Class PGSSpeechTransformer

ffsubsync/speech_transformers.py:1103–1208

Use PGS (Presentation Graphic Stream) subtitle timings as a sync reference. PGS subtitles are bitmap-based (e.g. Blu-ray / HDMV) and cannot be converted to text by ffmpeg, so they can't be fed through the normal subtitle pipeline. However, when muxed into an MKV the container still stor

class PGSSpeechTransformer(TransformerMixin, ComputeSpeechFrameBoundariesMixin):
    """Use PGS (Presentation Graphic Stream) subtitle timings as a sync reference.

    PGS subtitles are bitmap-based (e.g. Blu-ray / HDMV) and cannot be converted
    to text by ffmpeg, so they can't be fed through the normal subtitle pipeline.
    However, when muxed into an MKV the container still stores a presentation
    timestamp (``pts_time``) and ``duration_time`` for every subtitle packet, so
    we can recover *when* each caption is on screen without decoding the bitmaps
    or parsing the raw SUP/PCS binary at all.

    This transformer reads those per-packet timings via ``ffprobe`` (see
    :func:`_get_pgs_timings_via_ffprobe`), filtering out the tiny "clear" packets
    that carry no image, and builds the same kind of sparse binary speech signal
    that :class:`SubtitleSpeechTransformer` produces for text subtitles: 1.0 while
    a caption is displayed, 0.0 otherwise. That signal can then be aligned against
    the input subtitle file by the normal ffsubsync pipeline.

    The reference stream may be given explicitly via ``ref_stream`` (with or
    without a leading ``0:``), or left as ``None`` to auto-detect the first
    ``hdmv_pgs_subtitle`` track in the file.
    """

    # PGS is already in the MKV timebase so its duration cannot be compared
    # against the SRT to infer a framerate ratio.  Returning None here prevents
    # the duration-based framerate inference in try_sync from running.
    @property
    def num_frames(self) -> None:
        return None

    def __init__(
        self,
        sample_rate: int,
        start_seconds: int = 0,
        ffmpeg_path: Optional[str] = None,
        ref_stream: Optional[str] = None,
        gui_mode: bool = False,
    ) -> None:
        super(PGSSpeechTransformer, self).__init__()
        self.sample_rate: int = sample_rate
        self.start_seconds: int = start_seconds
        self.ffmpeg_path: Optional[str] = ffmpeg_path
        self.ref_stream: Optional[str] = ref_stream
        self.gui_mode: bool = gui_mode
        self.pgs_speech_results_: Optional[np.ndarray] = None

    def fit(self, fname: str, *_) -> "PGSSpeechTransformer":
        if self.ref_stream is None:
            stream = find_pgs_stream(fname, self.ffmpeg_path, self.gui_mode)
            if stream is None:
                raise ValueError(
                    "No hdmv_pgs_subtitle stream found in {}. "
                    "Specify one explicitly with --pgs-ref-stream.".format(fname)
                )
        else:
            stream = self.ref_stream
            if not stream.startswith("0:"):
                stream = "0:" + stream