mirror of
https://github.com/SickGear/SickGear.git
synced 2024-11-17 18:35:04 +00:00
163 lines
4.7 KiB
Python
163 lines
4.7 KiB
Python
|
"""
|
||
|
FLV video parser.
|
||
|
|
||
|
Documentation:
|
||
|
|
||
|
- FLV File format: http://osflash.org/flv
|
||
|
- libavformat from ffmpeg project
|
||
|
- flashticle: Python project to read Flash (SWF and FLV with AMF metadata)
|
||
|
http://undefined.org/python/#flashticle
|
||
|
|
||
|
Author: Victor Stinner
|
||
|
Creation date: 4 november 2006
|
||
|
"""
|
||
|
|
||
|
from hachoir.parser import Parser
|
||
|
from hachoir.field import (FieldSet,
|
||
|
UInt8, UInt24, UInt32, NullBits, NullBytes,
|
||
|
Bit, Bits, String, RawBytes, Enum)
|
||
|
from hachoir.core.endian import BIG_ENDIAN
|
||
|
from hachoir.parser.audio.mpeg_audio import Frame
|
||
|
from hachoir.parser.video.amf import AMFObject
|
||
|
from hachoir.core.tools import createDict
|
||
|
|
||
|
SAMPLING_RATE = {
|
||
|
0: (5512, "5.5 kHz"),
|
||
|
1: (11025, "11 kHz"),
|
||
|
2: (22050, "22.1 kHz"),
|
||
|
3: (44100, "44.1 kHz"),
|
||
|
}
|
||
|
SAMPLING_RATE_VALUE = createDict(SAMPLING_RATE, 0)
|
||
|
SAMPLING_RATE_TEXT = createDict(SAMPLING_RATE, 1)
|
||
|
|
||
|
AUDIO_CODEC_MP3 = 2
|
||
|
AUDIO_CODEC_NAME = {
|
||
|
0: "Uncompressed",
|
||
|
1: "ADPCM",
|
||
|
2: "MP3",
|
||
|
5: "Nellymoser 8kHz mono",
|
||
|
6: "Nellymoser",
|
||
|
}
|
||
|
|
||
|
VIDEO_CODEC_NAME = {
|
||
|
2: "Sorensen H.263",
|
||
|
3: "Screen video",
|
||
|
4: "On2 VP6",
|
||
|
}
|
||
|
|
||
|
FRAME_TYPE = {
|
||
|
1: "keyframe",
|
||
|
2: "inter frame",
|
||
|
3: "disposable inter frame",
|
||
|
}
|
||
|
|
||
|
|
||
|
class Header(FieldSet):
|
||
|
|
||
|
def createFields(self):
|
||
|
yield String(self, "signature", 3, "FLV format signature", charset="ASCII")
|
||
|
yield UInt8(self, "version")
|
||
|
|
||
|
yield NullBits(self, "reserved[]", 5)
|
||
|
yield Bit(self, "type_flags_audio")
|
||
|
yield NullBits(self, "reserved[]", 1)
|
||
|
yield Bit(self, "type_flags_video")
|
||
|
|
||
|
yield UInt32(self, "data_offset")
|
||
|
|
||
|
|
||
|
def parseAudio(parent, size):
|
||
|
yield Enum(Bits(parent, "codec", 4, "Audio codec"), AUDIO_CODEC_NAME)
|
||
|
yield Enum(Bits(parent, "sampling_rate", 2, "Sampling rate"), SAMPLING_RATE_TEXT)
|
||
|
yield Bit(parent, "is_16bit", "16-bit or 8-bit per sample")
|
||
|
yield Bit(parent, "is_stereo", "Stereo or mono channel")
|
||
|
|
||
|
size -= 1
|
||
|
if 0 < size:
|
||
|
if parent["codec"].value == AUDIO_CODEC_MP3:
|
||
|
yield Frame(parent, "music_data", size=size * 8)
|
||
|
else:
|
||
|
yield RawBytes(parent, "music_data", size)
|
||
|
|
||
|
|
||
|
def parseVideo(parent, size):
|
||
|
yield Enum(Bits(parent, "frame_type", 4, "Frame type"), FRAME_TYPE)
|
||
|
yield Enum(Bits(parent, "codec", 4, "Video codec"), VIDEO_CODEC_NAME)
|
||
|
if 1 < size:
|
||
|
yield RawBytes(parent, "data", size - 1)
|
||
|
|
||
|
|
||
|
def parseAMF(parent, size):
|
||
|
while parent.current_size < parent.size:
|
||
|
yield AMFObject(parent, "entry[]")
|
||
|
|
||
|
|
||
|
class Chunk(FieldSet):
|
||
|
tag_info = {
|
||
|
8: ("audio[]", parseAudio, ""),
|
||
|
9: ("video[]", parseVideo, ""),
|
||
|
18: ("metadata", parseAMF, ""),
|
||
|
}
|
||
|
|
||
|
def __init__(self, *args, **kw):
|
||
|
FieldSet.__init__(self, *args, **kw)
|
||
|
self._size = (11 + self["size"].value) * 8
|
||
|
tag = self["tag"].value
|
||
|
if tag in self.tag_info:
|
||
|
self._name, self.parser, self._description = self.tag_info[tag]
|
||
|
else:
|
||
|
self.parser = None
|
||
|
|
||
|
def createFields(self):
|
||
|
yield UInt8(self, "tag")
|
||
|
yield UInt24(self, "size", "Content size")
|
||
|
yield UInt24(self, "timestamp", "Timestamp in millisecond")
|
||
|
yield NullBytes(self, "reserved", 4)
|
||
|
size = self["size"].value
|
||
|
if size:
|
||
|
if self.parser:
|
||
|
yield from self.parser(self, size)
|
||
|
else:
|
||
|
yield RawBytes(self, "content", size)
|
||
|
|
||
|
def getSampleRate(self):
|
||
|
try:
|
||
|
return SAMPLING_RATE_VALUE[self["sampling_rate"].value]
|
||
|
except LookupError:
|
||
|
return None
|
||
|
|
||
|
|
||
|
class FlvFile(Parser):
|
||
|
PARSER_TAGS = {
|
||
|
"id": "flv",
|
||
|
"category": "video",
|
||
|
"file_ext": ("flv",),
|
||
|
"mime": ("video/x-flv",),
|
||
|
"min_size": 9 * 4,
|
||
|
"magic": (
|
||
|
# Signature, version=1, flags=5 (video+audio), header size=9
|
||
|
(b"FLV\1\x05\0\0\0\x09", 0),
|
||
|
# Signature, version=1, flags=5 (video), header size=9
|
||
|
(b"FLV\1\x01\0\0\0\x09", 0),
|
||
|
),
|
||
|
"description": "Macromedia Flash video"
|
||
|
}
|
||
|
endian = BIG_ENDIAN
|
||
|
|
||
|
def validate(self):
|
||
|
if self.stream.readBytes(0, 3) != b"FLV":
|
||
|
return "Wrong file signature"
|
||
|
if self["header/data_offset"].value != 9:
|
||
|
return "Unknown data offset in main header"
|
||
|
return True
|
||
|
|
||
|
def createFields(self):
|
||
|
yield Header(self, "header")
|
||
|
yield UInt32(self, "prev_size[]", "Size of previous chunk")
|
||
|
while not self.eof:
|
||
|
yield Chunk(self, "chunk[]")
|
||
|
yield UInt32(self, "prev_size[]", "Size of previous chunk")
|
||
|
|
||
|
def createDescription(self):
|
||
|
return "Macromedia Flash video version %s" % self["header/version"].value
|