mirror of
https://github.com/morpheus65535/bazarr
synced 2024-12-27 18:17:46 +00:00
186 lines
6 KiB
Python
186 lines
6 KiB
Python
#!/usr/bin/env python
|
|
# -*- coding: utf-8 -*-
|
|
"""
|
|
audio_codec, audio_profile and audio_channels property
|
|
"""
|
|
from rebulk.remodule import re
|
|
|
|
from rebulk import Rebulk, Rule, RemoveMatch
|
|
from ..common import dash
|
|
from ..common.validators import seps_before, seps_after
|
|
|
|
audio_properties = ['audio_codec', 'audio_profile', 'audio_channels']
|
|
|
|
|
|
def audio_codec():
|
|
"""
|
|
Builder for rebulk object.
|
|
:return: Created Rebulk object
|
|
:rtype: Rebulk
|
|
"""
|
|
rebulk = Rebulk().regex_defaults(flags=re.IGNORECASE, abbreviations=[dash]).string_defaults(ignore_case=True)
|
|
|
|
def audio_codec_priority(match1, match2):
|
|
"""
|
|
Gives priority to audio_codec
|
|
:param match1:
|
|
:type match1:
|
|
:param match2:
|
|
:type match2:
|
|
:return:
|
|
:rtype:
|
|
"""
|
|
if match1.name == 'audio_codec' and match2.name in ['audio_profile', 'audio_channels']:
|
|
return match2
|
|
if match1.name in ['audio_profile', 'audio_channels'] and match2.name == 'audio_codec':
|
|
return match1
|
|
return '__default__'
|
|
|
|
rebulk.defaults(name="audio_codec", conflict_solver=audio_codec_priority)
|
|
|
|
rebulk.regex("MP3", "LAME", r"LAME(?:\d)+-?(?:\d)+", value="MP3")
|
|
rebulk.regex('Dolby', 'DolbyDigital', 'Dolby-Digital', 'DD', 'AC3D?', value='AC3')
|
|
rebulk.regex("DolbyAtmos", "Dolby-Atmos", "Atmos", value="DolbyAtmos")
|
|
rebulk.string("AAC", value="AAC")
|
|
rebulk.string('EAC3', 'DDP', 'DD+', value="EAC3")
|
|
rebulk.string("Flac", value="FLAC")
|
|
rebulk.string("DTS", value="DTS")
|
|
rebulk.regex("True-?HD", value="TrueHD")
|
|
|
|
rebulk.defaults(name="audio_profile")
|
|
rebulk.string("HD", value="HD", tags="DTS")
|
|
rebulk.regex("HD-?MA", value="HDMA", tags="DTS")
|
|
rebulk.string("HE", value="HE", tags="AAC")
|
|
rebulk.string("LC", value="LC", tags="AAC")
|
|
rebulk.string("HQ", value="HQ", tags="AC3")
|
|
|
|
rebulk.defaults(name="audio_channels")
|
|
rebulk.regex(r'(7[\W_][01](?:ch)?)(?:[^\d]|$)', value='7.1', children=True)
|
|
rebulk.regex(r'(5[\W_][01](?:ch)?)(?:[^\d]|$)', value='5.1', children=True)
|
|
rebulk.regex(r'(2[\W_]0(?:ch)?)(?:[^\d]|$)', value='2.0', children=True)
|
|
rebulk.regex('7[01]', value='7.1', validator=seps_after, tags='weak-audio_channels')
|
|
rebulk.regex('5[01]', value='5.1', validator=seps_after, tags='weak-audio_channels')
|
|
rebulk.string('20', value='2.0', validator=seps_after, tags='weak-audio_channels')
|
|
rebulk.string('7ch', '8ch', value='7.1')
|
|
rebulk.string('5ch', '6ch', value='5.1')
|
|
rebulk.string('2ch', 'stereo', value='2.0')
|
|
rebulk.string('1ch', 'mono', value='1.0')
|
|
|
|
rebulk.rules(DtsRule, AacRule, Ac3Rule, AudioValidatorRule, HqConflictRule, AudioChannelsValidatorRule)
|
|
|
|
return rebulk
|
|
|
|
|
|
class AudioValidatorRule(Rule):
|
|
"""
|
|
