def make_reference_pipe(args: argparse.Namespace) -> Pipeline:
ref_format = _ref_format(args.reference)
if ref_format in SUBTITLE_EXTENSIONS:
if args.vad is not None:
logger.warning("Vad specified, but reference was not a movie")
return cast(
Pipeline,
make_subtitle_speech_pipeline(
fmt=ref_format,
**override(args,
encoding=args.reference_encoding
or DEFAULT_ENCODING),
),
)
elif ref_format in ("npy", "npz"):
if args.vad is not None:
logger.warning("Vad specified, but reference was not a movie")
return Pipeline([("deserialize",
DeserializeSpeechTransformer(args.non_speech_label))
])
else:
vad = args.vad or DEFAULT_VAD
if args.reference_encoding is not None:
logger.warning(
"Reference srt encoding specified, but reference was a video file"
)
ref_stream = args.reference_stream
if ref_stream is not None and not ref_stream.startswith("0:"):
ref_stream = "0:" + ref_stream
return Pipeline([
(
"speech_extract",
VideoSpeechTransformer(
vad=vad,
sample_rate=SAMPLE_RATE,
frame_rate=args.frame_rate,
non_speech_label=args.non_speech_label,
start_seconds=args.start_seconds,
ffmpeg_path=args.ffmpeg_path,
ref_stream=ref_stream,
vlc_mode=args.vlc_mode,
gui_mode=args.gui_mode,
),
),
])
def subpipe_maker(framerate_ratio):
return Pipeline([
('parse', parser),
('scale', SubtitleScaler(framerate_ratio)),
('speech_extract', SubtitleSpeechTransformer(
sample_rate=SAMPLE_RATE,
start_seconds=start_seconds,
framerate_ratio=framerate_ratio,
))
])
def subpipe_maker(framerate_ratio):
return Pipeline(
[
("parse", parser),
("scale", SubtitleScaler(framerate_ratio)),
(
"speech_extract",
SubtitleSpeechTransformer(
sample_rate=SAMPLE_RATE,
start_seconds=start_seconds,
framerate_ratio=framerate_ratio,
),
),
]
)
def try_sync(args: argparse.Namespace, reference_pipe: Optional[Pipeline], result: Dict[str, Any]) -> bool:
sync_was_successful = True
exc = None
try:
logger.info('extracting speech segments from %s...',
'stdin' if not args.srtin else 'subtitles file(s) {}'.format(args.srtin))
if not args.srtin:
args.srtin = [None]
for srtin in args.srtin:
skip_sync = args.skip_sync or reference_pipe is None
skip_infer_framerate_ratio = args.skip_infer_framerate_ratio or reference_pipe is None
srtout = srtin if args.overwrite_input else args.srtout
srt_pipe_maker = get_srt_pipe_maker(args, srtin)
framerate_ratios = get_framerate_ratios_to_try(args)
srt_pipes = [srt_pipe_maker(1.)] + [srt_pipe_maker(rat) for rat in framerate_ratios]
for srt_pipe in srt_pipes:
if callable(srt_pipe):
continue
else:
srt_pipe.fit(srtin)
if not skip_infer_framerate_ratio and hasattr(reference_pipe[-1], 'num_frames'):
inferred_framerate_ratio_from_length = float(reference_pipe[-1].num_frames) / cast(Pipeline, srt_pipes[0])[-1].num_frames
logger.info('inferred frameratio ratio: %.3f' % inferred_framerate_ratio_from_length)
srt_pipes.append(cast(Pipeline, srt_pipe_maker(inferred_framerate_ratio_from_length)).fit(srtin))
logger.info('...done')
logger.info('computing alignments...')
if skip_sync:
best_score = 0.
