feat: Implement AI script generation for audiobook projects

qwen3-tts-backend/core/audiobook_service.py

@@ -15,6 +15,9 @@ from db.models import AudiobookProject, AudiobookCharacter, User
 
 logger = logging.getLogger(__name__)
 
+_LINE_RE = re.compile(r'^【(.+?)】(.*)$')
+_EMO_RE = re.compile(r'（(开心|愤怒|悲伤|恐惧|厌恶|低沉|惊讶):([0-9.]+)）\s*$')
+
 # Cancellation events for batch operations, keyed by project_id
 _cancel_events: dict[int, asyncio.Event] = {}
 
@@ -161,6 +164,340 @@ def _split_into_chapters(text: str) -> list[str]:
     return chapters
 
 
+def parse_ai_script(script_text: str, char_map: dict) -> list[dict]:
+    results = []
+    for raw_line in script_text.splitlines():
+        line = raw_line.strip()
+        if not line:
+            continue
+        m = _LINE_RE.match(line)
+        if not m:
+            if results:
+                results[-1]["text"] = results[-1]["text"] + "　" + line
+            continue
+        speaker = m.group(1).strip()
+        content = m.group(2).strip()
+
+        emo_text = None
+        emo_alpha = None
+        emo_m = _EMO_RE.search(content)
+        if emo_m:
+            emo_text = emo_m.group(1)
+            try:
+                emo_alpha = float(emo_m.group(2))
+            except ValueError:
+                emo_alpha = None
+            content = content[:emo_m.start()].strip()
+
+        if content.startswith('"') and content.endswith('"'):
+            content = content[1:-1].strip()
+        elif content.startswith('"') and content.endswith('"'):
+            content = content[1:-1].strip()
+
+        character = speaker
+
+        results.append({
+            "character": character,
+            "text": content,
+            "emo_text": emo_text,
+            "emo_alpha": emo_alpha,
+        })
+    return results
+
+
+async def generate_ai_script(project_id: int, user: User, db: Session) -> None:
+    from core.database import SessionLocal
+
+    project = db.query(AudiobookProject).filter(AudiobookProject.id == project_id).first()
+    if not project or not project.script_config:
+        return
+
+    key = str(project_id)
+    ps.reset(key)
+    cfg = project.script_config
+
+    try:
+        crud.update_audiobook_project_status(db, project_id, "analyzing")
+        ps.append_line(key, f"[AI剧本] 项目「{project.title}」开始生成剧本")
+
+        llm = _get_llm_service(db)
+        _llm_model = crud.get_system_setting(db, "llm_model")
+        _user_id = user.id
+
+        def _log_usage(prompt_tokens: int, completion_tokens: int) -> None:
+            log_db = SessionLocal()
+            try:
+                crud.create_usage_log(log_db, _user_id, prompt_tokens, completion_tokens,
+                                      model=_llm_model, context="ai_script_generate")
+            finally:
+                log_db.close()
+
+        genre = cfg.get("genre", "")
+        subgenre = cfg.get("subgenre", "")
+        premise = cfg.get("premise", "")
+        style = cfg.get("style", "")
+        num_characters = cfg.get("num_characters", 5)
+        num_chapters = cfg.get("num_chapters", 8)
+
+        ps.append_line(key, f"\n[Step 1] 生成 {num_characters} 个角色...\n")
+        ps.append_line(key, "")
+
+        def on_token(token: str) -> None:
+            ps.append_token(key, token)
+
+        characters_data = await llm.generate_story_characters(
+            genre=genre, subgenre=subgenre, premise=premise, style=style,
+            num_characters=num_characters, usage_callback=_log_usage,
+        )
+
+        has_narrator = any(c.get("name") in ("narrator", "旁白") for c in characters_data)
+        if not has_narrator:
+            characters_data.insert(0, {
+                "name": "旁白",
+                "gender": "未知",
+                "description": "第三人称旁白叙述者",
+                "instruct": (
+                    "音色信息：浑厚醇厚的男性中低音，嗓音饱满有力，带有传统说书人的磁性与感染力\n"
+                    "身份背景：中国传统说书艺人，精通评书、章回小说叙述艺术，深谙故事节奏与听众心理\n"
+                    "年龄设定：中年男性，四五十岁，声音历经岁月沉淀，成熟稳重而不失活力\n"
+                    "外貌特征：面容沉稳，气度从容，台风大气，给人以可信赖的叙述者印象\n"
+                    "性格特质：沉稳睿智，叙事冷静客观，情到深处能引发共鸣，不动声色间娓娓道来\n"
+                    "叙事风格：语速适中偏慢，抑扬顿挫，擅长铺垫悬念，停顿恰到好处，语气庄重而生动，富有画面感"
+                )
+            })
+
+        ps.append_line(key, f"\n\n[完成] 角色列表：{', '.join(c.get('name', '') for c in characters_data)}")
+
+        crud.delete_audiobook_segments(db, project_id)
+        crud.delete_audiobook_characters(db, project_id)
+
+        backend_type = user.user_preferences.get("default_backend", "aliyun") if user.user_preferences else "aliyun"
+
+        for char_data in characters_data:
+            name = char_data.get("name", "旁白")
+            if name == "narrator":
+                name = "旁白"
+            instruct = char_data.get("instruct", "")
+            description = char_data.get("description", "")
+            gender = char_data.get("gender") or ("未知" if name == "旁白" else None)
+            try:
+                voice_design = crud.create_voice_design(
+                    db=db,
+                    user_id=user.id,
+                    name=f"[有声书] {project.title} - {name}",
+                    instruct=instruct,
+                    backend_type=backend_type,
+                    preview_text=description[:100] if description else None,
+                )
+                crud.create_audiobook_character(
+                    db=db,
+                    project_id=project_id,
+                    name=name,
+                    gender=gender,
+                    description=description,
+                    instruct=instruct,
+                    voice_design_id=voice_design.id,
+                )
+            except Exception as e:
+                logger.error(f"Failed to create char/voice for {name}: {e}")
+
+        crud.update_audiobook_project_status(db, project_id, "characters_ready")
+        ps.append_line(key, f"\n[状态] 角色创建完成，请确认角色后继续生成剧本")
+        ps.mark_done(key)
+
+        user_id = user.id
+
+        async def _generate_all_previews():
+            temp_db = SessionLocal()
+            try:
+                characters = crud.list_audiobook_characters(temp_db, project_id)
+                char_ids = [c.id for c in characters]
+            finally:
+                temp_db.close()
+            if not char_ids:
+                return
+            sem = asyncio.Semaphore(3)
+            async def _gen(char_id: int):
+                async with sem:
+                    local_db = SessionLocal()
+                    try:
+                        db_user = crud.get_user_by_id(local_db, user_id)
+                        await generate_character_preview(project_id, char_id, db_user, local_db)
+                    except Exception as e:
+                        logger.error(f"Background preview failed for char {char_id}: {e}")
+                    finally:
+                        local_db.close()
+            await asyncio.gather(*[_gen(cid) for cid in char_ids])
+
+        asyncio.create_task(_generate_all_previews())
+
+    except Exception as e:
+        logger.error(f"generate_ai_script failed for project {project_id}: {e}", exc_info=True)
+        ps.append_line(key, f"\n[错误] {e}")
+        ps.mark_done(key)
+        crud.update_audiobook_project_status(db, project_id, "error", error_message=str(e))
+
+
+async def generate_ai_script_chapters(project_id: int, user: User, db: Session) -> None:
+    from core.database import SessionLocal
+
+    project = db.query(AudiobookProject).filter(AudiobookProject.id == project_id).first()
+    if not project or not project.script_config:
+        return
+
+    key = str(project_id)
+    ps.reset(key)
+    cfg = project.script_config
+
+    try:
+        genre = cfg.get("genre", "")
+        subgenre = cfg.get("subgenre", "")
+        premise = cfg.get("premise", "")
+        style = cfg.get("style", "")
+        num_chapters = cfg.get("num_chapters", 8)
+
+        llm = _get_llm_service(db)
+        _llm_model = crud.get_system_setting(db, "llm_model")
+        _user_id = user.id
+
+        def _log_usage(prompt_tokens: int, completion_tokens: int) -> None:
+            log_db = SessionLocal()
+            try:
+                crud.create_usage_log(log_db, _user_id, prompt_tokens, completion_tokens,
+                                      model=_llm_model, context="ai_script_chapters")
+            finally:
+                log_db.close()
+
+        def on_token(token: str) -> None:
+            ps.append_token(key, token)
+
+        db_characters = crud.list_audiobook_characters(db, project_id)
+        characters_data = [
+            {"name": c.name, "gender": c.gender or "未知", "description": c.description or "", "instruct": c.instruct or ""}
+            for c in db_characters
+        ]
+        char_map = {c.name: c for c in db_characters}
+        backend_type = user.user_preferences.get("default_backend", "aliyun") if user.user_preferences else "aliyun"
+
+        ps.append_line(key, f"[AI剧本] 开始生成 {num_chapters} 章大纲...\n")
+        ps.append_line(key, "")
+
+        chapters_data = await llm.generate_chapter_outline(
+            genre=genre, subgenre=subgenre, premise=premise, style=style,
+            num_chapters=num_chapters, characters=characters_data, usage_callback=_log_usage,
+        )
+
+        ps.append_line(key, f"\n\n[完成] 大纲：{len(chapters_data)} 章")
+
+        crud.delete_audiobook_chapters(db, project_id)
+        crud.delete_audiobook_segments(db, project_id)
+
+        project_audio_dir = Path(settings.OUTPUT_DIR) / "audiobook" / str(project_id)
+        for subdir in ("segments", "chapters"):
+            d = project_audio_dir / subdir
+            if d.exists():
+                shutil.rmtree(d, ignore_errors=True)
+
+        for ch_data in chapters_data:
+            idx = ch_data.get("index", 0)
+            title = ch_data.get("title", f"第 {idx + 1} 章")
+            summary = ch_data.get("summary", "")
+            crud.create_audiobook_chapter(db, project_id, idx, summary, title=title)
+
