def ffmpeg_mix(tracks, width=None, height=None, audio_only=False, scale=1):
"""Mixes all tracks, returning a list of output streams."""
audio_streams = []
video_streams = []
xstack_layout = []
# FFMPEG filters
for t in tracks:
stream = ffmpeg.input(t['path'])
if t.get('has_audio'):
audio = stream.audio
if t.get('alignment_analysis'):
audio = audio.align_audio(t['alignment_analysis'])
# Volume adjust
in_loudness = t['filters'].get('loudness', {}).get('L')
if in_loudness:
# From https://github.com/mltframework/mlt/blob/7da01504d6844412b6e26c03b7c98214a1730343/src/modules/plus/filter_loudness.c#L160-L169
target_db = t['filters']['loudness']['program']
delta_db = target_db - in_loudness
audio = audio.volume(dB=delta_db)
audio_streams.append(audio)
if t.get('has_video') and not audio_only:
video = stream.video
if t.get('alignment_analysis'):
video = video.align_video(t['alignment_analysis'])
video = video.scale(w=int(t['width'] * scale), h=int(t['height'] * scale))
video_streams.append(video)
xstack_layout.append((int(t['left'] * scale), int(t['top'] * scale)))
# background for xstack
if video_streams:
video_tracks = [t for t in tracks if t.get('has_video')]
w = int((width or max(t['left'] + t['width'] for t in video_tracks)) * scale)
h = int((height or max(t['top'] + t['height'] for t in video_tracks)) * scale)
background = ffmpeg.input(f'color=black:size={w}x{h}:duration=1:rate=30',
format='lavfi')
video_streams.insert(0, background)
xstack_layout.insert(0, (0, 0))
# Combine outputs and return all the streams
streams = []
if audio_streams:
streams.append(ffmpeg.amix(audio_streams, dropout_transition=1000)
# amix reduces volume of each track to `1/n`, so increase the
# result by a factor of `n` to get it back to normal. Hopefully
# we've already applied some volume normalization, so this
# shouldn't be too loud. AFAICT Shotcut doesn't do any kind of
# volume reduction when it mixes tracks, so this should be
# closer to what it will sound like in Shotcut anyways.
.volume(len(audio_streams)))
if video_streams:
streams.append(ffmpeg.xstack(video_streams, layout=xstack_layout))
return streams
def load_bitmap(filename, h):
(stdout, _) = (ffmpeg.input(filename, ss=10).scale(
w=-2, h=h).output('pipe:',
vframes=1,
format='image2pipe',
vcodec='png').run(capture_stdout=True))
return stdout
def write_aligned_video(in_file, out_file, analysis, cfg):
stream = ffmpeg.input(in_file)
# Video filters
if ffmpeg.probe(in_file).video:
video = stream.video
# Detect crop at 20 seconds into the video
crop = ffmpeg.run_cropdetect(in_file, ss=20)
if crop:
video = video.crop(**crop)
if cfg.get('resize'):
video = video.scale(**cfg['resize'])
video = video.align_video(analysis)
else:
video = None
# Audio filters
audio = stream.audio
audio = audio.align_audio(analysis)
if analysis.get('loudnorm'):
audio = audio.loudnorm(analysis['loudnorm'],
resample=cfg.get('samplerate'))
elif cfg.get('samplerate'):
audio = audio.resample(cfg.get('samplerate'))
# Output
output_args = {}
if cfg.get('framerate'):
output_args['r'] = cfg.get('framerate')
output_args['movflags'] = '+faststart' # allow re-encoding on the fly
output_args['ac'] = 1
streams = [audio, video] if video else [audio]
return ffmpeg.output(*streams, out_file,
**output_args).run(overwrite_output=True)
def extract_audio_to_file(in_file, out_file, cfg):
return (ffmpeg.input(in_file).audio.filter(
'aresample', cfg.get('samplerate', 48000),
first_pts=0).filter('asetpts', 'PTS-STARTPTS').output(
out_file,
acodec=cfg.get('codec', 'aac'),
audio_bitrate=cfg.get('bitrate', '128k'),
ac=1,
).run(overwrite_output=True))
def write_aligned_audio(in_file, out_file, analysis, cfg):
audio = ffmpeg.input(in_file).audio.align_audio(analysis)
if analysis.get('loudnorm'):
audio = audio.loudnorm(analysis['loudnorm'],
resample=cfg.get('samplerate', 48000))
return audio.output(out_file, ac=1).run(overwrite_output=True)