Update app.py

app.py

@@ -20,6 +20,35 @@ model = whisper.load_model(model_size)
 print(f"[Startup] 模型加载完成 ✅", flush=True)
 
 
+import re
+
+# 每种语言的默认分段字数（用户未显式传 max_chars 时生效）
+DEFAULT_MAX_CHARS = {"zh": 12, "en": 55}
+FALLBACK_MAX_CHARS = 40  # 其他语言兜底值
+
+# 需要从最终字幕文本中清除的标点符号（中英文标点、引号、括号等）
+# 特意不包含直引号撇号 ' ，避免破坏英文缩写（don't / it's）
+_PUNCT_CHARS = "，。！？、；：“”‘’「」『』（）()《》【】[]<>—–…‥·,.!?;:\""
+_PUNCT_PATTERN = re.compile("[" + re.escape(_PUNCT_CHARS) + "]")
+
+
+def strip_punctuation(text: str) -> str:
+    """去除字幕文本中的标点符号，仅做视觉清洗，不影响时间戳/断句逻辑"""
+    if not text:
+        return text
+    cleaned = _PUNCT_PATTERN.sub("", text)
+    cleaned = re.sub(r"[ \t]+", " ", cleaned).strip()
+    return cleaned
+
+
+def resolve_max_chars(user_max_chars: int, lang_code: str) -> int:
+    """决定实际使用的 max_chars：用户显式指定则用用户的，否则按语言给默认值"""
+    if user_max_chars and user_max_chars > 0:
+        return user_max_chars
+    base_lang = (lang_code or "").split("-")[0].lower()
+    return DEFAULT_MAX_CHARS.get(base_lang, FALLBACK_MAX_CHARS)
+
+
 def format_timestamp(seconds: float) -> str:
     hours = int(seconds // 3600)
     minutes = int((seconds % 3600) // 60)
@@ -60,6 +89,10 @@ def group_by_chars(result, max_chars: int):
     if buf_text:
         segments.append({"start": buf_start, "end": buf_end, "text": buf_text})
 
+    # 断句依赖标点判断已经用完了，这里统一清洗掉标点，不影响时间戳
+    for seg in segments:
+        seg["text"] = strip_punctuation(seg["text"])
+
     return segments
 
 
@@ -102,7 +135,7 @@ async def generate_subtitle(
     file: UploadFile = File(...),
     output_format: str = Form("srt"),
     language: Optional[str] = Form(None),
-    max_chars: int = Form(0)  # 0 = 不拆分，用 Whisper 原始分段
+    max_chars: int = Form(0)  # 0/未传 = 按语言自动使用默认值（zh=12, en=55, 其他=40）
 ):
     valid_formats = ["srt", "vtt", "txt", "json"]
     if output_format not in valid_formats:
@@ -116,12 +149,11 @@ async def generate_subtitle(
 
     file_mb = len(content) / 1024 / 1024
     print(f"[Transcribe] 文件: {file.filename} ({file_mb:.1f} MB)", flush=True)
-    print(f"[Transcribe] 格式: {output_format} | 语言: {language or '自动检测'} | max_chars: {max_chars or '不拆分'}", flush=True)
+    print(f"[Transcribe] 格式: {output_format} | 语言: {language or '自动检测'} | max_chars: {max_chars if max_chars > 0 else '未指定(将按语言自动选择)'}", flush=True)
 
     try:
-        transcribe_options = {"verbose": True}
-        if max_chars > 0:
-            transcribe_options["word_timestamps"] = True
+        # word_timestamps 永远开启：分组逻辑现在始终执行，不再有"完全不分组"的原始模式
+        transcribe_options = {"verbose": True, "word_timestamps": True}
         if language:
             transcribe_options["language"] = language
 
@@ -134,11 +166,12 @@ async def generate_subtitle(
         detected_lang = result.get("language", "unknown")
         print(f"[Transcribe] 完成 ✅ 耗时 {elapsed:.1f}s | 语言: {detected_lang} | 原始段数: {seg_count}", flush=True)
 
-        if max_chars > 0:
-            segments = group_by_chars(result, max_chars)
-            print(f"[Transcribe] 按字数拆分完成，共 {len(segments)} 段（max_chars={max_chars}）", flush=True)
-        else:
-            segments = result["segments"]
+        effective_max_chars = resolve_max_chars(max_chars, language or detected_lang)
+        source = "用户指定" if (max_chars and max_chars > 0) else f"按语言({language or detected_lang})自动默认"
+        segments = group_by_chars(result, effective_max_chars)
+        print(f"[Transcribe] 按字数拆分完成，共 {len(segments)} 段（max_chars={effective_max_chars}，{source}）", flush=True)
+
+        full_text = strip_punctuation(result["text"])
 
         if output_format == "srt":
             subtitle_text = generate_srt(segments)
@@ -149,14 +182,15 @@ async def generate_subtitle(
             media_type = "text/plain"
             filename = "subtitle.vtt"
         elif output_format == "txt":
-            subtitle_text = result["text"]
+            subtitle_text = full_text
             media_type = "text/plain"
             filename = "subtitle.txt"
         elif output_format == "json":
             subtitle_text = json.dumps({
-                "text": result["text"],
+                "text": full_text,
                 "segments": segments,
-                "language": result["language"]
+                "language": result["language"],
+                "max_chars_used": effective_max_chars
             }, ensure_ascii=False, indent=2)
             media_type = "application/json"
             filename = "subtitle.json"