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/audio/generate.py

@@ -4,6 +4,7 @@ from __future__ import annotations
 import platform
 import shutil
 import subprocess
+import wave
 from pathlib import Path
 
 from loguru import logger
@@ -15,6 +16,9 @@ def generate_rep_audio_files(
     rate: int,
     output_dir: Path,
     overwrite: bool = False,
+    trim_leading_silence: bool = True,
+    trim_silence_threshold: int = 500,
+    trim_silence_padding_ms: int = 20,
 ) -> None:
     """
     确保 0~max_count 的运动次数语音 wav 文件存在。
@@ -40,29 +44,37 @@ def generate_rep_audio_files(
 
     if not missing_counts:
         logger.info("Rep audio files already prepared: {}", output_dir)
-        return
-
-    logger.info(
-        "Preparing rep audio files, system={}, count={}, output_dir={}",
-        system,
-        len(missing_counts),
-        output_dir,
-    )
-
-    if system == "darwin":
-        _generate_with_macos_say(
-            counts=missing_counts,
-            output_dir=output_dir,
-            rate=rate,
-        )
     else:
-        _generate_with_pyttsx3(
-            counts=missing_counts,
-            output_dir=output_dir,
-            rate=rate,
+        logger.info(
+            "Preparing rep audio files, system={}, count={}, output_dir={}",
+            system,
+            len(missing_counts),
+            output_dir,
         )
 
-    logger.info("Rep audio files prepared: {}", output_dir)
+        if system == "darwin":
+            _generate_with_macos_say(
+                counts=missing_counts,
+                output_dir=output_dir,
+                rate=rate,
+            )
+        else:
+            _generate_with_pyttsx3(
+                counts=missing_counts,
+                output_dir=output_dir,
+                rate=rate,
+            )
+
+        logger.info("Rep audio files prepared: {}", output_dir)
+
+    if trim_leading_silence and suffix == ".wav":
+        _trim_leading_silence_files(
+            counts=list(range(0, max_count + 1)),
+            output_dir=output_dir,
+            suffix=suffix,
+            threshold=trim_silence_threshold,
+            padding_ms=trim_silence_padding_ms,
+        )
 
 
 def _generate_with_macos_say(
@@ -127,3 +139,89 @@ def _generate_with_pyttsx3(
 
 def _audio_path(output_dir: Path, count: int, *, suffix: str) -> Path:
     return output_dir / f"{count}{suffix}"
+
+
+def _trim_leading_silence_files(
+    *,
+    counts: list[int],
+    output_dir: Path,
+    suffix: str,
+    threshold: int,
+    padding_ms: int,
+) -> None:
+    trimmed = 0
+    total_removed_ms = 0.0
+
+    for count in counts:
+        audio_file = _audio_path(output_dir, count, suffix=suffix)
+        if not audio_file.exists():
+            continue
+        removed_ms = _trim_leading_silence(audio_file, threshold=threshold, padding_ms=padding_ms)
+        if removed_ms > 0:
+            trimmed += 1
+            total_removed_ms += removed_ms
+
+    logger.info(
+        "Rep audio leading silence trim complete: files_trimmed={}, total_removed_ms={:.1f}, threshold={}, padding_ms={}",
+        trimmed,
+        total_removed_ms,
+        threshold,
+        padding_ms,
+    )
+
+
+def _trim_leading_silence(audio_file: Path, *, threshold: int, padding_ms: int) -> float:
+    with wave.open(str(audio_file), "rb") as reader:
+        params = reader.getparams()
+        frames = reader.readframes(params.nframes)
+
+    frame_size = params.sampwidth * params.nchannels
+    if params.nframes <= 0 or frame_size <= 0:
+        return 0.0
+
+    chunk_frames = max(1, params.framerate // 100)
+    leading_frames = 0
+    offset = 0
+    chunk_size = chunk_frames * frame_size
+
+    while offset < len(frames):
+        chunk = frames[offset : offset + chunk_size]
+        if _pcm_rms(chunk, params.sampwidth) > threshold:
+            break
+        chunk_frame_count = len(chunk) // frame_size
+        leading_frames += chunk_frame_count
+        offset += chunk_size
+
+    padding_frames = int(params.framerate * max(0, padding_ms) / 1000)
+    remove_frames = max(0, leading_frames - padding_frames)
+    if remove_frames <= 0:
+        return 0.0
+
+    start = min(len(frames), remove_frames * frame_size)
+    trimmed_frames = frames[start:]
+    if not trimmed_frames:
+        return 0.0
+
+    with wave.open(str(audio_file), "wb") as writer:
+        writer.setparams(params)
+        writer.writeframes(trimmed_frames)
+
+    return remove_frames / params.framerate * 1000
+
+
+def _pcm_rms(chunk: bytes, sample_width: int) -> float:
+    if not chunk:
+        return 0.0
+
+    if sample_width == 2:
+        sample_count = len(chunk) // 2
+        if sample_count == 0:
+            return 0.0
+        total = 0
+        for i in range(0, sample_count * 2, 2):
+            sample = int.from_bytes(chunk[i : i + 2], "little", signed=True)
+            total += sample * sample
+        return (total / sample_count) ** 0.5
+
+    peak = max(abs(byte - 128) for byte in chunk)
+    return float(peak)