Update Commit

requirements.txt

@@ -1,4 +1,5 @@
 beautifulsoup4>=4.12
+deep-translator>=1.11
 lxml>=5.0
 requests>=2.30
 urllib3>=2.0

translate_epub.py

@@ -10,7 +10,7 @@ def _ipv4(h, p, f=0, t=0, pr=0, fl=0):
     return _orig(h, p, socket.AF_INET, t, pr, fl)
 socket.getaddrinfo = _ipv4
 
-import os, sys, time, glob, json, random
+import os, sys, time, glob, json, random, re
 import zipfile, tempfile, shutil
 from bs4 import BeautifulSoup
 from deep_translator import GoogleTranslator
@@ -30,6 +30,7 @@ PROXY_PASS = 'anakmanis'
 SEP = '[[SPLIT]]'
 MAX_CHUNK = 4000
 DELAY = 0.3
+SENTENCE_RE = re.compile(r'(?<=[.!?])\s+')
 
 BASE_DIR = os.path.dirname(os.path.abspath(__file__))
 SOURCE_DIR = os.path.join(BASE_DIR, 'original')
@@ -147,31 +148,57 @@ def process_epub(epub_path, output_path, state_info=None):
 
                 all_texts = [e.get_text().strip() for e in elems]
 
+                # Split each element's text into sentences
+                elem_sentences = []  # list of (elem_idx, [sentence1, sentence2, ...])
+                for ei, t in enumerate(all_texts):
+                    if not t:
+                        elem_sentences.append((ei, []))
+                    elif len(t) <= MAX_CHUNK:
+                        elem_sentences.append((ei, [t]))
+                    else:
+                        parts = SENTENCE_RE.split(t)
+                        elem_sentences.append((ei, parts))
+
+                # Flatten sentences into a list, tracking which element each belongs to
+                flat = []  # (elem_idx, sentence_idx_in_elem, text)
+                for ei, sentences in elem_sentences:
+                    for si, s in enumerate(sentences):
+                        flat.append((ei, si, s.strip()))
+
+                # Build batches of sentences up to MAX_CHUNK
                 batches = []
                 cur, curlen = [], 0
-                for t in all_texts:
-                    t2 = t.strip()
-                    if not t2:
-                        cur.append('')
+                for ei, si, s in flat:
+                    if not s:
                         continue
-                    if curlen + len(t2) + len(SEP) > MAX_CHUNK and cur:
+                    if curlen + len(s) + len(SEP) > MAX_CHUNK and cur:
                         batches.append(cur)
                         cur, curlen = [], 0
-                    cur.append(t2)
-                    curlen += len(t2) + len(SEP)
+                    cur.append((ei, si, s))
+                    curlen += len(s) + len(SEP)
                 if cur:
                     batches.append(cur)
 
-                results = [None] * len(all_texts)
-                idx = 0
+                # Translate sentence batches and reassemble per element
+                elem_translated_sentences = {}
                 for batch in batches:
-                    translated = translate_batch(batch)
-                    for j, trans in enumerate(translated):
-                        results[idx + j] = trans
-                    idx += len(batch)
+                    texts = [s for _, _, s in batch]
+                    translated = translate_batch(texts)
+                    for (ei, si, _), trans in zip(batch, translated):
+                        elem_translated_sentences.setdefault(ei, {})[si] = trans
                     batch_total += 1
                     time.sleep(DELAY)
 
+                # Reconstruct per-element text
+                results = []
+                for ei, (_, sentences) in enumerate(elem_sentences):
+                    if ei in elem_translated_sentences and sentences:
+                        trans_dict = elem_translated_sentences[ei]
+                        parts = [trans_dict.get(si, s) for si, s in enumerate(sentences)]
+                        results.append(' '.join(parts))
+                    else:
+                        results.append(all_texts[ei])
+
                 changed = False
                 for e, orig, trans in zip(elems, all_texts, results):
                     trans = trans.strip() if trans else ''