apps/worker/video_processing/unified_render_service.py

@@ -198,8 +198,13 @@ class UnifiedRenderService:
         # 2. 分组为 RenderLayers
         layers = self._group_clips_into_layers(resolved)
 
+        # 2.5 配音时长对齐：如果有配音素材，调整片段时长以匹配配音时长
+        voice_duration = self._get_voice_audio_duration()
+        if voice_duration > 0:
+            self._align_clips_to_voice_duration(layers, voice_duration)
+
         # 3. 计算视频总时长（用于字幕显示时长）
-        video_duration = self._estimate_total_duration(layers)
+        video_duration_final = self._estimate_total_duration(layers)
         # Debug: 输出各图层时长明细
         for layer in layers:
             layer_total = sum(UnifiedRenderService._clip_adjusted_duration(c) for c in layer.clips)
@@ -215,16 +220,16 @@ class UnifiedRenderService:
                 self.transition_duration,
                 ", ".join(clip_details),
             )
-        logger.info("[debug] estimated video_duration=%.3f", video_duration)
+        logger.info("[debug] estimated video_duration=%.3f", video_duration_final)
 
         # 3.5 TTS 配音生成（如果配置了）
-        self._maybe_add_voiceover_layer(layers, video_duration=video_duration)
+        self._maybe_add_voiceover_layer(layers, video_duration=video_duration_final)
 
         # 3.6 配音素材库音频（如果传入了本地路径）
-        self._maybe_add_voice_library_layer(layers, video_duration=video_duration)
+        self._maybe_add_voice_library_layer(layers, video_duration=video_duration_final)
 
         # 4. 生成 ASS 字幕文件（如果有 title/subtitle 配置）
-        ass_path = self._maybe_generate_ass(video_duration)
+        ass_path = self._maybe_generate_ass(video_duration_final)
 
         # 4.5 解析画中画配置
         pip_config = PiPConfig.from_dict((self.plan.config or {}).get("pip_config"))
@@ -257,7 +262,7 @@ class UnifiedRenderService:
             # 先尝试 stream copy 优化（无重编码，性能提升 10 倍+）
             # 条件不满足或失败时回退到带滤镜的直通渲染
             stream_copy_ok = self._try_render_stream_copy(
-                layers, output_path, ass_path=ass_path, video_duration=video_duration
+                layers, output_path, ass_path=ass_path, video_duration=video_duration_final
             )
             if stream_copy_ok:
                 used_stream_copy = True
@@ -271,7 +276,7 @@ class UnifiedRenderService:
                     layers,
                     output_path,
                     ass_path=ass_path,
-                    video_duration=video_duration,
+                    video_duration=video_duration_final,
                 )
         else:
             filter_complex, input_args = self._build_filter_complex(layers, ass_path=ass_path)
@@ -327,7 +332,7 @@ class UnifiedRenderService:
                         from video_processing.ffmpeg_utils import run_ffmpeg
 
                         run_ffmpeg(extract_cmd)
-                        final_audio = mix_bgm_with_main(ctx, main_audio_path, bgm_cfg, video_duration)
+                        final_audio = mix_bgm_with_main(ctx, main_audio_path, bgm_cfg, video_duration_final)
                         # 合并回视频
 
                         bgm_output = self.work_dir / f"rendered_{self.plan.id}_bgm.mp4"
@@ -353,7 +358,7 @@ class UnifiedRenderService:
             audio_path = mix_audio(
                 ctx,
                 layers,
-                video_duration,
+                video_duration_final,
                 bgm_path=self.bgm_path,
                 bgm_config=bgm_config,
                 audio_tracks_config=audio_tracks_config,
@@ -487,6 +492,147 @@ class UnifiedRenderService:
         """
         return _estimate_total_duration_pure(layers, self.transition_duration)
 
