From ec07088c13257cba8f0bad0be1c76f5d547ac76a Mon Sep 17 00:00:00 2001
From: Sakura-RanChen <1908198662@qq.com>
Date: Sat, 1 Nov 2025 17:59:16 +0800
Subject: [PATCH] =?UTF-8?q?fix:=20=E6=B5=81=E6=8E=A7=E7=8A=B6=E6=80=81?=
 =?UTF-8?q?=E6=9C=AA=E9=87=8D=E7=BD=AE=20update:=20=E5=A2=9E=E5=8A=A0?=
 =?UTF-8?q?=E6=B5=81=E5=BC=8F=E9=A2=84=E7=BC=93=E5=86=B2?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 .../core/handle/sendAudioHandle.py            | 33 ++++++++++---------
 1 file changed, 18 insertions(+), 15 deletions(-)

diff --git a/main/xiaozhi-server/core/handle/sendAudioHandle.py b/main/xiaozhi-server/core/handle/sendAudioHandle.py
index a820ee46..97573373 100644
--- a/main/xiaozhi-server/core/handle/sendAudioHandle.py
+++ b/main/xiaozhi-server/core/handle/sendAudioHandle.py
@@ -94,37 +94,40 @@ async def sendAudio(conn, audios, frame_duration=60):
     send_delay = conn.config.get("tts_audio_send_delay", -1) / 1000.0
 
     if isinstance(audios, bytes):
-        if conn.client_abort:
-            return
-
-        conn.last_activity_time = time.time() * 1000
-
-        # 获取或初始化流控状态
-        if not hasattr(conn, "audio_flow_control"):
+        # 重置流控状态,第一次读取和会话发生转变时
+        if not hasattr(conn, "audio_flow_control") or conn.audio_flow_control.get("sentence_id") != conn.sentence_id:
             conn.audio_flow_control = {
                 "last_send_time": 0,
                 "packet_count": 0,
                 "start_time": time.perf_counter(),
                 "sequence": 0,  # 添加序列号
+                "sentence_id": conn.sentence_id,
             }
 
+        if conn.client_abort:
+            return
+
+        conn.last_activity_time = time.time() * 1000
+
+        # 预缓冲：前5个包直接发送，不做延迟
+        pre_buffer_count = 5
         flow_control = conn.audio_flow_control
         current_time = time.perf_counter()
-        
-        if send_delay > 0:
+
+        if flow_control["packet_count"] < pre_buffer_count:
+            # 预缓冲阶段，直接发送不延迟
+            pass
+        elif send_delay > 0:
             # 使用固定延迟
             await asyncio.sleep(send_delay)
         else:
-            # 计算预期发送时间
+            effective_packet = flow_control["packet_count"] - pre_buffer_count
             expected_time = flow_control["start_time"] + (
-                flow_control["packet_count"] * frame_duration / 1000
+                effective_packet * frame_duration / 1000
             )
             delay = expected_time - current_time
             if delay > 0:
                 await asyncio.sleep(delay)
-            else:
-                # 纠正误差
-                flow_control["start_time"] += abs(delay)
 
         if conn.conn_from_mqtt_gateway:
             # 计算时间戳和序列号
@@ -150,7 +153,7 @@ async def sendAudio(conn, audios, frame_duration=60):
         play_position = 0
 
         # 执行预缓冲
-        pre_buffer_frames = min(3, len(audios))
+        pre_buffer_frames = min(5, len(audios))
         for i in range(pre_buffer_frames):
             if conn.conn_from_mqtt_gateway:
                 # 计算时间戳和序列号