From ba828b782d83690d0429e4d5025e1848aec6fb3a Mon Sep 17 00:00:00 2001
From: Del Wang <wjg12341234@163.com>
Date: Mon, 26 May 2025 23:12:04 +0800
Subject: [PATCH] =?UTF-8?q?fix:=20=E4=BC=98=E5=8C=96=20VAD=20=E8=AF=AD?=
 =?UTF-8?q?=E9=9F=B3=E7=89=87=E6=AE=B5=E6=94=B6=E9=9B=86?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 examples/xiaozhi/xiaozhi/event.py             |  6 +---
 .../xiaozhi/services/audio/vad/__init__.py    | 29 ++++++++++++++-----
 2 files changed, 23 insertions(+), 12 deletions(-)

diff --git a/examples/xiaozhi/xiaozhi/event.py b/examples/xiaozhi/xiaozhi/event.py
index f10eafd..88ea0e2 100644
--- a/examples/xiaozhi/xiaozhi/event.py
+++ b/examples/xiaozhi/xiaozhi/event.py
@@ -27,7 +27,6 @@ class __EventManager:
         self.session_id = 0
         self.current_step = Step.idle
         self.next_step_future = None
-        self.tts_session_id = ""
 
     def update_step(self, step: Step, step_data=None):
         if get_xiaoai().mode == "xiaozhi":
@@ -80,12 +79,9 @@ class __EventManager:
 
     def on_tts_end(self, session_id):
         """TTS结束"""
-        if self.current_step == Step.on_interrupt:
+        if self.current_step in [Step.on_interrupt, Step.on_tts_end]:
             # 当前 session 已经被打断了，不再处理
             return
-        if self.tts_session_id == session_id:
-            return
-        self.tts_session_id = session_id
         self.session_id = self.session_id + 1
         self.update_step(Step.on_tts_end)
         self.start_session()
diff --git a/examples/xiaozhi/xiaozhi/services/audio/vad/__init__.py b/examples/xiaozhi/xiaozhi/services/audio/vad/__init__.py
index 8738391..4613cff 100644
--- a/examples/xiaozhi/xiaozhi/services/audio/vad/__init__.py
+++ b/examples/xiaozhi/xiaozhi/services/audio/vad/__init__.py
@@ -33,16 +33,16 @@ class _VAD:
         self.stream = None
 
         # 暂存的语音片段
-        self.temp_frames = []
-        self.speech_buffer = []
+        self.silence_frames = []  # 静音片段
+        self.speech_frames = []  # 语音片段
         self.target = None  # 检测目标 speech/silence
 
     def _reset_state(self):
         """重置状态"""
         self.speech_count = 0
         self.silence_count = 0
-        self.speech_buffer = []
-        self.temp_frames = []
+        self.speech_frames = []
+        self.silence_frames = []
 
     def start(self):
         """启动VAD检测器"""
@@ -78,22 +78,37 @@ class _VAD:
         """处理语音帧"""
         self.speech_count += len(frames)
         self.silence_count = 0
-        self.speech_buffer.extend(frames)
 
-        speech_bytes = bytes(self.speech_buffer)
+        if not self.speech_frames:
+            # 加入静音片段（潜在的语音片段）
+            self.speech_frames.extend(self.silence_frames)
+
+        # 加入语音片段
+        self.speech_frames.extend(frames)
+
+        speech_bytes = bytes(self.speech_frames)
 
         if (
             self.target == "speech"
             and self.speech_count > self.min_speech_duration * self.sample_rate / 1000
         ):
             self.pause()
+            # !FIXME: 需要保证音频流的连续性（在发消息期间）
             EventManager.on_speech(speech_bytes)
 
     def _handle_silence_frame(self, frames):
         """处理静音帧"""
         self.silence_count += len(frames)
         self.speech_count = 0
-        self.speech_buffer = []
+
+        if not self.speech_frames:
+            # 如果之前没有语音片段，则将当前帧加入静音片段
+            self.silence_frames.extend(frames)
+            # 确保静音片段长度不超过 3s
+            self.silence_frames = self.silence_frames[: 3 * self.sample_rate]
+        else:
+            # 如果之前有语音片段，则将当前帧加入语音片段
+            self.speech_frames.extend(frames)
 
         if (
             self.target == "silence"