mirror of
https://github.com/morpheus65535/bazarr.git
synced 2024-11-15 12:14:51 +08:00
376 lines
16 KiB
Python
376 lines
16 KiB
Python
# -*- coding: utf-8 -*-
|
|
from .exceptions import ParserError, MalformedMKVError
|
|
from .parsers import ebml
|
|
from datetime import timedelta
|
|
import logging
|
|
|
|
|
|
__all__ = ['VIDEO_TRACK', 'AUDIO_TRACK', 'SUBTITLE_TRACK', 'MKV', 'Info', 'Track', 'VideoTrack',
|
|
'AudioTrack', 'SubtitleTrack', 'Tag', 'SimpleTag', 'Chapter']
|
|
logger = logging.getLogger(__name__)
|
|
|
|
|
|
# Track types
|
|
VIDEO_TRACK, AUDIO_TRACK, SUBTITLE_TRACK = 0x01, 0x02, 0x11
|
|
|
|
|
|
class MKV(object):
|
|
"""Matroska Video file
|
|
|
|
:param stream: seekable file-like object
|
|
|
|
"""
|
|
def __init__(self, stream, recurse_seek_head=False):
|
|
# default attributes
|
|
self.info = None
|
|
self.video_tracks = []
|
|
self.audio_tracks = []
|
|
self.subtitle_tracks = []
|
|
self.chapters = []
|
|
self.tags = []
|
|
|
|
# keep track of the elements parsed
|
|
self.recurse_seek_head = recurse_seek_head
|
|
self._parsed_positions = set()
|
|
|
|
try:
|
|
# get the Segment element
|
|
logger.info('Reading Segment element')
|
|
specs = ebml.get_matroska_specs()
|
|
segments = ebml.parse(stream, specs, ignore_element_names=['EBML'], max_level=0)
|
|
if not segments:
|
|
raise MalformedMKVError('No Segment found')
|
|
if len(segments) > 1:
|
|
logger.warning('%d segments found, using the first one', len(segments))
|
|
segment = segments[0]
|
|
|
|
# get and recursively parse the SeekHead element
|
|
logger.info('Reading SeekHead element')
|
|
stream.seek(segment.position)
|
|
seek_head = ebml.parse_element(stream, specs)
|
|
if seek_head.name != 'SeekHead':
|
|
raise MalformedMKVError('No SeekHead found')
|
|
seek_head.load(stream, specs, ignore_element_names=['Void', 'CRC-32'])
|
|
self._parse_seekhead(seek_head, segment, stream, specs)
|
|
except ParserError as e:
|
|
raise MalformedMKVError('Parsing error: %s' % e)
|
|
|
|
def _parse_seekhead(self, seek_head, segment, stream, specs):
|
|
for seek in seek_head:
|
|
element_id = ebml.read_element_id(seek['SeekID'].data)
|
|
element_name = specs[element_id][1]
|
|
element_position = seek['SeekPosition'].data + segment.position
|
|
if element_position in self._parsed_positions:
|
|
logger.warning('Skipping already parsed %s element at position %d', element_name, element_position)
|
|
continue
|
|
if element_name == 'Info':
|
|
logger.info('Processing element %s from SeekHead at position %d', element_name, element_position)
|
|
element = self._load_element(stream, specs, element_position)
|
|
self.info = Info.fromelement(element)
|
|
elif element_name == 'Tracks':
|
|
logger.info('Processing element %s from SeekHead at position %d', element_name, element_position)
|
|
tracks = self._load_element(stream, specs, element_position)
|
|
self.video_tracks.extend([VideoTrack.fromelement(t) for t in tracks if t['TrackType'].data == VIDEO_TRACK])
|
|
self.audio_tracks.extend([AudioTrack.fromelement(t) for t in tracks if t['TrackType'].data == AUDIO_TRACK])
|
|
self.subtitle_tracks.extend([SubtitleTrack.fromelement(t) for t in tracks if t['TrackType'].data == SUBTITLE_TRACK])
|
|
elif element_name == 'Chapters':
|
|
logger.info('Processing element %s from SeekHead at position %d', element_name, element_position)
|
|
element = self._load_element(stream, specs, element_position)
|
|
self.chapters.extend([Chapter.fromelement(c) for c in element[0] if c.name == 'ChapterAtom'])
|
|
elif element_name == 'Tags':
|
|
logger.info('Processing element %s from SeekHead at position %d', element_name, element_position)
|
|
element = self._load_element(stream, specs, element_position)
|
|
self.tags.extend([Tag.fromelement(t) for t in element])
|
|
elif element_name == 'SeekHead' and self.recurse_seek_head:
|
|
logger.info('Processing element %s from SeekHead at position %d', element_name, element_position)
