perf(video): 优化视频处理性能监控和音频播放

- 添加视频处理性能计时和统计功能
- 实现帧处理时间监控和慢帧警告
- 添加音频文件静音修剪功能
- 优化Windows平台音频播放实现
- 调整默认日志输出频率减少冗余信息
- 修复MediaPipe GPU委托在Windows上的兼容性问题

app/exercises/dead_bug/detector.py

@@ -52,6 +52,7 @@ class DeadBugDetector:
         self._result_event = threading.Event()
         self._inflight = False
         self._inflight_started_at = 0.0
+        self.last_timing: dict[str, float | bool] = {}
 
         def on_result(pose_result, _image, _timestamp_ms):
             with self._result_lock:
@@ -79,7 +80,9 @@ class DeadBugDetector:
 
     def process_frame(self, bgr_frame: np.ndarray, timestamp_ms: int) -> tuple[np.ndarray, DeadBugResult]:
         """处理单帧：姿态检测、指标计算、状态机更新、可视化叠加"""
+        total_started = time.perf_counter()
         timestamp_ms = self._normalize_timestamp(timestamp_ms)
+        normalize_done = time.perf_counter()
 
         with self._result_lock:
             if self._inflight and time.monotonic() - self._inflight_started_at > 0.5:
@@ -90,9 +93,11 @@ class DeadBugDetector:
             if should_submit:
                 self._inflight = True
                 self._inflight_started_at = time.monotonic()
+        lock_done = time.perf_counter()
 
         if should_submit:
             rgba_frame = bgr_to_rgba(bgr_frame)
+            convert_done = time.perf_counter()
             mp_image = mp.Image(image_format=mp.ImageFormat.SRGBA, data=rgba_frame)
             self._result_event.clear()
             try:
@@ -102,12 +107,20 @@ class DeadBugDetector:
                     self._inflight = False
                     self._inflight_started_at = 0.0
                 raise
+            submit_done = time.perf_counter()
             self._result_event.wait(timeout=0.08)
+            wait_done = time.perf_counter()
+        else:
+            convert_done = lock_done
+            submit_done = lock_done
+            wait_done = lock_done
 
         with self._result_lock:
             pose_result = self._latest_result
+        result_read_done = time.perf_counter()
 
         annotated = bgr_frame.copy()
+        copy_done = time.perf_counter()
 
         if pose_result is None or not pose_result.pose_landmarks:
             self._state.mark_no_pose()
@@ -120,6 +133,18 @@ class DeadBugDetector:
                 metrics=None,
             )
             draw_status_overlay(annotated, result)
+            self._record_timing(
+                total_started,
+                normalize_done,
+                lock_done,
+                convert_done,
+                submit_done,
+                wait_done,
+                result_read_done,
+                copy_done,
+                time.perf_counter(),
+                should_submit,
+            )
             return annotated, result
 
         landmarks = [Point(lm.x, lm.y, lm.z, getattr(lm, "visibility", 1.0)) for lm in pose_result.pose_landmarks[0]]
@@ -136,6 +161,18 @@ class DeadBugDetector:
                 metrics=None,
             )
             draw_status_overlay(annotated, result)
+            self._record_timing(
+                total_started,
+                normalize_done,
+                lock_done,
+                convert_done,
+                submit_done,
+                wait_done,
+                result_read_done,
+                copy_done,
+                time.perf_counter(),
+                should_submit,
+            )
             return annotated, result
 
         raw = calculate_metrics(
@@ -169,8 +206,46 @@ class DeadBugDetector:
 
         result = self._state.update(metrics)
         draw_status_overlay(annotated, result)
+        self._record_timing(
+            total_started,
+            normalize_done,
+            lock_done,
+            convert_done,
+            submit_done,
+            wait_done,
+            result_read_done,
+            copy_done,
+            time.perf_counter(),
+            should_submit,
+        )
         return annotated, result
 
+    def _record_timing(
+        self,
+        total_started: float,
+        normalize_done: float,
+        lock_done: float,
+        convert_done: float,
+        submit_done: float,
+        wait_done: float,
+        result_read_done: float,
+        copy_done: float,
+        finished: float,
+        submitted: bool,
+    ) -> None:
+        self.last_timing = {
+            "total_ms": (finished - total_started) * 1000,
+            "timestamp_ms": (normalize_done - total_started) * 1000,
+            "lock_ms": (lock_done - normalize_done) * 1000,
+            "convert_ms": (convert_done - lock_done) * 1000,
+            "submit_ms": (submit_done - convert_done) * 1000,
+            "wait_ms": (wait_done - submit_done) * 1000,
+            "result_read_ms": (result_read_done - wait_done) * 1000,
+            "copy_ms": (copy_done - result_read_done) * 1000,
+            "postprocess_draw_ms": (finished - copy_done) * 1000,
+            "submitted": submitted,
+        }
+
     def _normalize_timestamp(self, timestamp_ms: int) -> int:
         """确保时间戳严格递增（MediaPipe要求）"""
         if timestamp_ms <= self._last_timestamp_ms: