bazarr/libs/ffsubsync/aligners.py

159 lines
5.6 KiB
Python

# -*- coding: utf-8 -*-
import logging
import math
from typing import List, Optional, Tuple, Type, Union
import numpy as np
from ffsubsync.golden_section_search import gss
from ffsubsync.sklearn_shim import Pipeline, TransformerMixin
logging.basicConfig(level=logging.INFO)
logger: logging.Logger = logging.getLogger(__name__)
MIN_FRAMERATE_RATIO = 0.9
MAX_FRAMERATE_RATIO = 1.1
class FailedToFindAlignmentException(Exception):
pass
class FFTAligner(TransformerMixin):
def __init__(self, max_offset_samples: Optional[int] = None) -> None:
self.max_offset_samples: Optional[int] = max_offset_samples
self.best_offset_: Optional[int] = None
self.best_score_: Optional[float] = None
self.get_score_: bool = False
def _eliminate_extreme_offsets_from_solutions(
self, convolve: np.ndarray, substring: np.ndarray
) -> np.ndarray:
convolve = np.copy(convolve)
if self.max_offset_samples is None:
return convolve
def _offset_to_index(offset):
return len(convolve) - 1 + offset - len(substring)
convolve[: _offset_to_index(-self.max_offset_samples)] = float("-inf")
convolve[_offset_to_index(self.max_offset_samples) :] = float("-inf")
return convolve
def _compute_argmax(self, convolve: np.ndarray, substring: np.ndarray) -> None:
best_idx = int(np.argmax(convolve))
self.best_offset_ = len(convolve) - 1 - best_idx - len(substring)
self.best_score_ = convolve[best_idx]
def fit(self, refstring, substring, get_score: bool = False) -> "FFTAligner":
refstring, substring = [
list(map(int, s)) if isinstance(s, str) else s
for s in [refstring, substring]
]
refstring, substring = map(
lambda s: 2 * np.array(s).astype(float) - 1, [refstring, substring]
)
total_bits = math.log(len(substring) + len(refstring), 2)
total_length = int(2 ** math.ceil(total_bits))
extra_zeros = total_length - len(substring) - len(refstring)
subft = np.fft.fft(np.append(np.zeros(extra_zeros + len(refstring)), substring))
refft = np.fft.fft(
np.flip(np.append(refstring, np.zeros(len(substring) + extra_zeros)), 0)
)
convolve = np.real(np.fft.ifft(subft * refft))
self._compute_argmax(
self._eliminate_extreme_offsets_from_solutions(convolve, substring),
substring,
)
self.get_score_ = get_score
return self
def transform(self, *_) -> Union[int, Tuple[float, int]]:
if self.get_score_:
return self.best_score_, self.best_offset_
else:
return self.best_offset_
class MaxScoreAligner(TransformerMixin):
def __init__(
self,
base_aligner: Union[FFTAligner, Type[FFTAligner]],
srtin: Optional[str] = None,
sample_rate=None,
max_offset_seconds=None,
) -> None:
self.srtin: Optional[str] = srtin
if sample_rate is None or max_offset_seconds is None:
self.max_offset_samples: Optional[int] = None
else:
self.max_offset_samples = abs(int(max_offset_seconds * sample_rate))
if isinstance(base_aligner, type):
self.base_aligner: FFTAligner = base_aligner(
max_offset_samples=self.max_offset_samples
)
else:
self.base_aligner = base_aligner
self.max_offset_seconds: Optional[int] = max_offset_seconds
self._scores: List[Tuple[Tuple[float, int], Pipeline]] = []
def fit_gss(self, refstring, subpipe_maker):
def opt_func(framerate_ratio, is_last_iter):
subpipe = subpipe_maker(framerate_ratio)
substring = subpipe.fit_transform(self.srtin)
score = self.base_aligner.fit_transform(
refstring, substring, get_score=True
)
logger.info(
"got score %.0f (offset %d) for ratio %.3f",
score[0],
score[1],
framerate_ratio,
)
if is_last_iter:
self._scores.append((score, subpipe))
return -score[0]
gss(opt_func, MIN_FRAMERATE_RATIO, MAX_FRAMERATE_RATIO)
return self
def fit(
self, refstring, subpipes: Union[Pipeline, List[Pipeline]]
) -> "MaxScoreAligner":
if not isinstance(subpipes, list):
subpipes = [subpipes]
for subpipe in subpipes:
if callable(subpipe):
self.fit_gss(refstring, subpipe)
continue
elif hasattr(subpipe, "transform"):
substring = subpipe.transform(self.srtin)
else:
substring = subpipe
self._scores.append(
(
self.base_aligner.fit_transform(
refstring, substring, get_score=True
),
subpipe,
)
)
return self
def transform(self, *_) -> Tuple[Tuple[float, float], Pipeline]:
scores = self._scores
if self.max_offset_samples is not None:
scores = list(
filter(lambda s: abs(s[0][1]) <= self.max_offset_samples, scores)
)
if len(scores) == 0:
raise FailedToFindAlignmentException(
"Synchronization failed; consider passing "
"--max-offset-seconds with a number larger than "
"{}".format(self.max_offset_seconds)
)
(score, offset), subpipe = max(scores, key=lambda x: x[0][0])
return (score, offset), subpipe