xiaoxia/xiaoxia-saas
1
0
Fork 0
Code Issues 19 Pull Requests 1 Actions Packages Projects Releases 7 Wiki Activity

wip(#1749): render_audio 四路 apad 静音补齐 + stream-copy tpad + worker 重渲传配音时长（checkpoint）

Browse Source
This commit is contained in:
saas-backend-agent
2026-09-07 01:11:57 +08:00
parent 06ca57c8ae
commit eef372073a
3 changed files with 71 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files
apps/worker/video_processing/render_audio.py
+57 -13
View File
@@ -47,13 +47,23 @@ class RenderContext:
def clip_effective_duration(clip: ResolvedClip) -> float:
"""计算 clip 的有效时长.
"""计算 clip 的有效时长（#1749：目标段长始终为准）。
与 UnifiedRenderService._clip_effective_duration 逻辑一致。
与 UnifiedRenderService._clip_effective_duration 逻辑一致；素材短于段长的
部分由末帧冻结 tpad / 音频 apad 铺满，不在此处钳制。
"""
if clip.duration > 0:
return min(clip.duration, clip.actual_duration) if clip.actual_duration > 0 else clip.duration
return clip.actual_duration if clip.actual_duration > 0 else 0.0
return float(clip.duration)
return float(clip.actual_duration) if clip.actual_duration > 0 else 0.0
def _clip_freeze_seconds(clip: ResolvedClip) -> float:
"""读取 #1749 末帧冻结秒数（_resolve_clips 写入 config['_freeze_seconds']）。"""
cfg = getattr(clip, "config", None) or {}
try:
return max(0.0, float(cfg.get("_freeze_seconds", 0.0) or 0.0))
except (TypeError, ValueError):
return 0.0
def clip_has_audio(ctx: RenderContext, clip: ResolvedClip) -> bool:
@@ -273,6 +283,9 @@ def concat_main_audio(
if video_duration > 0 and (final_duration <= 0 or final_duration > video_duration):
final_duration = video_duration
# #1749 末帧冻结秒数（音频需 apad 补静音与视频等长）
freeze_seconds = _clip_freeze_seconds(clip)
# 音频倒放
reverse_config = ReverseConfig.from_dict(clip.config.get("reverse"))
has_reverse = reverse_config.enabled and reverse_config.reverse_audio
@@ -281,20 +294,30 @@ def concat_main_audio(
if not has_speed and not has_reverse:
# 无调速无倒放：根据是否需要裁剪/音量选择最高效的路径。
vol = _clip_volume(clip)
need_trim = trim_start > 0 or (effective_duration > 0 and final_duration < adjusted_duration)
need_trim = (
trim_start > 0
or freeze_seconds > 0
or (effective_duration > 0 and final_duration < adjusted_duration)
)
need_volume = abs(vol - 1.0) >= 1e-6
if need_trim:
# 需要裁剪：用 atrim 滤镜在滤镜链中精确裁剪（采样点级精度，不浪费解码）。
# 滤镜顺序：atrim → asetpts → volume（先裁剪再调音量，避免处理被丢弃的数据）。
# 滤镜顺序：atrim → asetpts → apad(冻结补静音) → volume。
af_parts: list[str] = []
if trim_start > 0 and effective_duration > 0:
af_parts.append(f"atrim=start={trim_start:.3f}:duration={final_duration:.3f}")
# 冻结场景：素材内可用时长 = 目标段长 − freeze（截掉超出素材的部分后补静音）
atrim_dur = final_duration
if freeze_seconds > 0:
atrim_dur = max(0.0, final_duration - freeze_seconds)
if trim_start > 0 and atrim_dur > 0:
af_parts.append(f"atrim=start={trim_start:.3f}:duration={atrim_dur:.3f}")
elif trim_start > 0:
af_parts.append(f"atrim=start={trim_start:.3f}")
elif final_duration > 0:
af_parts.append(f"atrim=duration={final_duration:.3f}")
elif atrim_dur > 0:
af_parts.append(f"atrim=duration={atrim_dur:.3f}")
af_parts.append("asetpts=PTS-STARTPTS")
if freeze_seconds > 0:
af_parts.append(f"apad=whole_dur={final_duration:.3f}")
if need_volume:
af_parts.append(f"volume={vol:.4f}")
command = [
@@ -334,8 +357,13 @@ def concat_main_audio(
"-ac",
"2",
]
af_simple: list[str] = []
if need_volume:
command.extend(["-af", f"volume={vol:.4f}"])
af_simple.append(f"volume={vol:.4f}")
if freeze_seconds > 0:
af_simple.append(f"apad=whole_dur={final_duration:.3f}")
if af_simple:
command.extend(["-af", ",".join(af_simple)])
if final_duration > 0:
command.extend(["-t", f"{final_duration:.3f}"])