+    def _get_voice_audio_duration(self) -> float:
+        """获取配音音频文件的时长（秒）。
+
+        Returns:
+            配音音频时长，如果无配音或探测失败则返回 0.0
+        """
+        if not self.voiceover_audio_path:
+            return 0.0
+
+        audio_path = Path(self.voiceover_audio_path)
+        if not audio_path.exists() or audio_path.stat().st_size == 0:
+            return 0.0
+
+        try:
+            duration = probe_duration(audio_path)
+            logger.info("[voice-align] 配音音频时长: %.3fs path=%s", duration, self.voiceover_audio_path)
+            return duration
+        except Exception as e:
+            logger.warning("[voice-align] 探测配音音频时长失败: %s", e)
+            return 0.0
+
+    def _align_clips_to_voice_duration(
+        self,
+        layers: list[RenderLayer],
+        voice_duration: float,
+    ) -> None:
+        """调整片段时长以对齐配音时长。
+
+        核心逻辑：
+        - 计算片段总时长与配音时长的比例
+        - ±5% 以内不调整
+        - ratio < 1（片段比配音长）：按比例裁剪每段末尾
+        - ratio > 1（片段比配音短）：按比例慢放每段
+
+        Args:
+            layers: 渲染图层列表
+            voice_duration: 配音时长（秒）
+        """
+        if voice_duration <= 0:
+            return
+
+        # 只调整视频图层（main/broll/background），不调整音频图层
+        video_layers = [layer for layer in layers if layer.role in ("main", "broll", "background")]
+        if not video_layers:
+            return
+
+        # 计算所有视频图层的总时长
+        total_clips_duration = 0.0
+        for layer in video_layers:
+            for clip in layer.clips:
+                clip_dur = self._clip_adjusted_duration(clip)
+                total_clips_duration += clip_dur
+
+        if total_clips_duration <= 0:
+            return
+
+        ratio = voice_duration / total_clips_duration
+
+        # ±5% 以内不调整
+        if abs(ratio - 1.0) <= 0.05:
+            logger.info(
+                "[voice-align] 比例接近1:1，跳过调整: ratio=%.4f voice=%.3f clips=%.3f",
+                ratio,
+                voice_duration,
+                total_clips_duration,
+            )
+            return
+
+        logger.info(
+            "[voice-align] 开始调整片段时长: ratio=%.4f voice=%.3f clips=%.3f",
+            ratio,
+            voice_duration,
+            total_clips_duration,
+        )
+
+        # 收集所有视频 clip
+        all_clips: list[tuple[RenderLayer, ResolvedClip]] = []
+        for layer in video_layers:
+            for clip in layer.clips:
+                all_clips.append((layer, clip))
+
+        if not all_clips:
+            return
+
+        if ratio < 1.0:
+            # 片段比配音长，按比例裁剪每段末尾
+            # 减少每个 clip 的 duration
+            for _layer, clip in all_clips:
+                old_duration = clip.duration if clip.duration > 0 else clip.actual_duration
+                new_duration = old_duration * ratio
+
+                # 更新 duration
+                clip.duration = max(0.1, new_duration)  # 至少 0.1s
+
+                # 如果有 trim_config，也需要调整
+                if clip.trim_config is not None:
+                    new_trim_duration = clip.trim_config.duration * ratio
+                    clip.trim_config = TrimConfig(
+                        start_time=clip.trim_config.start_time,
+                        duration=max(0.1, new_trim_duration),
+                    )
+
+                logger.debug(
+                    "[voice-align] trim clip=%s: %.3f -> %.3f",
+                    clip.clip_id,
+                    old_duration,
+                    clip.duration,
+                )
+
+        else:
+            # ratio > 1.0: 片段比配音短，按比例慢放每段
+            # 降低 playback_speed
+            for _layer, clip in all_clips:
+                old_speed = clip.playback_speed if clip.playback_speed > 0 else 1.0
+                # speed = old_speed / ratio 会使视频变慢（ratio > 1 时）
+                new_speed = old_speed / ratio
+
+                # 下限 0.25x（避免过慢）
+                new_speed = max(0.25, round(new_speed, 4))
+                clip.playback_speed = new_speed
+
+                logger.debug(
+                    "[voice-align] slowdown clip=%s: speed %.4f -> %.4f",
+                    clip.clip_id,
+                    old_speed,
+                    new_speed,
+                )
+
+        # 调整后重新计算总时长用于日志
+        new_total = 0.0
+        for layer in video_layers:
+            for clip in layer.clips:
+                new_total += self._clip_adjusted_duration(clip)
+
+        logger.info(
+            "[voice-align] 调整完成: 新总时长=%.3fs (目标=%.3fs, 差异=%.3fs)",
+            new_total,
+            voice_duration,
+            abs(new_total - voice_duration),
+        )
+
     def _maybe_generate_ass(self, video_duration: float) -> Path | None:
         """根据 plan.config 生成 ASS 字幕文件。
 

tests/unit/test_voice_duration_alignment.py

@@ -0,0 +1,206 @@
+"""Tests for voice duration alignment feature.
+
+Tests the _align_clips_to_voice_duration method in UnifiedRenderService.
+"""
+
+from __future__ import annotations
+
+from pathlib import Path
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+from video_processing.unified_render_service import ResolvedClip, RenderLayer, UnifiedRenderService
+
+
+class TestAlignClipsToVoiceDuration:
+    """Test clip duration alignment to voice audio."""
+
+    def _make_clip(
+        self,
+        clip_id: str,
+        duration: float,
+        actual_duration: float = 0.0,
+        playback_speed: float = 1.0,
+    ) -> ResolvedClip:
+        """Helper to create a ResolvedClip for testing."""
+        return ResolvedClip(
+            clip_id=clip_id,
+            asset_id=f"asset_{clip_id}",
+            local_path=Path(f"/tmp/{clip_id}.mp4"),
+            clip_type="main",
+            order=0,
+            duration=duration,
+            actual_duration=actual_duration or duration,
+            playback_speed=playback_speed,
+        )
+
+    def _make_layer(self, role: str, clips: list[ResolvedClip]) -> RenderLayer:
+        """Helper to create a RenderLayer for testing."""
+        return RenderLayer(role=role, clips=clips, z_index=0)
+
+    def _make_service(self, voiceover_path: str | None = None) -> UnifiedRenderService:
+        """Helper to create a mock UnifiedRenderService."""
+        plan = MagicMock()
+        plan.id = "test_plan"
+        plan.config = {}
+        
+        with patch.object(UnifiedRenderService, "__init__", lambda self, **kwargs: None):
+            service = UnifiedRenderService.__new__(UnifiedRenderService)
+            service.plan = plan
+            service.voiceover_audio_path = voiceover_path
+            service.transition_duration = 0.0
+            return service
+
+    def test_no_voice_audio_no_adjustment(self):
+        """No voice audio → no adjustment."""
+        service = self._make_service(voiceover_path=None)
+        clips = [self._make_clip("c1", 10.0), self._make_clip("c2", 10.0)]
+        layers = [self._make_layer("main", clips)]
+
+        service._align_clips_to_voice_duration(layers, voice_duration=0.0)
+
+        # No change
+        assert clips[0].duration == 10.0
+        assert clips[1].duration == 10.0
+
+    def test_ratio_within_5_percent_no_adjustment(self):
+        """Ratio within ±5% → no adjustment."""
+        service = self._make_service()
+        clips = [self._make_clip("c1", 10.0)]
+        layers = [self._make_layer("main", clips)]
+
+        # Total clips = 10s, voice = 10.3s → ratio = 1.03 (within 5%)
+        service._align_clips_to_voice_duration(layers, voice_duration=10.3)
+
+        assert clips[0].duration == 10.0  # Unchanged
+
+    def test_ratio_less_than_1_trim_clips(self):
+        """Ratio < 1 (clips too long) → trim clips proportionally."""
+        service = self._make_service()
+        clips = [self._make_clip("c1", 10.0), self._make_clip("c2", 10.0)]
+        layers = [self._make_layer("main", clips)]
+
+        # Total clips = 20s, voice = 15s → ratio = 0.75
+        service._align_clips_to_voice_duration(layers, voice_duration=15.0)
+
+        # Each clip should be trimmed to 75%
+        assert abs(clips[0].duration - 7.5) < 0.01
+        assert abs(clips[1].duration - 7.5) < 0.01
+
+    def test_ratio_greater_than_1_slowdown_clips(self):
+        """Ratio > 1 (clips too short) → slow down clips."""
+        service = self._make_service()
+        clips = [self._make_clip("c1", 10.0), self._make_clip("c2", 10.0)]
+        layers = [self._make_layer("main", clips)]
+
+        # Total clips = 20s, voice = 25s → ratio = 1.25
+        service._align_clips_to_voice_duration(layers, voice_duration=25.0)
+
+        # Each clip's speed should be reduced: 1.0 / 1.25 = 0.8
+        assert abs(clips[0].playback_speed - 0.8) < 0.01
+        assert abs(clips[1].playback_speed - 0.8) < 0.01
+
+    def test_speed_lower_bound_025(self):
+        """Playback speed should not go below 0.25x."""
+        service = self._make_service()
+        clips = [self._make_clip("c1", 5.0)]
+        layers = [self._make_layer("main", clips)]
+
+        # Total clips = 5s, voice = 50s → ratio = 10.0
+        # Speed would be 1.0 / 10 = 0.1, but should be clamped to 0.25
+        service._align_clips_to_voice_duration(layers, voice_duration=50.0)
+
+        assert clips[0].playback_speed == 0.25
+
+    def test_only_video_layers_adjusted(self):
+        """Only main/broll/background layers are adjusted, not audio."""
+        service = self._make_service()
+        
+        video_clips = [self._make_clip("v1", 10.0)]
+        audio_clips = [self._make_clip("a1", 10.0)]
+        
+        layers = [
+            self._make_layer("main", video_clips),
+            self._make_layer("audio", audio_clips),
+        ]
+
+        # ratio = 0.5 → should trim video but not audio
+        service._align_clips_to_voice_duration(layers, voice_duration=5.0)
+
+        assert abs(video_clips[0].duration - 5.0) < 0.01  # Trimmed
+        assert audio_clips[0].duration == 10.0  # Unchanged
+
+    def test_multiple_video_layers_all_adjusted(self):
+        """All video layers (main, broll, background) are adjusted."""
+        service = self._make_service()
+        
+        main_clips = [self._make_clip("m1", 10.0)]
+        broll_clips = [self._make_clip("b1", 10.0)]
+        bg_clips = [self._make_clip("bg1", 10.0)]
+        
+        layers = [
+            self._make_layer("main", main_clips),
+            self._make_layer("broll", broll_clips),
+            self._make_layer("background", bg_clips),
+        ]
+
+        # Total video = 30s, voice = 15s → ratio = 0.5
+        service._align_clips_to_voice_duration(layers, voice_duration=15.0)
+
+        # All should be trimmed to 50%
+        assert abs(main_clips[0].duration - 5.0) < 0.01
+        assert abs(broll_clips[0].duration - 5.0) < 0.01
+        assert abs(bg_clips[0].duration - 5.0) < 0.01
+
+    def test_trim_config_also_adjusted(self):
+        """When clip has trim_config, it should also be adjusted."""
+        from video_processing.trim_engine import TrimConfig
+        
+        service = self._make_service()
+        
+        clip = self._make_clip("c1", 10.0)
+        clip.trim_config = TrimConfig(start_time=0.0, duration=10.0)
+        
+        layers = [self._make_layer("main", [clip])]
+
+        # ratio = 0.5
+        service._align_clips_to_voice_duration(layers, voice_duration=5.0)
+
+        assert abs(clip.duration - 5.0) < 0.01
+        assert clip.trim_config is not None
+        assert abs(clip.trim_config.duration - 5.0) < 0.01
+
+
+class TestGetVoiceAudioDuration:
+    """Test voice audio duration probing."""
+
+    def test_no_voiceover_path_returns_zero(self):
+        """No voiceover path → return 0."""
+        with patch.object(UnifiedRenderService, "__init__", lambda self, **kwargs: None):
+            service = UnifiedRenderService.__new__(UnifiedRenderService)
+            service.voiceover_audio_path = None
+
+        assert service._get_voice_audio_duration() == 0.0
+
+    def test_nonexistent_file_returns_zero(self):
+        """Nonexistent file → return 0."""
+        with patch.object(UnifiedRenderService, "__init__", lambda self, **kwargs: None):
+            service = UnifiedRenderService.__new__(UnifiedRenderService)
+            service.voiceover_audio_path = "/nonexistent/path.mp3"
+
+        assert service._get_voice_audio_duration() == 0.0
+
+    @patch("video_processing.unified_render_service.probe_duration")
+    @patch("video_processing.unified_render_service.Path.exists", return_value=True)
+    @patch("video_processing.unified_render_service.Path.stat")
+    def test_probes_duration_from_file(self, mock_stat, mock_exists, mock_probe):
+        """Valid file → probe duration."""
+        mock_stat.return_value.st_size = 1000  # Non-empty file
+        mock_probe.return_value = 42.5
+
+        with patch.object(UnifiedRenderService, "__init__", lambda self, **kwargs: None):
+            service = UnifiedRenderService.__new__(UnifiedRenderService)
+            service.voiceover_audio_path = "/tmp/voice.mp3"
+
+        assert service._get_voice_audio_duration() == 42.5