Subtitle extraction from streaming media manifests #247

Authored by fstirlitz
Modified from: https://github.com/ytdl-org/youtube-dl/pull/6144

Closes: #73
Fixes:
https://github.com/ytdl-org/youtube-dl/issues/6106
https://github.com/ytdl-org/youtube-dl/issues/14977
https://github.com/ytdl-org/youtube-dl/issues/21438
https://github.com/ytdl-org/youtube-dl/issues/23609
https://github.com/ytdl-org/youtube-dl/issues/28132

Might also fix (untested):
https://github.com/ytdl-org/youtube-dl/issues/15424
https://github.com/ytdl-org/youtube-dl/issues/18267
https://github.com/ytdl-org/youtube-dl/issues/23899
https://github.com/ytdl-org/youtube-dl/issues/24375
https://github.com/ytdl-org/youtube-dl/issues/24595
https://github.com/ytdl-org/youtube-dl/issues/27899

Related:
https://github.com/ytdl-org/youtube-dl/issues/22379
https://github.com/ytdl-org/youtube-dl/pull/24517
https://github.com/ytdl-org/youtube-dl/pull/24886
https://github.com/ytdl-org/youtube-dl/pull/27215

Notes:
* The functions `extractor.common._extract_..._formats` are still kept for compatibility
* Only some extractors have currently been moved to using `_extract_..._formats_and_subtitles`
* Direct subtitle manifests (without a master) are not supported and are wrongly identified as containing video formats
* AES support is untested
* The fragmented TTML subtitles extracted from DASH/ISM are valid, but are unsupported by `ffmpeg` and most video players
    * Their XML fragments can be dumped using `ffmpeg -i in.mp4 -f data -map 0 -c copy out.ttml`.
        Once the unnecessary headers are stripped out of this, it becomes a valid self-contained ttml file
    * The ttml subs downloaded from DASH manifests can also be directly opened with <https://github.com/SubtitleEdit>
* Fragmented WebVTT files extracted from DASH/ISM are also unsupported by most tools
    * Unlike the ttml files, the XML fragments of these cannot be dumped using `ffmpeg`
    * The webtt subs extracted from DASH can be parsed by <https://github.com/gpac/gpac>
    * But validity of the those extracted from ISM are untested

1 files changed, 378 insertions, 0 deletions

diff --git a/yt_dlp/webvtt.py b/yt_dlp/webvtt.py
new file mode 100644
index 000000000..a184ee369
--- /dev/null
+++ b/yt_dlp/webvtt.py
@@ -0,0 +1,378 @@
+# coding: utf-8
+from __future__ import unicode_literals, print_function, division
+
+"""
+A partial parser for WebVTT segments. Interprets enough of the WebVTT stream
+to be able to assemble a single stand-alone subtitle file, suitably adjusting
+timestamps on the way, while everything else is passed through unmodified.
+
+Regular expressions based on the W3C WebVTT specification
+<https://www.w3.org/TR/webvtt1/>. The X-TIMESTAMP-MAP extension is described
+in RFC 8216 §3.5 <https://tools.ietf.org/html/rfc8216#section-3.5>.
+"""
+
+import re
+import io
+from .utils import int_or_none
+from .compat import (
+    compat_str as str,
+    compat_Pattern,
+    compat_Match,
+)
+
+
+class _MatchParser(object):
+    """
+    An object that maintains the current parsing position and allows
+    conveniently advancing it as syntax elements are successfully parsed.
+    """
+
+    def __init__(self, string):
+        self._data = string
+        self._pos = 0
+
+    def match(self, r):
+        if isinstance(r, compat_Pattern):
+            return r.match(self._data, self._pos)
+        if isinstance(r, str):
+            if self._data.startswith(r, self._pos):
+                return len(r)
+            return None
+        raise ValueError(r)
+
+    def advance(self, by):
+        if by is None:
+            amt = 0
+        elif isinstance(by, compat_Match):
+            amt = len(by.group(0))
+        elif isinstance(by, str):
+            amt = len(by)
+        elif isinstance(by, int):
+            amt = by
+        else:
+            raise ValueError(by)
+        self._pos += amt
+        return by
+
+    def consume(self, r):
+        return self.advance(self.match(r))
+
+    def child(self):
+        return _MatchChildParser(self)
+
+
+class _MatchChildParser(_MatchParser):
+    """
+    A child parser state, which advances through the same data as
+    its parent, but has an independent position. This is useful when
+    advancing through syntax elements we might later want to backtrack
+    from.
+    """
+
+    def __init__(self, parent):
+        super(_MatchChildParser, self).__init__(parent._data)
+        self.__parent = parent
+        self._pos = parent._pos
+
+    def commit(self):
+        """
+        Advance the parent state to the current position of this child state.
+        """
+        self.__parent._pos = self._pos
+        return self.__parent
+
+
+class ParseError(Exception):
+    def __init__(self, parser):
+        super(ParseError, self).__init__("Parse error at position %u (near %r)" % (
+            parser._pos, parser._data[parser._pos:parser._pos + 20]
+        ))
+
+
+_REGEX_TS = re.compile(r'''(?x)
+    (?:([0-9]{2,}):)?
+    ([0-9]{2}):
+    ([0-9]{2})\.
+    ([0-9]{3})?
+''')
+_REGEX_EOF = re.compile(r'\Z')
+_REGEX_NL = re.compile(r'(?:\r\n|[\r\n])')
+_REGEX_BLANK = re.compile(r'(?:\r\n|[\r\n])+')
+
+
+def _parse_ts(ts):
+    """
+    Convert a parsed WebVTT timestamp (a re.Match obtained from _REGEX_TS)
+    into an MPEG PES timestamp: a tick counter at 90 kHz resolution.
+    """
+
+    h, min, s, ms = ts.groups()
+    return 90 * (
+        int(h or 0) * 3600000 +  # noqa: W504,E221,E222
+        int(min)    *   60000 +  # noqa: W504,E221,E222
+        int(s)      *    1000 +  # noqa: W504,E221,E222
+        int(ms)                  # noqa: W504,E221,E222
+    )
+
+
+def _format_ts(ts):
+    """
+    Convert an MPEG PES timestamp into a WebVTT timestamp.
+    This will lose sub-millisecond precision.
+    """
+
+    ts = int((ts + 45) // 90)
+    ms , ts = divmod(ts, 1000)  # noqa: W504,E221,E222,E203
+    s  , ts = divmod(ts, 60)    # noqa: W504,E221,E222,E203
+    min, h  = divmod(ts, 60)    # noqa: W504,E221,E222
+    return '%02u:%02u:%02u.%03u' % (h, min, s, ms)
+
+
+class Block(object):
+    """
+    An abstract WebVTT block.
+    """
+
+    def __init__(self, **kwargs):
+        for key, val in kwargs.items():
+            setattr(self, key, val)
+
+    @classmethod
+    def parse(cls, parser):
+        m = parser.match(cls._REGEX)
+        if not m:
+            return None
+        parser.advance(m)
+        return cls(raw=m.group(0))
+
+    def write_into(self, stream):
+        stream.write(self.raw)
+
+
+class HeaderBlock(Block):
+    """
+    A WebVTT block that may only appear in the header part of the file,
+    i.e. before any cue blocks.
+    """
+
+    pass
+
+
+class Magic(HeaderBlock):
+    _REGEX = re.compile(r'\ufeff?WEBVTT([ \t][^\r\n]*)?(?:\r\n|[\r\n])')
+
+    # XXX: The X-TIMESTAMP-MAP extension is described in RFC 8216 §3.5
+    # <https://tools.ietf.org/html/rfc8216#section-3.5>, but the RFC
+    # doesn’t specify the exact grammar nor where in the WebVTT
+    # syntax it should be placed; the below has been devised based
+    # on usage in the wild
+    #
+    # And strictly speaking, the presence of this extension violates
+    # the W3C WebVTT spec. Oh well.
+
+    _REGEX_TSMAP = re.compile(r'X-TIMESTAMP-MAP=')
+    _REGEX_TSMAP_LOCAL = re.compile(r'LOCAL:')
+    _REGEX_TSMAP_MPEGTS = re.compile(r'MPEGTS:([0-9]+)')
+
+    @classmethod
+    def __parse_tsmap(cls, parser):
+        parser = parser.child()
+
+        while True:
+            m = parser.consume(cls._REGEX_TSMAP_LOCAL)
+            if m:
+                m = parser.consume(_REGEX_TS)