command.append(str(output_path))
@@ -369,6 +397,10 @@ def concat_main_audio(
if abs(vol - 1.0) >= 1e-6:
audio_filters.append(f"volume={vol:.4f}")
# #1749 冻结补静音（调速后时长已变，apad 补齐到最终时长）
if freeze_seconds > 0:
audio_filters.append(f"apad=whole_dur={final_duration:.3f}")
# aformat 归一化：统一输出格式为 48000Hz + stereo + fltp
audio_filters.append("aformat=sample_rates=48000:channel_layouts=stereo:sample_fmts=fltp")
@@ -410,10 +442,13 @@ def concat_main_audio(
speed = getattr(clip, "playback_speed", 1.0) or 1.0
if not isinstance(speed, (int, float)) or speed <= 0:
speed = 1.0
# #1749 末帧冻结：素材内可用时长 = 目标段长 − freeze，apad 补静音
freeze_seconds = _clip_freeze_seconds(clip)
atrim_dur = max(0.0, effective_duration - freeze_seconds) if freeze_seconds > 0 else effective_duration
audio_filters: list[str] = []
if effective_duration > 0:
audio_filters.append(f"atrim=start={trim_start:.3f}:duration={effective_duration:.3f}")
audio_filters.append(f"atrim=start={trim_start:.3f}:duration={atrim_dur:.3f}")
audio_filters.append("asetpts=PTS-STARTPTS")
# 音频调速 — atempo 多级串联
@@ -435,6 +470,10 @@ def concat_main_audio(
if reverse_filter:
audio_filters.append(reverse_filter)
# #1749 冻结补静音：视频 tpad 延长后音频等长补齐（concat 时间轴对齐）
if freeze_seconds > 0:
audio_filters.append(f"apad=whole_dur={effective_duration:.3f}")
# 音量（0=静音，1=原声）
vol = _clip_volume(clip)
if abs(vol - 1.0) >= 1e-6:
@@ -506,12 +545,17 @@ def mix_with_independent_audio(
input_args.extend(["-i", str(clip.local_path)])
effective_duration = clip_effective_duration(clip)
trim_start = getattr(clip, "start_time", 0) or 0
freeze_seconds = _clip_freeze_seconds(clip)
atrim_dur = max(0.0, effective_duration - freeze_seconds) if freeze_seconds > 0 else effective_duration
clip_filters = []
if effective_duration > 0:
clip_filters.append(f"atrim=start={trim_start:.3f}:duration={effective_duration:.3f}")
clip_filters.append(f"atrim=start={trim_start:.3f}:duration={atrim_dur:.3f}")
clip_filters.append("asetpts=PTS-STARTPTS")
else:
clip_filters.append("asetpts=PTS-STARTPTS")
# #1749 冻结补静音（与视频 tpad 等长）
if freeze_seconds > 0:
clip_filters.append(f"apad=whole_dur={effective_duration:.3f}")
vol = _clip_volume(clip)
if abs(vol - 1.0) >= 1e-6:
clip_filters.append(f"volume={vol:.4f}")
apps/worker/video_processing/unified_render_service.py
+8 -2
View File
@@ -1297,10 +1297,16 @@ class UnifiedRenderService:
# trim
effective_duration = UnifiedRenderService._clip_effective_duration(clip)
freeze_seconds = float((clip.config or {}).get("_freeze_seconds", 0.0) or 0.0)
# #1749 冻结时素材内截取时长 = 目标段长 − freeze
trim_dur = max(0.0, effective_duration - freeze_seconds) if freeze_seconds > 0 else effective_duration
if effective_duration > 0:
filters.append(f"trim=duration={effective_duration}")
if trim_dur > 0:
filters.append(f"trim=duration={trim_dur}")
filters.append("setpts=PTS-STARTPTS")
# #1749 末帧冻结（禁慢放）
if freeze_seconds > 0:
filters.append(f"tpad=stop_mode=clone:stop_duration={freeze_seconds:.3f}")
# 调速 — 与 filter_complex 路径一致
speed = UnifiedRenderService._clip_speed(clip)
apps/worker/worker_app/tasks/generation.py
+6
View File
@@ -535,11 +535,17 @@ def _reselect_plan_for_batch_retry(task_id: str, plan_id: str, task_info: dict)
try:
svc = EditPlanService(db)
asset_pool = list(task_info.get("task_asset_ids") or [])
_voice_dur = 0.0
try:
_voice_dur = float(task_info.get("voice_duration", 0.0) or 0.0)
except (TypeError, ValueError):
_voice_dur = 0.0
new_plan = svc.reselect_plan_for_variant(
plan_id,
asset_pool,
created_by_user_id=task_info.get("user_id", ""),
name_suffix="重渲变体",
voice_duration=_voice_dur,
)
return new_plan.id
finally: