feat: BGM音轨混音能力(音量/淡入淡出/人声闪避/预设BGM库) (#291)
CI/CD Pipeline / Integration Tests (push) Has been cancelled
CI/CD Pipeline / Build & Push Staging (Watchtower auto-deploy) (push) Has been cancelled
CI/CD Pipeline / Staging E2E Tests (push) Has been cancelled
CI/CD Pipeline / Staging API Integration Tests (push) Has been cancelled
CI/CD Pipeline / Production Browser E2E (push) Failing after 1540h49m57s
CI/CD Pipeline / Build Production Runtime Images (push) Failing after 1540h50m2s
CI/CD Pipeline / Validate Code Quality And Tests (push) Has been skipped
CI/CD Pipeline / Unit Tests (push) Has been skipped
CI/CD Pipeline / Deploy Production (push) Failing after 1541h21m36s
CI/CD Pipeline / Frontend Lint (push) Has been skipped

This commit was merged in pull request #291.
This commit is contained in:
2026-07-14 10:37:24 +08:00
parent c840f37a44
commit 3cd26e98db
7 changed files with 1037 additions and 6 deletions
+313
View File
@@ -0,0 +1,313 @@
"""BGM 混音模块 — 背景音乐与主音频混合.
基于 FFmpeg 实现:
- BGM 音量调节
- 淡入淡出(afade)
- 循环播放(aloop,短 BGM 铺长视频)
- 人声闪避(sidechaincompress,有人声时BGM自动降低音量)
- amix 混音
作为 render_audio.py 的增强模块,在 mix_audio 后处理阶段被调用。
"""
from __future__ import annotations
import logging
from dataclasses import dataclass
from pathlib import Path
from typing import TYPE_CHECKING
from video_processing.ffmpeg_utils import FFMPEG_BIN, probe_duration, run_ffmpeg
if TYPE_CHECKING:
from video_processing.render_audio import RenderContext
logger = logging.getLogger(__name__)
@dataclass
class BGMConfig:
"""BGM 混音配置(内部使用,从 plan.config.bgm 转换而来)"""
bgm_path: str # BGM 本地文件路径
volume: float = 0.3 # 0.0 ~ 1.0
fade_in: float = 0.0 # 淡入时长(秒)
fade_out: float = 0.0 # 淡出时长(秒)
loop_enabled: bool = True # 是否循环铺满
sidechain_enabled: bool = False # 人声闪避
sidechain_ratio: float = 0.3 # 闪避时音量降低比例
sidechain_attack: float = 0.02 # 攻击时间
sidechain_release: float = 0.5 # 释放时间
sidechain_threshold: float = -25.0 # 触发阈值(dB)
@classmethod
def from_config_dict(cls, bgm_path: str, config: dict) -> "BGMConfig":
"""从 plan.config.bgm 字典创建 BGMConfig。"""
return cls(
bgm_path=bgm_path,
volume=float(config.get("volume", 0.3)),
fade_in=float(config.get("fade_in", 0.0)),
fade_out=float(config.get("fade_out", 0.0)),
loop_enabled=bool(config.get("loop_enabled", True)),
sidechain_enabled=bool(config.get("sidechain_enabled", False)),
sidechain_ratio=float(config.get("sidechain_ratio", 0.3)),
sidechain_attack=float(config.get("sidechain_attack", 0.02)),
sidechain_release=float(config.get("sidechain_release", 0.5)),
sidechain_threshold=float(config.get("sidechain_threshold", -25.0)),
)
# ── BGM 预处理 ────────────────────────────────────────────────────────────────
def prepare_bgm_track(
ctx: "RenderContext",
bgm: BGMConfig,
target_duration: float,
) -> Path:
"""预处理 BGM 轨道:循环/截断 + 音量 + 淡入淡出.
生成一个时长精确等于 target_duration 的 BGM 音频文件。
后续再与主音频混音。
Args:
ctx: 渲染上下文
bgm: BGM 配置
target_duration: 目标时长(秒),通常等于视频总时长
Returns:
处理后的 BGM 音频文件路径
"""
output_path = ctx.work_dir / f"bgm_processed_{ctx.plan_id}.aac"
if target_duration <= 0:
target_duration = 5.0 # 兜底
bgm_dur = probe_duration(bgm.bgm_path)
needs_loop = bgm.loop_enabled and bgm_dur > 0 and bgm_dur < target_duration * 0.9
# 构建滤镜链
filter_parts: list[str] = []
input_looped: bool = False
if needs_loop:
# 计算需要循环多少次才能铺满
loop_count = max(1, int(target_duration / bgm_dur) + 2)
# aloop 滤镜:循环指定次数
filter_parts.append(f"aloop=loop={loop_count}:size=0")
input_looped = True
# 音量调节
volume = max(0.0, min(1.0, bgm.volume))
if abs(volume - 1.0) > 0.001:
filter_parts.append(f"volume={volume:.3f}")
# 淡入
if bgm.fade_in > 0:
filter_parts.append(f"afade=t=in:st=0:d={bgm.fade_in:.3f}")
# 淡出(从 target_duration - fade_out 开始)
if bgm.fade_out > 0 and target_duration > bgm.fade_out:
fade_start = target_duration - bgm.fade_out
filter_parts.append(f"afade=t=out:st={fade_start:.3f}:d={bgm.fade_out:.3f}")
# 最终截断到目标时长
filter_parts.append(f"atrim=0:{target_duration:.3f}")
filter_parts.append("asetpts=N/SR/TB") # 重置时间戳
filter_str = ",".join(filter_parts)
command = [
FFMPEG_BIN,
"-y",
"-i",
bgm.bgm_path,
"-filter:a",
filter_str,
"-c:a",
"aac",
"-b:a",
"128k",
str(output_path),
]
logger.info(
"[bgm] prepare BGM track: path=%s dur=%.2f target=%.2f loop=%s fade_in=%.2f fade_out=%.2f",
bgm.bgm_path[-40:],
bgm_dur,
target_duration,
needs_loop,
bgm.fade_in,
bgm.fade_out,
)
run_ffmpeg(command)
return output_path
# ── BGM + 主音频混音 ──────────────────────────────────────────────────────────
def mix_bgm_with_main(
ctx: "RenderContext",
main_audio_path: Path,
bgm: BGMConfig,
target_duration: float,
) -> Path:
"""将 BGM 与主音频混合.
两种模式:
1. 普通混音(sidechain 关闭):amix 两路音频
2. 人声闪避(sidechain 开启):用 sidechaincompress 让 BGM 跟随主音频音量自动调整
Args:
ctx: 渲染上下文
main_audio_path: 主音频文件路径(人声/原始音频)
bgm: BGM 配置
target_duration: 目标时长
Returns:
混音后的音频文件路径
"""
output_path = ctx.work_dir / f"audio_with_bgm_{ctx.plan_id}.aac"
# 先预处理 BGM 轨道(循环/音量/淡入淡出/截断)
bgm_processed = prepare_bgm_track(ctx, bgm, target_duration)
if not bgm.sidechain_enabled:
# 普通 amix 混音
_mix_simple(main_audio_path, bgm_processed, output_path)
else:
# sidechain 人声闪避混音
_mix_sidechain(main_audio_path, bgm_processed, output_path, bgm)
return output_path
def _mix_simple(main_path: Path, bgm_path: Path, output_path: Path) -> None:
"""简单 amix 混音:主音频 + BGM = 输出.
主音频权重 1.0,BGM 已经在预处理阶段调好了音量。
amix 会自动归一化,需要用 volume 补偿。
"""
# 使用 amix:inputs=2,duration=first(以主音频时长为准)
# 然后用 volume=2 补偿 amix 的衰减(2路输入每路平均乘0.5)
filter_complex = "[0:a][1:a]amix=inputs=2:duration=first:dropout_transition=0[outa];" "[outa]volume=2[final]"
command = [
FFMPEG_BIN,
"-y",
"-i",
str(main_path),
"-i",
str(bgm_path),
"-filter_complex",
filter_complex,
"-map",
"[final]",
"-c:a",
"aac",
"-b:a",
"128k",
str(output_path),
]
logger.info("[bgm] simple amix mix")
run_ffmpeg(command)
def _mix_sidechain(
main_path: Path,
bgm_path: Path,
output_path: Path,
bgm: BGMConfig,
) -> None:
"""sidechain 人声闪避混音.
原理:
- 主音频作为 sidechain 信号源
- BGM 轨道经过 sidechaincompress,根据主音频音量动态调整 BGM 音量
- 最后 amix 混音
FFmpeg sidechaincompress 参数:
- threshold: 触发阈值(dB),主音频超过此值时开始压缩
- ratio: 压缩比,越高压缩越狠
- attack: 攻击时间(秒)
- release: 释放时间(秒)
"""
# sidechain_ratio 表示闪避时 BGM 音量降低比例
# ratio = 1 / (1 - sidechain_ratio),但实际压缩比需要更精细调整
# 简化处理:把 ratio 映射到 2:1 ~ 10:1 范围
ratio = max(2.0, min(10.0, 1.0 / (1.0 - bgm.sidechain_ratio)))
filter_complex = (
# BGM 经过 sidechain 压缩,用主音频做触发
f"[1:a][0:a]sidechaincompress="
f"threshold={bgm.sidechain_threshold}dB:"
f"ratio={ratio:.1f}:"
f"attack={bgm.sidechain_attack:.3f}:"
f"release={bgm.sidechain_release:.3f}:"
f"knee=6[bgm_comp];"
# 主音频 + 压缩后的 BGM 混音
f"[0:a][bgm_comp]amix=inputs=2:duration=first:dropout_transition=0[outa];"
f"[outa]volume=1.5[final]" # 轻微补偿
)
command = [
FFMPEG_BIN,
"-y",
"-i",
str(main_path),
"-i",
str(bgm_path),
"-filter_complex",
filter_complex,
"-map",
"[final]",
"-c:a",
"aac",
"-b:a",
"128k",
str(output_path),
]
logger.info(
"[bgm] sidechain mix: threshold=%.1fdB ratio=%.1f attack=%.3f release=%.3f",
bgm.sidechain_threshold,
ratio,
bgm.sidechain_attack,
bgm.sidechain_release,
)
run_ffmpeg(command)
# ── 纯 BGM 模式(无主音频) ──────────────────────────────────────────────────
def build_bgm_only(
ctx: "RenderContext",
bgm: BGMConfig,
target_duration: float,
) -> Path:
"""只有 BGM、没有主音频时,直接生成 BGM 音频.
Args:
ctx: 渲染上下文
bgm: BGM 配置
target_duration: 目标时长
Returns:
BGM 音频文件路径
"""
output_path = ctx.work_dir / f"bgm_only_{ctx.plan_id}.aac"
if target_duration <= 0:
target_duration = 5.0
bgm_processed = prepare_bgm_track(ctx, bgm, target_duration)
# 直接复制
import shutil
shutil.copy2(bgm_processed, output_path)
return output_path
+33 -3
View File
@@ -70,6 +70,9 @@ def mix_audio(
ctx: RenderContext,
layers: list[RenderLayer],
video_duration: float,
*,
bgm_path: str | None = None,
bgm_config: dict | None = None,
) -> Path | None:
"""音频后处理混音.