|
|
element = self._load_element(stream, specs, element_position)
|
|
self._parse_seekhead(element, segment, stream, specs)
|
|
else:
|
|
logger.debug('Element %s ignored', element_name)
|
|
self._parsed_positions.add(element_position)
|
|
|
|
def _load_element(self,stream, specs, position):
|
|
stream.seek(position)
|
|
element = ebml.parse_element(stream,specs)
|
|
element.load(stream, specs, ignore_element_names=['Void', 'CRC-32'])
|
|
return element
|
|
|
|
def get_srt_subtitles_track_by_language(self):
|
|
"""get a dictionary of the SRT subtitles track id's indexed by language"""
|
|
|
|
subtitles = dict()
|
|
for track in self.subtitle_tracks:
|
|
|
|
logger.info("Found subtitle language %s, with codec %s and lacing %s",
|
|
track.language,track.codec_id,track.lacing)
|
|
|
|
if not track.is_srt():
|
|
logger.debug("Ignoring subtitle language %s with codec %s",track.language,track.codec_id)
|
|
elif track.lacing:
|
|
logger.info("Ignoring subtitle language %s with lacing %s",track.language,track.lacing)
|
|
else:
|
|
subtitles[track.language] = track
|
|
|
|
return subtitles
|
|
|
|
def to_dict(self):
|
|
return {'info': self.info.__dict__, 'video_tracks': [t.__dict__ for t in self.video_tracks],
|
|
'audio_tracks': [t.__dict__ for t in self.audio_tracks], 'subtitle_tracks': [t.__dict__ for t in self.subtitle_tracks],
|
|
'chapters': [c.__dict__ for c in self.chapters], 'tags': [t.__dict__ for t in self.tags]}
|
|
|
|
def __repr__(self):
|
|
return '<%s [%r, %r, %r, %r]>' % (self.__class__.__name__, self.info, self.video_tracks, self.audio_tracks, self.subtitle_tracks)
|
|
|
|
|
|
class Info(object):
|
|
"""Object for the Info EBML element"""
|
|
def __init__(self, title=None, duration=None, date_utc=None, timecode_scale=None, muxing_app=None, writing_app=None):
|
|
self.title = title
|
|
self.timecode_scale = timecode_scale
|
|
self.duration = timedelta(microseconds=duration * (timecode_scale or 1000000) // 1000) if duration else None
|
|
self.date_utc = date_utc
|
|
self.muxing_app = muxing_app
|
|
self.writing_app = writing_app
|
|
|
|
@classmethod
|
|
def fromelement(cls, element):
|
|
"""Load the :class:`Info` from an :class:`~enzyme.parsers.ebml.Element`
|
|
|
|
:param element: the Info element
|
|
:type element: :class:`~enzyme.parsers.ebml.Element`
|
|
|
|
"""
|
|
title = element.get('Title')
|
|
duration = element.get('Duration')
|
|
date_utc = element.get('DateUTC')
|
|
timecode_scale = element.get('TimecodeScale',1000000)
|
|
muxing_app = element.get('MuxingApp')
|
|
writing_app = element.get('WritingApp')
|
|
return cls(title, duration, date_utc, timecode_scale, muxing_app, writing_app)
|
|
|
|
def __repr__(self):
|
|
return '<%s [title=%r, duration=%s, date=%s]>' % (self.__class__.__name__, self.title, self.duration, self.date_utc)
|
|
|
|
def __str__(self):
|
|
return repr(self.__dict__)
|
|
|
|
|
|
class Track(object):
|
|
"""Base object for the Tracks EBML element"""
|
|
def __init__(self, type=None, number=None, name=None, language=None, enabled=None, default=None, forced=None, lacing=None,
|
|
codec_id=None, codec_name=None):
|
|
self.type = type
|
|
self.number = number
|
|
self.name = name
|
|
self.language = language
|
|
self.enabled = enabled
|
|
self.default = default
|
|
self.forced = forced
|
|
self.lacing = lacing
|
|
self.codec_id = codec_id
|
|
self.codec_name = codec_name
|
|
|
|
@classmethod
|
|
def fromelement(cls, element):
|
|
"""Load the :class:`Track` from an :class:`~enzyme.parsers.ebml.Element`
|
|
|
|
:param element: the Track element
|
|
:type element: :class:`~enzyme.parsers.ebml.Element`
|
|
|
|
"""
|
|
type = element.get('TrackType')
|
|
number = element.get('TrackNumber', 0)
|
|
name = element.get('Name')
|
|
language = element.get('Language','eng')
|
|
enabled = bool(element.get('FlagEnabled', 1))
|
|
default = bool(element.get('FlagDefault', 1))
