mirror of
https://github.com/ynput/ayon-core.git
synced 2025-12-24 21:04:40 +01:00
Global: finalizing otio audio extractor
This commit is contained in:
parent
b0c72f38bf
commit
6f39b0f046
1 changed files with 68 additions and 38 deletions
|
|
@ -16,9 +16,9 @@ class ExtractOtioAudioTracks(pyblish.api.ContextPlugin):
|
|||
asset length and add it to each marked instance data representation. This
|
||||
is influenced by instance data audio attribute """
|
||||
|
||||
order = pyblish.api.CollectorOrder - 0.571
|
||||
order = pyblish.api.ExtractorOrder - 0.44
|
||||
label = "Extract OTIO Audio Tracks"
|
||||
hosts = ["hiero"]
|
||||
hosts = ["hiero", "resolve"]
|
||||
|
||||
# FFmpeg tools paths
|
||||
ffmpeg_path = get_ffmpeg_tool_path("ffmpeg")
|
||||
|
|
@ -29,6 +29,13 @@ class ExtractOtioAudioTracks(pyblish.api.ContextPlugin):
|
|||
Args:
|
||||
context (pyblish.Context): context of publisher
|
||||
"""
|
||||
# split the long audio file to peces devided by isntances
|
||||
audio_instances = self.get_audio_instances(context)
|
||||
self.log.debug("Audio instances: {}".format(len(audio_instances)))
|
||||
|
||||
if len(audio_instances) < 1:
|
||||
self.log.info("No audio instances available")
|
||||
return
|
||||
|
||||
# get sequence
|
||||
otio_timeline = context.data["otioTimeline"]
|
||||
|
|
@ -59,56 +66,77 @@ class ExtractOtioAudioTracks(pyblish.api.ContextPlugin):
|
|||
# remove empty
|
||||
os.remove(empty["mediaPath"])
|
||||
|
||||
# split the long audio file to peces devided by isntances
|
||||
audio_instances = self.get_audio_instances(context)
|
||||
|
||||
# cut instance framerange and add to representations
|
||||
self.create_audio_representations(audio_temp_fpath, audio_instances)
|
||||
self.add_audio_to_instances(audio_temp_fpath, audio_instances)
|
||||
|
||||
def create_audio_representations(self, audio_file, instances):
|
||||
# remove full mixed audio file
|
||||
os.remove(audio_temp_fpath)
|
||||
|
||||
def add_audio_to_instances(self, audio_file, instances):
|
||||
created_files = []
|
||||
for inst in instances:
|
||||
# create empty representation attr
|
||||
if "representations" not in inst.data:
|
||||
inst.data["representations"] = []
|
||||
name = inst.data["asset"]
|
||||
|
||||
name = inst.data["name"]
|
||||
recycling_file = [f for f in created_files if name in f]
|
||||
|
||||
# frameranges
|
||||
timeline_in_h = inst.data["clipInH"]
|
||||
timeline_out_h = inst.data["clipOutH"]
|
||||
fps = inst.data["fps"]
|
||||
|
||||
# seconds
|
||||
# create duration
|
||||
duration = (timeline_out_h - timeline_in_h) + 1
|
||||
start_sec = float(timeline_in_h / fps)
|
||||
duration_sec = float(duration / fps)
|
||||
|
||||
# temp audio file
|
||||
audio_fpath = self.create_temp_file(name)
|
||||
# ffmpeg generate new file only if doesnt exists already
|
||||
if not recycling_file:
|
||||
# convert to seconds
|
||||
start_sec = float(timeline_in_h / fps)
|
||||
duration_sec = float(duration / fps)
|
||||
|
||||
cmd = " ".join([
|
||||
self.ffmpeg_path,
|
||||
"-ss {}".format(start_sec),
|
||||
"-t {}".format(duration_sec),
|
||||
"-i {}".format(audio_file),
|
||||
audio_fpath
|
||||
])
|
||||
# temp audio file
|
||||
audio_fpath = self.create_temp_file(name)
|
||||
|
||||
# run subprocess
|
||||
self.log.debug("Executing: {}".format(cmd))
|
||||
openpype.api.run_subprocess(
|
||||
cmd, shell=True, logger=self.log
|
||||
)
|
||||
cmd = " ".join([
|
||||
self.ffmpeg_path,
|
||||
"-ss {}".format(start_sec),
|
||||
"-t {}".format(duration_sec),
|
||||
"-i {}".format(audio_file),
|
||||
audio_fpath
|
||||
])
|
||||
|
||||
# add to representations
|
||||
inst.data["representations"].append({
|
||||
"files": os.path.basename(audio_fpath),
|
||||
"name": "wav",
|
||||
"ext": "wav",
|
||||
"stagingDir": os.path.dirname(audio_fpath),
|
||||
"frameStart": 0,
|
||||
"frameEnd": duration
|
||||
})
|
||||
# run subprocess
|
||||
self.log.debug("Executing: {}".format(cmd))
|
||||
openpype.api.run_subprocess(
|
||||
cmd, shell=True, logger=self.log
|
||||
)
|
||||
else:
|
||||
audio_fpath = recycling_file.pop()
|
||||
|
||||
if "audio" in (inst.data["families"] + [inst.data["family"]]):
|
||||
# create empty representation attr
|
||||
if "representations" not in inst.data:
|
||||
inst.data["representations"] = []
|
||||
# add to representations
|
||||
inst.data["representations"].append({
|
||||
"files": os.path.basename(audio_fpath),
|
||||
"name": "wav",
|
||||
"ext": "wav",
|
||||
"stagingDir": os.path.dirname(audio_fpath),
|
||||
"frameStart": 0,
|
||||
"frameEnd": duration
|
||||
})
|
||||
|
||||
elif "reviewAudio" in inst.data.keys():
|
||||
audio_attr = inst.data.get("audio") or []
|
||||
audio_attr.append({
|
||||
"filename": audio_fpath,
|
||||
"offset": 0
|
||||
})
|
||||
inst.data["audio"] = audio_attr
|
||||
|
||||
# add generated audio file to created files for recycling
|
||||
if audio_fpath not in created_files:
|
||||
created_files.append(audio_fpath)
|
||||
|
||||
def get_audio_instances(self, context):
|
||||
"""Return only instances which are having audio in families
|
||||
|
|
@ -121,7 +149,9 @@ class ExtractOtioAudioTracks(pyblish.api.ContextPlugin):
|
|||
"""
|
||||
return [
|
||||
_i for _i in context
|
||||
if bool("audio" in _i.data.get("families", []))
|
||||
if bool("audio" in (
|
||||
_i.data.get("families", []) + [_i.data["family"]])
|
||||
) or _i.data.get("reviewAudio")
|
||||
]
|
||||
|
||||
def get_audio_track_items(self, otio_timeline):
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue