Update pipeline/postprocess.py
Browse files- pipeline/postprocess.py +33 -14
pipeline/postprocess.py
CHANGED
|
@@ -1,10 +1,10 @@
|
|
| 1 |
-
# ─── Post-Processing Engine
|
| 2 |
|
| 3 |
-
MIN_DISPLAY =
|
| 4 |
-
MAX_DISPLAY = 7.0
|
| 5 |
-
MERGE_MAX_CHARS = 72
|
| 6 |
-
|
| 7 |
-
|
| 8 |
|
| 9 |
|
| 10 |
def format_timestamp(seconds: float) -> str:
|
|
@@ -16,6 +16,7 @@ def format_timestamp(seconds: float) -> str:
|
|
| 16 |
|
| 17 |
|
| 18 |
def merge_segments(segments: list) -> list:
|
|
|
|
| 19 |
if not segments:
|
| 20 |
return []
|
| 21 |
|
|
@@ -36,9 +37,8 @@ def merge_segments(segments: list) -> list:
|
|
| 36 |
combined_text = current["text"] + " " + seg["text"].strip()
|
| 37 |
combined_duration = seg["end"] - current["start"]
|
| 38 |
ends_sentence = current["text"].rstrip().endswith((".", "!", "?"))
|
| 39 |
-
|
| 40 |
-
|
| 41 |
-
is_short = current_duration < 1.5
|
| 42 |
|
| 43 |
should_merge = (
|
| 44 |
not ends_sentence
|
|
@@ -107,22 +107,34 @@ def apply_min_display(segments: list) -> list:
|
|
| 107 |
return result
|
| 108 |
|
| 109 |
|
| 110 |
-
def split_long_segment(seg: dict
|
|
|
|
|
|
|
|
|
|
|
|
|
| 111 |
words = seg.get("words", [])
|
| 112 |
text = seg["text"].strip()
|
| 113 |
|
| 114 |
-
#
|
| 115 |
-
if
|
| 116 |
return [seg]
|
| 117 |
|
| 118 |
-
|
|
|
|
|
|
|
| 119 |
best_gap = -1
|
|
|
|
|
|
|
| 120 |
for i in range(1, len(words)):
|
| 121 |
gap = words[i]["start"] - words[i - 1]["end"]
|
| 122 |
if gap > best_gap:
|
| 123 |
best_gap = gap
|
| 124 |
best_idx = i
|
| 125 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 126 |
words1 = words[:best_idx]
|
| 127 |
words2 = words[best_idx:]
|
| 128 |
|
|
@@ -132,6 +144,12 @@ def split_long_segment(seg: dict, depth: int = 0) -> list:
|
|
| 132 |
start2 = words2[0]["start"]
|
| 133 |
end1 = max(start2 - 0.080, words1[-1]["end"])
|
| 134 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 135 |
seg1 = {
|
| 136 |
"text": " ".join(w["word"].strip() for w in words1),
|
| 137 |
"start": words1[0]["start"],
|
|
@@ -145,7 +163,8 @@ def split_long_segment(seg: dict, depth: int = 0) -> list:
|
|
| 145 |
"words": words2,
|
| 146 |
}
|
| 147 |
|
| 148 |
-
|
|
|
|
| 149 |
|
| 150 |
|
| 151 |
def build_srt(segments: list) -> str:
|
|
|
|
| 1 |
+
# ─── Post-Processing Engine v7 ─────────────────────────────────────────────────
|
| 2 |
|
| 3 |
+
MIN_DISPLAY = 1.5 # حداقل زمان نمایش (ثانیه)
|
| 4 |
+
MAX_DISPLAY = 7.0 # حداکثر زمان نمایش
|
| 5 |
+
MERGE_MAX_CHARS = 72 # حداکثر کاراکتر برای merge
|
| 6 |
+
MAX_GAP = 1.5 # حداکثر فاصله بین دو segment برای merge
|
| 7 |
+
MAX_LINE_CHARS = 42 # استاندارد Netflix — حداکثر کاراکتر در هر خط
|
| 8 |
|
| 9 |
|
| 10 |
def format_timestamp(seconds: float) -> str:
|
|
|
|
| 16 |
|
| 17 |
|
| 18 |
def merge_segments(segments: list) -> list:
|
| 19 |
+
"""ادغام segmentهای کوتاه به جملات کامل"""
|
| 20 |
if not segments:
|
| 21 |
return []
|
| 22 |
|
|
|
|
| 37 |
combined_text = current["text"] + " " + seg["text"].strip()
|
| 38 |
combined_duration = seg["end"] - current["start"]
|
| 39 |
ends_sentence = current["text"].rstrip().endswith((".", "!", "?"))
|
| 40 |
+
current_duration = current["end"] - current["start"]
|
| 41 |
+
is_short = current_duration < 1.5
|
|
|
|
| 42 |
|
| 43 |
should_merge = (
|
| 44 |
not ends_sentence
|
|
|
|
| 107 |
return result
|
| 108 |
|
| 109 |
|
| 110 |
+
def split_long_segment(seg: dict) -> list:
|
| 111 |
+
"""
|
| 112 |
+
اگه segment بیشتر از MAX_LINE_CHARS کاراکتر داشت،
|
| 113 |
+
تبدیل میشه به دو subtitle جداگانه با timestamp مجزا.
|
| 114 |
+
"""
|
| 115 |
words = seg.get("words", [])
|
| 116 |
text = seg["text"].strip()
|
| 117 |
|
| 118 |
+
# اگه کوتاهه یا کلمه کافی نداره، نشکن
|
| 119 |
+
if len(text) <= MAX_LINE_CHARS or len(words) < 4:
|
| 120 |
return [seg]
|
| 121 |
|
| 122 |
+
# پیدا کردن بهترین نقطه split
|
| 123 |
+
mid = len(words) // 2
|
| 124 |
+
best_idx = mid
|
| 125 |
best_gap = -1
|
| 126 |
+
|
| 127 |
+
# اول دنبال بزرگترین gap بگرد
|
| 128 |
for i in range(1, len(words)):
|
| 129 |
gap = words[i]["start"] - words[i - 1]["end"]
|
| 130 |
if gap > best_gap:
|
| 131 |
best_gap = gap
|
| 132 |
best_idx = i
|
| 133 |
|
| 134 |
+
# اگه gap خیلی کوچیکه (زیر ۵۰ms)، از وسط بشکن
|
| 135 |
+
if best_gap < 0.050:
|
| 136 |
+
best_idx = mid
|
| 137 |
+
|
| 138 |
words1 = words[:best_idx]
|
| 139 |
words2 = words[best_idx:]
|
| 140 |
|
|
|
|
| 144 |
start2 = words2[0]["start"]
|
| 145 |
end1 = max(start2 - 0.080, words1[-1]["end"])
|
| 146 |
|
| 147 |
+
# safety: حداقل MIN_DISPLAY برای هر بخش
|
| 148 |
+
if end1 - words1[0]["start"] < MIN_DISPLAY:
|
| 149 |
+
end1 = words1[0]["start"] + MIN_DISPLAY
|
| 150 |
+
if end1 >= start2:
|
| 151 |
+
end1 = start2 - 0.040
|
| 152 |
+
|
| 153 |
seg1 = {
|
| 154 |
"text": " ".join(w["word"].strip() for w in words1),
|
| 155 |
"start": words1[0]["start"],
|
|
|
|
| 163 |
"words": words2,
|
| 164 |
}
|
| 165 |
|
| 166 |
+
# recursive — اگه هنوز طولانیه، دوباره بشکن
|
| 167 |
+
return split_long_segment(seg1) + split_long_segment(seg2)
|
| 168 |
|
| 169 |
|
| 170 |
def build_srt(segments: list) -> str:
|