Remove audio properties if not surrounded by separators and not next each others
|
|
"""
|
|
priority = 64
|
|
consequence = RemoveMatch
|
|
|
|
def when(self, matches, context):
|
|
ret = []
|
|
|
|
audio_list = matches.range(predicate=lambda match: match.name in audio_properties)
|
|
for audio in audio_list:
|
|
if not seps_before(audio):
|
|
valid_before = matches.range(audio.start - 1, audio.start,
|
|
lambda match: match.name in audio_properties)
|
|
if not valid_before:
|
|
ret.append(audio)
|
|
continue
|
|
if not seps_after(audio):
|
|
valid_after = matches.range(audio.end, audio.end + 1,
|
|
lambda match: match.name in audio_properties)
|
|
if not valid_after:
|
|
ret.append(audio)
|
|
continue
|
|
|
|
return ret
|
|
|
|
|
|
class AudioProfileRule(Rule):
|
|
"""
|
|
Abstract rule to validate audio profiles
|
|
"""
|
|
priority = 64
|
|
dependency = AudioValidatorRule
|
|
consequence = RemoveMatch
|
|
|
|
def __init__(self, codec):
|
|
super(AudioProfileRule, self).__init__()
|
|
self.codec = codec
|
|
|
|
def when(self, matches, context):
|
|
profile_list = matches.named('audio_profile', lambda match: self.codec in match.tags)
|
|
ret = []
|
|
for profile in profile_list:
|
|
codec = matches.previous(profile, lambda match: match.name == 'audio_codec' and match.value == self.codec)
|
|
if not codec:
|
|
codec = matches.next(profile, lambda match: match.name == 'audio_codec' and match.value == self.codec)
|
|
if not codec:
|
|
ret.append(profile)
|
|
return ret
|
|
|
|
|
|
class DtsRule(AudioProfileRule):
|
|
"""
|
|
Rule to validate DTS profile
|
|
"""
|
|
|
|
def __init__(self):
|
|
super(DtsRule, self).__init__("DTS")
|
|
|
|
|
|
class AacRule(AudioProfileRule):
|
|
"""
|
|
Rule to validate AAC profile
|
|
"""
|
|
|
|
def __init__(self):
|
|
super(AacRule, self).__init__("AAC")
|
|
|
|
|
|
class Ac3Rule(AudioProfileRule):
|
|
"""
|
|
Rule to validate AC3 profile
|
|
"""
|
|
|
|
def __init__(self):
|
|
super(Ac3Rule, self).__init__("AC3")
|
|
|
|
|
|
class HqConflictRule(Rule):
|
|
"""
|
|
Solve conflict between HQ from other property and from audio_profile.
|
|
"""
|
|
|
|
dependency = [DtsRule, AacRule, Ac3Rule]
|
|
consequence = RemoveMatch
|
|
|
|
def when(self, matches, context):
|
|
hq_audio = matches.named('audio_profile', lambda match: match.value == 'HQ')
|
|
hq_audio_spans = [match.span for match in hq_audio]
|
|
hq_other = matches.named('other', lambda match: match.span in hq_audio_spans)
|
|
|
|
if hq_other:
|
|
return hq_other
|
|
|
|
|
|
class AudioChannelsValidatorRule(Rule):
|
|
"""
|
|
Remove audio_channel if no audio codec as previous match.
|
|
"""
|
|
priority = 128
|
|
consequence = RemoveMatch
|
|
|
|
def when(self, matches, context):
|
|
ret = []
|
|
|
|
for audio_channel in matches.tagged('weak-audio_channels'):
|
|
valid_before = matches.range(audio_channel.start - 1, audio_channel.start,
|
|
lambda match: match.name == 'audio_codec')
|
|
if not valid_before:
|
|
ret.append(audio_channel)
|
|
|
|
return ret
|