|
|
forced = bool(element.get('FlagForced', 0))
|
|
lacing = bool(element.get('FlagLacing', 1))
|
|
codec_id = element.get('CodecID')
|
|
codec_name = element.get('CodecName')
|
|
return cls(type=type, number=number, name=name, language=language, enabled=enabled, default=default,
|
|
forced=forced, lacing=lacing, codec_id=codec_id, codec_name=codec_name)
|
|
|
|
def __repr__(self):
|
|
return '<%s [%d, name=%r, language=%s]>' % (self.__class__.__name__, self.number, self.name, self.language)
|
|
|
|
def __str__(self):
|
|
return str(self.__dict__)
|
|
|
|
|
|
class VideoTrack(Track):
|
|
"""Object for the Tracks EBML element with :data:`VIDEO_TRACK` TrackType"""
|
|
def __init__(self, width=0, height=0, interlaced=False, stereo_mode=None, crop=None,
|
|
display_width=None, display_height=None, display_unit=None, aspect_ratio_type=None, **kwargs):
|
|
super(VideoTrack, self).__init__(**kwargs)
|
|
self.width = width
|
|
self.height = height
|
|
self.interlaced = interlaced
|
|
self.stereo_mode = stereo_mode
|
|
self.crop = crop
|
|
self.display_width = display_width
|
|
self.display_height = display_height
|
|
self.display_unit = display_unit
|
|
self.aspect_ratio_type = aspect_ratio_type
|
|
|
|
@classmethod
|
|
def fromelement(cls, element):
|
|
"""Load the :class:`VideoTrack` from an :class:`~enzyme.parsers.ebml.Element`
|
|
|
|
:param element: the Track element with :data:`VIDEO_TRACK` TrackType
|
|
:type element: :class:`~enzyme.parsers.ebml.Element`
|
|
|
|
"""
|
|
videotrack = super(VideoTrack, cls).fromelement(element)
|
|
videotrack.width = element['Video'].get('PixelWidth', 0)
|
|
videotrack.height = element['Video'].get('PixelHeight', 0)
|
|
videotrack.interlaced = bool(element['Video'].get('FlagInterlaced', False))
|
|
videotrack.stereo_mode = element['Video'].get('StereoMode')
|
|
videotrack.crop = {}
|
|
if 'PixelCropTop' in element['Video']:
|
|
videotrack.crop['top'] = element['Video']['PixelCropTop']
|
|
if 'PixelCropBottom' in element['Video']:
|
|
videotrack.crop['bottom'] = element['Video']['PixelCropBottom']
|
|
if 'PixelCropLeft' in element['Video']:
|
|
videotrack.crop['left'] = element['Video']['PixelCropLeft']
|
|
if 'PixelCropRight' in element['Video']:
|
|
videotrack.crop['right'] = element['Video']['PixelCropRight']
|
|
videotrack.display_width = element['Video'].get('DisplayWidth')
|
|
videotrack.display_height = element['Video'].get('DisplayHeight')
|
|
videotrack.display_unit = element['Video'].get('DisplayUnit')
|
|
videotrack.aspect_ratio_type = element['Video'].get('AspectRatioType')
|
|
return videotrack
|
|
|
|
def __repr__(self):
|
|
return '<%s [%d, %dx%d, %s, name=%r, language=%s]>' % (self.__class__.__name__, self.number, self.width, self.height,
|
|
self.codec_id, self.name, self.language)
|
|
|
|
def __str__(self):
|
|
return str(self.__dict__)
|
|
|
|
|
|
class AudioTrack(Track):
|
|
"""Object for the Tracks EBML element with :data:`AUDIO_TRACK` TrackType"""
|
|
def __init__(self, sampling_frequency=None, channels=None, output_sampling_frequency=None, bit_depth=None, **kwargs):
|
|
super(AudioTrack, self).__init__(**kwargs)
|
|
self.sampling_frequency = sampling_frequency
|
|
self.channels = channels
|
|
self.output_sampling_frequency = output_sampling_frequency
|
|
self.bit_depth = bit_depth
|
|
|
|
@classmethod
|
|
def fromelement(cls, element):
|
|
"""Load the :class:`AudioTrack` from an :class:`~enzyme.parsers.ebml.Element`
|
|
|
|
:param element: the Track element with :data:`AUDIO_TRACK` TrackType
|
|
:type element: :class:`~enzyme.parsers.ebml.Element`
|
|
|
|
"""
|
|
audiotrack = super(AudioTrack, cls).fromelement(element)
|
|
audiotrack.sampling_frequency = element['Audio'].get('SamplingFrequency', 8000.0)
|
|
audiotrack.channels = element['Audio'].get('Channels', 1)
|
|
audiotrack.output_sampling_frequency = element['Audio'].get('OutputSamplingFrequency')
|
|
audiotrack.bit_depth = element['Audio'].get('BitDepth')
|
|
return audiotrack
|
|
|
|
def __repr__(self):
|
|
return '<%s [%d, %d channel(s), %.0fHz, %s, name=%r, language=%s]>' % (self.__class__.__name__, self.number, self.channels,
|
|
self.sampling_frequency, self.codec_id, self.name, self.language)
|
|
|
|
|
|
class SubtitleTrack(Track):
|
|
"""Object for the Tracks EBML element with :data:`SUBTITLE_TRACK` TrackType"""
|
|
|
|
def is_srt(self):
|
|
return self.codec_id == 'S_TEXT/UTF8'
|
|
|
|
class Tag(object):
|
|
"""Object for the Tag EBML element"""
|
|
def __init__(self, targets=None, simpletags=None):
|
|
self.targets = targets if targets is not None else []
|
|
self.simpletags = simpletags if simpletags is not None else []
|
|
|
|
@classmethod
|
|
def fromelement(cls, element):
|
|
"""Load the :class:`Tag` from an :class:`~enzyme.parsers.ebml.Element`
|
|
|
|
:param element: the Tag element
|
|
:type element: :class:`~enzyme.parsers.ebml.Element`
|
|
|
|
"""
|
|
targets = element['Targets'] if 'Targets' in element else []
|
|
simpletags = [SimpleTag.fromelement(s) for s in element if s.name == 'SimpleTag']
|
|
return cls(targets, simpletags)
|
|
|
|
def __repr__(self):
|
|
return '<%s [targets=%r, simpletags=%r]>' % (self.__class__.__name__, self.targets, self.simpletags)
|
|
|
|
|
|
class SimpleTag(object):
|
|
"""Object for the SimpleTag EBML element"""
|
|
def __init__(self, name, language='und', default=True, string=None, binary=None):
|
|
self.name = name
|
|
self.language = language
|
|
self.default = default
|
|
self.string = string
|
|
self.binary = binary
|
|
|
|
@classmethod
|
|
def fromelement(cls, element):
|
|
"""Load the :class:`SimpleTag` from an :class:`~enzyme.parsers.ebml.Element`
|
|
|
|
:param element: the SimpleTag element
|
|
:type element: :class:`~enzyme.parsers.ebml.Element`
|
|
|
|
"""
|
|
name = element.get('TagName')
|
|
language = element.get('TagLanguage', 'und')
|
|
default = element.get('TagDefault', True)
|
|
string = element.get('TagString')
|
|
binary = element.get('TagBinary')
|
|
return cls(name, language, default, string, binary)
|
|
|
|
def __repr__(self):
|
|
return '<%s [%s, language=%s, default=%s, string=%s]>' % (self.__class__.__name__, self.name, self.language, self.default, self.string)
|
|
|
|
|
|
class Chapter(object):
|
|
"""Object for the ChapterAtom and ChapterDisplay EBML element
|
|
|
|
.. note::
|
|
For the sake of simplicity, it is assumed that the ChapterAtom element
|
|
has no more than 1 ChapterDisplay child element and informations it contains
|
|
are merged into the :class:`Chapter`
|
|
|
|
"""
|
|
def __init__(self, start, hidden=False, enabled=False, end=None, string=None, language=None):
|
|
self.start = start
|
|
self.hidden = hidden
|
|
self.enabled = enabled
|
|
self.end = end
|
|
self.string = string
|
|
self.language = language
|
|
|
|
@classmethod
|
|
def fromelement(cls, element):
|
|
"""Load the :class:`Chapter` from an :class:`~enzyme.parsers.ebml.Element`
|
|
|
|
:param element: the ChapterAtom element
|
|
:type element: :class:`~enzyme.parsers.ebml.Element`
|
|
|
|
"""
|
|
start = timedelta(microseconds=element.get('ChapterTimeStart') // 1000)
|
|
hidden = element.get('ChapterFlagHidden', False)
|
|
enabled = element.get('ChapterFlagEnabled', True)
|
|
end = element.get('ChapterTimeEnd')
|
|
chapterdisplays = [c for c in element if c.name == 'ChapterDisplay']
|
|
if len(chapterdisplays) > 1:
|
|
logger.warning('More than 1 (%d) ChapterDisplay element in the ChapterAtom, using the first one', len(chapterdisplays))
|
|
if chapterdisplays:
|
|
string = chapterdisplays[0].get('ChapString')
|
|
language = chapterdisplays[0].get('ChapLanguage')
|
|
return cls(start, hidden, enabled, end, string, language)
|
|
return cls(start, hidden, enabled, end)
|
|
|
|
def __repr__(self):
|
|
return '<%s [%s, enabled=%s]>' % (self.__class__.__name__, self.start, self.enabled)
|