+                if m is None:
+                    raise ParseError(parser)
+                local = _parse_ts(m)
+                if local is None:
+                    raise ParseError(parser)
+            else:
+                m = parser.consume(cls._REGEX_TSMAP_MPEGTS)
+                if m:
+                    mpegts = int_or_none(m.group(1))
+                    if mpegts is None:
+                        raise ParseError(parser)
+                else:
+                    raise ParseError(parser)
+            if parser.consume(','):
+                continue
+            if parser.consume(_REGEX_NL):
+                break
+            raise ParseError(parser)
+
+        parser.commit()
+        return local, mpegts
+
+    @classmethod
+    def parse(cls, parser):
+        parser = parser.child()
+
+        m = parser.consume(cls._REGEX)
+        if not m:
+            raise ParseError(parser)
+
+        extra = m.group(1)
+        local, mpegts = None, None
+        if parser.consume(cls._REGEX_TSMAP):
+            local, mpegts = cls.__parse_tsmap(parser)
+        if not parser.consume(_REGEX_NL):
+            raise ParseError(parser)
+        parser.commit()
+        return cls(extra=extra, mpegts=mpegts, local=local)
+
+    def write_into(self, stream):
+        stream.write('WEBVTT')
+        if self.extra is not None:
+            stream.write(self.extra)
+        stream.write('\n')
+        if self.local or self.mpegts:
+            stream.write('X-TIMESTAMP-MAP=LOCAL:')
+            stream.write(_format_ts(self.local if self.local is not None else 0))
+            stream.write(',MPEGTS:')
+            stream.write(str(self.mpegts if self.mpegts is not None else 0))
+            stream.write('\n')
+        stream.write('\n')
+
+
+class StyleBlock(HeaderBlock):
+    _REGEX = re.compile(r'''(?x)
+        STYLE[\ \t]*(?:\r\n|[\r\n])
+        ((?:(?!-->)[^\r\n])+(?:\r\n|[\r\n]))*
+        (?:\r\n|[\r\n])
+    ''')
+
+
+class RegionBlock(HeaderBlock):
+    _REGEX = re.compile(r'''(?x)
+        REGION[\ \t]*
+        ((?:(?!-->)[^\r\n])+(?:\r\n|[\r\n]))*
+        (?:\r\n|[\r\n])
+    ''')
+
+
+class CommentBlock(Block):
+    _REGEX = re.compile(r'''(?x)
+        NOTE(?:\r\n|[\ \t\r\n])
+        ((?:(?!-->)[^\r\n])+(?:\r\n|[\r\n]))*
+        (?:\r\n|[\r\n])
+    ''')
+
+
+class CueBlock(Block):
+    """
+    A cue block. The payload is not interpreted.
+    """
+
+    _REGEX_ID = re.compile(r'((?:(?!-->)[^\r\n])+)(?:\r\n|[\r\n])')
+    _REGEX_ARROW = re.compile(r'[ \t]+-->[ \t]+')
+    _REGEX_SETTINGS = re.compile(r'[ \t]+((?:(?!-->)[^\r\n])+)')
+    _REGEX_PAYLOAD = re.compile(r'[^\r\n]+(?:\r\n|[\r\n])?')
+
+    @classmethod
+    def parse(cls, parser):
+        parser = parser.child()
+
+        id = None
+        m = parser.consume(cls._REGEX_ID)
+        if m:
+            id = m.group(1)
+
+        m0 = parser.consume(_REGEX_TS)
+        if not m0:
+            return None
+        if not parser.consume(cls._REGEX_ARROW):
+            return None
+        m1 = parser.consume(_REGEX_TS)
+        if not m1:
+            return None
+        m2 = parser.consume(cls._REGEX_SETTINGS)
+        if not parser.consume(_REGEX_NL):
+            return None
+
+        start = _parse_ts(m0)
+        end = _parse_ts(m1)
+        settings = m2.group(1) if m2 is not None else None
+
+        text = io.StringIO()
+        while True:
+            m = parser.consume(cls._REGEX_PAYLOAD)
+            if not m:
+                break
+            text.write(m.group(0))
+
+        parser.commit()
+        return cls(
+            id=id,
+            start=start, end=end, settings=settings,
+            text=text.getvalue()
+        )
+
+    def write_into(self, stream):
+        if self.id is not None:
+            stream.write(self.id)
+            stream.write('\n')
+        stream.write(_format_ts(self.start))
+        stream.write(' --> ')
+        stream.write(_format_ts(self.end))
+        if self.settings is not None:
+            stream.write(' ')
+            stream.write(self.settings)
+        stream.write('\n')
+        stream.write(self.text)
+        stream.write('\n')
+
+    @property
+    def as_json(self):
+        return {
+            'id': self.id,
+            'start': self.start,
+            'end': self.end,
+            'text': self.text,
+            'settings': self.settings,
+        }
+
+
+def parse_fragment(frag_content):
+    """
+    A generator that yields (partially) parsed WebVTT blocks when given
+    a bytes object containing the raw contents of a WebVTT file.
+    """
+
+    parser = _MatchParser(frag_content.decode('utf-8'))
+
+    yield Magic.parse(parser)
+
+    while not parser.match(_REGEX_EOF):
+        if parser.consume(_REGEX_BLANK):
+            continue
+
+        block = RegionBlock.parse(parser)
+        if block:
+            yield block
+            continue
+        block = StyleBlock.parse(parser)
+        if block:
+            yield block
+            continue
+        block = CommentBlock.parse(parser)
+        if block:
+            yield block  # XXX: or skip
+            continue
+
+        break
+
+    while not parser.match(_REGEX_EOF):
+        if parser.consume(_REGEX_BLANK):
+            continue
+
+        block = CommentBlock.parse(parser)
+        if block:
+            yield block  # XXX: or skip
+            continue
+        block = CueBlock.parse(parser)
+        if block:
+            yield block
+            continue
+
+        raise ParseError(parser)