best_srt_pipe = cast(Pipeline, srt_pipes[0])
offset_samples = 0
else:
(best_score, offset_samples), best_srt_pipe = MaxScoreAligner(
FFTAligner, srtin, SAMPLE_RATE, args.max_offset_seconds
).fit_transform(
reference_pipe.transform(args.reference),
srt_pipes,
)
logger.info('...done')
offset_seconds = offset_samples / float(SAMPLE_RATE) + args.apply_offset_seconds
scale_step = best_srt_pipe.named_steps['scale']
logger.info('score: %.3f', best_score)
logger.info('offset seconds: %.3f', offset_seconds)
logger.info('framerate scale factor: %.3f', scale_step.scale_factor)
output_steps: List[Tuple[str, TransformerMixin]] = [('shift', SubtitleShifter(offset_seconds))]
if args.merge_with_reference:
output_steps.append(
('merge', SubtitleMerger(reference_pipe.named_steps['parse'].subs_))
)
output_pipe = Pipeline(output_steps)
out_subs = output_pipe.fit_transform(scale_step.subs_)
if args.output_encoding != 'same':
out_subs = out_subs.set_encoding(args.output_encoding)
suppress_output_thresh = args.suppress_output_if_offset_less_than
if (
suppress_output_thresh is None
or (
scale_step.scale_factor == 1.0
and offset_seconds >= suppress_output_thresh
)
):
logger.info('writing output to {}'.format(srtout or 'stdout'))
out_subs.write_file(srtout)
else:
logger.warning('suppressing output because offset %s was less than suppression threshold %s',
offset_seconds, args.suppress_output_if_offset_less_than)
except FailedToFindAlignmentException as e:
sync_was_successful = False
logger.error(e)
except Exception as e:
exc = e
sync_was_successful = False
logger.error(e)
else:
result['offset_seconds'] = offset_seconds
result['framerate_scale_factor'] = scale_step.scale_factor
finally:
if exc is not None:
raise exc
result['sync_was_successful'] = sync_was_successful
return sync_was_successful
def try_sync(args: argparse.Namespace, reference_pipe: Optional[Pipeline],
result: Dict[str, Any]) -> bool:
sync_was_successful = True
exc = None
try:
logger.info(
"extracting speech segments from %s...",
"stdin"
if not args.srtin else "subtitles file(s) {}".format(args.srtin),
)
if not args.srtin:
args.srtin = [None]
for srtin in args.srtin:
skip_sync = args.skip_sync or reference_pipe is None
skip_infer_framerate_ratio = (args.skip_infer_framerate_ratio
or reference_pipe is None)
srtout = srtin if args.overwrite_input else args.srtout
srt_pipe_maker = get_srt_pipe_maker(args, srtin)
framerate_ratios = get_framerate_ratios_to_try(args)
srt_pipes = [srt_pipe_maker(1.0)
] + [srt_pipe_maker(rat) for rat in framerate_ratios]
for srt_pipe in srt_pipes:
if callable(srt_pipe):
continue
else:
srt_pipe.fit(srtin)
if not skip_infer_framerate_ratio and hasattr(
reference_pipe[-1], "num_frames"):
inferred_framerate_ratio_from_length = (
float(reference_pipe[-1].num_frames) /
cast(Pipeline, srt_pipes[0])[-1].num_frames)
logger.info("inferred frameratio ratio: %.3f" %
inferred_framerate_ratio_from_length)
srt_pipes.append(
cast(Pipeline,
srt_pipe_maker(
inferred_framerate_ratio_from_length)).fit(srtin))
logger.info("...done")
logger.info("computing alignments...")
if skip_sync:
best_score = 0.0
best_srt_pipe = cast(Pipeline, srt_pipes[0])
offset_samples = 0
else:
(best_score, offset_samples), best_srt_pipe = MaxScoreAligner(
FFTAligner, srtin, SAMPLE_RATE,
args.max_offset_seconds).fit_transform(
reference_pipe.transform(args.reference),
srt_pipes,
)
logger.info("...done")
offset_seconds = (offset_samples / float(SAMPLE_RATE) +
args.apply_offset_seconds)
scale_step = best_srt_pipe.named_steps["scale"]
logger.info("score: %.3f", best_score)
logger.info("offset seconds: %.3f", offset_seconds)
logger.info("framerate scale factor: %.3f",
scale_step.scale_factor)
output_steps: List[Tuple[str, TransformerMixin]] = [
("shift", SubtitleShifter(offset_seconds))
]
if args.merge_with_reference:
output_steps.append(
("merge",
SubtitleMerger(
reference_pipe.named_steps["parse"].subs_)))
output_pipe = Pipeline(output_steps)
out_subs = output_pipe.fit_transform(scale_step.subs_)
if args.output_encoding != "same":
out_subs = out_subs.set_encoding(args.output_encoding)
suppress_output_thresh = args.suppress_output_if_offset_less_than
if suppress_output_thresh is None or (
scale_step.scale_factor == 1.0
and offset_seconds >= suppress_output_thresh):
logger.info("writing output to {}".format(srtout or "stdout"))
out_subs.write_file(srtout)
else:
logger.warning(
"suppressing output because offset %s was less than suppression threshold %s",
offset_seconds,
args.suppress_output_if_offset_less_than,
)
except FailedToFindAlignmentException:
sync_was_successful = False
logger.exception("failed to find alignment")
except Exception as e:
exc = e
sync_was_successful = False
else:
result["offset_seconds"] = offset_seconds
result["framerate_scale_factor"] = scale_step.scale_factor
finally:
if exc is not None:
raise exc
result["sync_was_successful"] = sync_was_successful
return sync_was_successful