二进制数据转为opus/pcm数据，支持wav、mp3、p3
+    """
+    if file_type == "p3":
+        # 直接用p3解码
+        return p3.decode_opus_from_bytes_stream(audio_bytes, callback)
+    else:
+        # 其他格式用pydub
+        audio = AudioSegment.from_file(
+            BytesIO(audio_bytes), format=file_type, parameters=["-nostdin"]
+        )
+        audio = audio.set_channels(1).set_frame_rate(16000).set_sample_width(2)
+        raw_data = audio.raw_data
+        pcm_to_data_stream(raw_data, is_opus, callback)
+
+
+def pcm_to_data_stream(raw_data, is_opus=True, callback: Callable[[Any], Any] = None):
+    # 初始化Opus编码器
+    encoder = opuslib_next.Encoder(16000, 1, opuslib_next.APPLICATION_AUDIO)
+
+    # 编码参数
+    frame_duration = 60  # 60ms per frame
+    frame_size = int(16000 * frame_duration / 1000)  # 960 samples/frame
+
+    # 按帧处理所有音频数据（包括最后一帧可能补零）
+    for i in range(0, len(raw_data), frame_size * 2):  # 16bit=2bytes/sample
+        # 获取当前帧的二进制数据
+        chunk = raw_data[i : i + frame_size * 2]
+
+        # 如果最后一帧不足，补零
+        if len(chunk) < frame_size * 2:
+            chunk += b"\x00" * (frame_size * 2 - len(chunk))
+
+        if is_opus:
+            # 转换为numpy数组处理
+            np_frame = np.frombuffer(chunk, dtype=np.int16)
+            # 编码Opus数据
+            frame_data = encoder.encode(np_frame.tobytes(), frame_size)
+            callback(frame_data)
+        else:
+            frame_data = chunk if isinstance(chunk, bytes) else bytes(chunk)
+            callback(frame_data)
 
 def opus_datas_to_wav_bytes(opus_datas, sample_rate=16000, channels=1):
     """
@@ -307,7 +350,6 @@ def opus_datas_to_wav_bytes(opus_datas, sample_rate=16000, channels=1):
         wf.writeframes(pcm_bytes)
     return wav_buffer.getvalue()
 
-
 def check_vad_update(before_config, new_config):
     if (
         new_config.get("selected_module") is None