File size: 8,283 Bytes
5806e12 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 |
import json
import os
import re
from typing import List, Dict, Any, Callable, Tuple
def map_special_tokens_to_word_positions(text: str, word_list: List[Dict[str, Any]]) -> List[Dict[str, Any]]:
special_token_map: List[Dict[str, Any]] = []
for m in re.finditer(r'<[^>]*?>', text):
special_token_map.append({
"token": m.group(),
"char_start": m.start(), # index in original text
})
if not special_token_map:
return []
visible_offset_map = {}
visible_idx = 0
i = 0
while i < len(text):
if text[i] == '<':
j = text.find('>', i) + 1
i = j
continue
visible_offset_map[i] = visible_idx
visible_idx += 1
i += 1
clean_text = re.sub(r'<[^>]*?>', '', text)
# locate each word in clean_text
word_positions = []
cur = 0
for w in word_list:
pos = clean_text.find(w["word"], cur)
if pos != -1:
word_positions.append({
"word": w["word"],
"start": pos,
"end": pos + len(w["word"])
})
cur = pos + len(w["word"])
# map each token
for sp in special_token_map:
# how many visible chars are before this token?
raw_idx = sp["char_start"]
visible_before = 0
# find largest key <= raw_idx in visible_offset_map
keys = [k for k in visible_offset_map.keys() if k < raw_idx]
if keys:
visible_before = visible_offset_map[max(keys)] + 1 # +1 because map stores idx of char at k
insert_after = -1
for i, wp in enumerate(word_positions):
if visible_before >= wp["end"]:
insert_after = i
else:
break
sp["insert_after_word_idx"] = insert_after
return special_token_map
def reorganize_transcription_c_unit(
session_id: str,
segment_func: Callable[[str], List[int]],
base_dir: str = "session_data",
device: str = "cuda"
) -> Tuple[int, int]:
"""Segment utterances into C-units with rules:
1. Boundaries inside <REPSTART>…<REPEND> or <REVSTART>…<REVEND> are ignored.
2. Trailing <PAUSE> <REVSTART> <REPSTART> moves to next C-unit prefix.
Returns (total_cunit_count, ignored_boundary_count).
"""
session_dir = os.path.join(base_dir, session_id)
input_file = os.path.join(session_dir, "transcription.json")
output_file = os.path.join(session_dir, "transcription_cunit.json")
if not os.path.exists(input_file):
raise FileNotFoundError(input_file)
with open(input_file, "r", encoding="utf-8") as f:
data = json.load(f)
# Handle both old and new format
if "segments" in data:
transcription_data = data["segments"]
else:
transcription_data = data
cunit_data: List[Dict[str, Any]] = []
ignored_boundary_count = 0
for utt in transcription_data:
original_text = utt["text"]
words_meta = utt.get("words", [])
clean_text = re.sub(r'<[^>]*?>', '', original_text).strip()
if not clean_text:
continue
# build word list
if words_meta:
word_data = [w for w in words_meta if w["word"] not in {"?", ",", ".", "!"}]
word_texts = [w["word"] for w in word_data]
else:
word_texts = re.sub(r'[\?\.,!]', '', clean_text).split()
word_data = [{"word": w, "start": utt["start"], "end": utt["end"]} for w in word_texts]
if not word_texts:
continue
# token positions & special ranges
special_token_map = map_special_tokens_to_word_positions(original_text, word_data)
rep_ranges, rev_ranges = _build_special_ranges(special_token_map)
def inside_special(idx: int) -> bool:
return any(s <= idx <= e for s, e in rep_ranges) or any(s <= idx <= e for s, e in rev_ranges)
# segmentation labels
labels = segment_func(' '.join(word_texts))
if len(labels) != len(word_texts):
raise ValueError(
f"Segmentation length mismatch: {len(word_texts)} words vs {len(labels)} labels"
)
current_words: List[str] = []
current_meta: List[Dict[str, Any]] = []
cunit_start_idx = 0 # global word idx of first word in current c‑unit
cunit_start_time = word_data[0]["start"]
carry_over_tokens: List[str] = []
for i, (word, label) in enumerate(zip(word_texts, labels)):
current_words.append(word)
current_meta.append(word_data[i])
is_last_word = i == len(word_texts) - 1
boundary_from_model = label == 1 and not inside_special(i)
if label == 1 and inside_special(i):
ignored_boundary_count += 1
make_boundary = boundary_from_model or is_last_word
if not make_boundary:
continue
# -------- assemble C‑unit --------
text_parts: List[str] = []
# 2a. prefix: carried‑over <PAUSE>
if carry_over_tokens:
text_parts.extend(carry_over_tokens)
carry_over_tokens = []
for j, w in enumerate(current_words):
global_word_idx = cunit_start_idx + j
# sentence‑initial tokens & ‑1 insertion
if global_word_idx == 0:
text_parts.extend(
[sp["token"] for sp in special_token_map if sp["insert_after_word_idx"] == -1]
)
text_parts.append(w)
# tokens that follow this word
text_parts.extend(
[sp["token"] for sp in special_token_map if sp["insert_after_word_idx"] == global_word_idx]
)
# 2b. move trailing <PAUSE> to next c‑unit
while text_parts and text_parts[-1].upper() == '<PAUSE>':
carry_over_tokens.insert(0, text_parts.pop())
# 2c. move trailing <REPSTART> or <REVSTART> to next c‑unit
while text_parts and text_parts[-1].upper() in {'<REPSTART>', '<REVSTART>'}:
carry_over_tokens.insert(0, text_parts.pop())
# Create text_token (with special tokens) and text (only words)
text_token = ' '.join(text_parts)
text_words_only = ' '.join(current_words)
cunit_data.append({
"start": cunit_start_time,
"end": current_meta[-1]["end"],
"speaker": "", # Initialize as empty
"text_token": text_token,
"text": text_words_only,
"words": [
{
"word": word["word"],
"start": word["start"],
"end": word["end"]
} for word in current_meta
]
})
# reset for next C‑unit
cunit_start_idx = i + 1
current_words, current_meta = [], []
if cunit_start_idx < len(word_data):
cunit_start_time = word_data[cunit_start_idx]["start"]
# Wrap in segments structure to match original format
output_data = {
"segments": cunit_data
}
with open(output_file, "w", encoding="utf-8") as f:
json.dump(output_data, f, indent=2, ensure_ascii=False)
print(f"C-unit segmentation done → {output_file}")
return len(cunit_data), ignored_boundary_count
def _build_special_ranges(special_token_map: List[Dict[str, Any]]):
rep_ranges, rev_ranges = [], []
rep_start, rev_start = None, None
for sp in special_token_map:
tok = sp["token"].upper()
idx = sp["insert_after_word_idx"]
if tok == '<REPSTART>':
rep_start = idx + 1
elif tok == '<REPEND>' and rep_start is not None:
rep_ranges.append((rep_start, idx))
rep_start = None
elif tok == '<REVSTART>':
rev_start = idx + 1
elif tok == '<REVEND>' and rev_start is not None:
rev_ranges.append((rev_start, idx))
rev_start = None
return rep_ranges, rev_ranges
|