@@ -79,11 +82,14 @@ def mix_audio(
3. 独立音频轨(audio role)用 amix 混入
4. 输出时长截断到 video_duration
5. 无音频流的 clip 会被自动跳过,避免 FFmpeg 引用 [i:a] 失败
6. 如果提供了 bgm_path,则额外混入 BGM(支持淡入淡出、循环、人声闪避)
Args:
ctx: 渲染上下文
layers: 图层列表
video_duration: 视频总时长(用于截断音频)
bgm_path: BGM 音频本地路径,为 None 时不混入 BGM
bgm_config: BGM 配置字典(volume/fade_in/fade_out/sidechain 等)
Returns:
混音后的音频文件路径,无音频时返回 None
@@ -116,6 +122,15 @@ def mix_audio(
audio_clips = [c for c in audio_clips if clip_has_audio(ctx, c)]
if not main_clips and not audio_clips:
# 没有主音频也没有独立音频 → 检查是否有 BGM
if bgm_path and bgm_config and bgm_config.get("enabled", False):
from video_processing.bgm_mixer import BGMConfig, build_bgm_only
bgm_cfg = BGMConfig.from_config_dict(bgm_path, bgm_config)
try:
return build_bgm_only(ctx, bgm_cfg, video_duration)
except Exception:
logger.exception("[bgm] 纯BGM生成失败: plan_id=%s", ctx.plan_id)
return None
# 构建音频处理命令
@@ -124,10 +139,25 @@ def mix_audio(
# 简单场景:只有主图层 + 无独立音频 → 直接从视频提取音频并拼接
if main_clips and not audio_clips:
concat_main_audio(ctx, main_clips, output_path, video_duration)
return output_path
else:
# 有独立音频轨 → amix 混音
mix_with_independent_audio(ctx, main_clips, audio_clips, output_path, video_duration)
# ── BGM 混音 ──
if bgm_path and bgm_config and bgm_config.get("enabled", False):
from video_processing.bgm_mixer import BGMConfig, mix_bgm_with_main
bgm_cfg = BGMConfig.from_config_dict(bgm_path, bgm_config)
bgm_output = ctx.work_dir / f"audio_with_bgm_{ctx.plan_id}.aac"
try:
# 这里 main_audio 就是 output_path,先有主音频再混 BGM
final_path = mix_bgm_with_main(ctx, output_path, bgm_cfg, video_duration)
return final_path
except Exception:
logger.exception("[bgm] BGM 混音失败,回退到无 BGM 音频: plan_id=%s", ctx.plan_id)
return output_path
# 有独立音频轨 → amix 混音
mix_with_independent_audio(ctx, main_clips, audio_clips, output_path, video_duration)
return output_path
@@ -165,6 +165,7 @@ class UnifiedRenderService:
output_fps: int = DEFAULT_FPS,
transition_duration: float = DEFAULT_TRANSITION_DURATION,
asr_service: Any = None, # ASRService 实例,用于自动生成字幕
bgm_path: str | None = None, # BGM 本地文件路径
):
self.plan = plan
self.clips = clips
@@ -175,6 +176,7 @@ class UnifiedRenderService:
self.output_fps = output_fps
self.transition_duration = transition_duration
self.asr_service = asr_service
self.bgm_path = bgm_path
def render(self) -> RenderResult:
"""执行渲染,返回 RenderResult.
@@ -288,9 +290,55 @@ class UnifiedRenderService:
if is_pass_through:
# 直通场景已在一次调用中完成视频+音频
has_audio = pass_through_has_audio
# 直通模式下也支持 BGM 混音:提取音频 → 混 BGM → 合并回视频
if self.bgm_path and pass_through_has_audio:
config = self.plan.config or {}
bgm_config = config.get("bgm", {}) or {}
if bgm_config.get("enabled", False):
ctx = RenderContext(work_dir=self.work_dir, plan_id=self.plan.id)
from video_processing.bgm_mixer import BGMConfig, mix_bgm_with_main
bgm_cfg = BGMConfig.from_config_dict(self.bgm_path, bgm_config)
# 从直通输出中提取音频
main_audio_path = self.work_dir / f"pass_through_audio_{self.plan.id}.aac"
extract_cmd = [
FFMPEG_BIN,
"-y",
"-i",
str(output_path),
"-vn",
"-acodec",
"aac",
"-b:a",
"128k",
str(main_audio_path),
]
try:
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)
# 合并回视频
bgm_output = self.work_dir / f"rendered_{self.plan.id}_bgm.mp4"
merge_audio_video(ctx, output_path, final_audio, bgm_output)
output_path = bgm_output
logger.info("[unified-render] pass-through BGM mix done: plan_id=%s", self.plan.id)
except Exception:
logger.exception(
"[unified-render] pass-through BGM mix failed, skipping: plan_id=%s", self.plan.id
)
else:
ctx = RenderContext(work_dir=self.work_dir, plan_id=self.plan.id)
audio_path = mix_audio(ctx, layers, video_duration)
config = self.plan.config or {}
bgm_config = config.get("bgm", {}) or {}
audio_path = mix_audio(
ctx,
layers,
video_duration,
bgm_path=self.bgm_path,
bgm_config=bgm_config,
)
t_audio_end = time.time()
audio_mix_ms = int((t_audio_end - t_audio_start) * 1000)
has_audio = audio_path is not None