+        crud.update_audiobook_project_status(db, project_id, "ready")
+
+        ps.append_line(key, f"\n[Step 2] 逐章生成对话脚本...\n")
+
+        for ch_data in chapters_data:
+            idx = ch_data.get("index", 0)
+            title = ch_data.get("title", f"第 {idx + 1} 章")
+            summary = ch_data.get("summary", "")
+
+            ps.append_line(key, f"\n第 {idx + 1} 章「{title}」→ ")
+            ps.append_line(key, "")
+
+            chapter_obj = crud.get_audiobook_chapter_by_index(db, project_id, idx)
+            if not chapter_obj:
+                continue
+
+            try:
+                script_text = await llm.generate_chapter_script(
+                    genre=genre, premise=premise,
+                    chapter_index=idx, chapter_title=title, chapter_summary=summary,
+                    characters=characters_data, on_token=on_token, usage_callback=_log_usage,
+                )
+
+                chapter_obj.source_text = script_text
+                db.commit()
+
+                segments_data = parse_ai_script(script_text, char_map)
+
+                unknown_speakers = {
+                    seg["character"] for seg in segments_data
+                    if seg["character"] != "旁白" and seg["character"] not in char_map
+                }
+                for speaker_name in sorted(unknown_speakers):
+                    try:
+                        npc_instruct = (
+                            "音色信息：普通自然的中性成年人声音，语调平和\n"
+                            "身份背景：故事中的路人或配角\n"
+                            "年龄设定：成年人\n"
+                            "外貌特征：普通外貌\n"
+                            "性格特质：平淡自然\n"
+                            "叙事风格：语速正常，语气自然"
+                        )
+                        npc_voice = crud.create_voice_design(
+                            db=db, user_id=user.id,
+                            name=f"[有声书] {project.title} - {speaker_name}",
+                            instruct=npc_instruct, backend_type=backend_type,
+                        )
+                        npc_char = crud.create_audiobook_character(
+                            db=db, project_id=project_id, name=speaker_name,
+                            description=f"配角：{speaker_name}",
+                            instruct=npc_instruct, voice_design_id=npc_voice.id,
+                        )
+                        char_map[speaker_name] = npc_char
+                        ps.append_line(key, f"\n[NPC] 自动创建配角：{speaker_name}")
+                    except Exception as e:
+                        logger.error(f"Failed to create NPC {speaker_name}: {e}")
+
+                crud.delete_audiobook_segments_for_chapter(db, project_id, idx)
+
+                seg_counter = 0
+                for seg in segments_data:
+                    seg_text = seg.get("text", "").strip()
+                    if not seg_text:
+                        continue
+                    char = char_map.get(seg.get("character", "旁白")) or char_map.get("旁白")
+                    if not char:
+                        continue
+                    crud.create_audiobook_segment(
+                        db, project_id, char.id, seg_text,
+                        chapter_index=idx, segment_index=seg_counter,
+                        emo_text=seg.get("emo_text"), emo_alpha=seg.get("emo_alpha"),
+                    )
+                    seg_counter += 1
+
+                crud.update_audiobook_chapter_status(db, chapter_obj.id, "ready")
+                ps.append_line(key, f"\n✓ {seg_counter} 段")
+
+            except Exception as e:
+                logger.error(f"Chapter {idx} script generation failed: {e}", exc_info=True)
+                ps.append_line(key, f"\n[错误] {e}")
+                crud.update_audiobook_chapter_status(db, chapter_obj.id, "error", error_message=str(e))
+
+        crud.update_audiobook_project_status(db, project_id, "ready")
+        ps.append_line(key, f"\n\n[完成] AI剧本生成完毕，项目已就绪")
+        ps.mark_done(key)
+        logger.info(f"AI script chapters generation complete for project {project_id}")
+
+    except Exception as e:
+        logger.error(f"generate_ai_script_chapters failed for project {project_id}: {e}", exc_info=True)
+        ps.append_line(key, f"\n[错误] {e}")
+        ps.mark_done(key)
+        crud.update_audiobook_project_status(db, project_id, "error", error_message=str(e))
+
+
 async def analyze_project(project_id: int, user: User, db: Session, turbo: bool = False) -> None:
     project = db.query(AudiobookProject).filter(AudiobookProject.id == project_id).first()
     if not project:
@@ -242,10 +579,10 @@ async def analyze_project(project_id: int, user: User, db: Session, turbo: bool
             usage_callback=_log_analyze_usage,
         )
 
-        has_narrator = any(c.get("name") == "narrator" for c in characters_data)
+        has_narrator = any(c.get("name") in ("narrator", "旁白") for c in characters_data)
         if not has_narrator:
             characters_data.insert(0, {
-                "name": "narrator",
+                "name": "旁白",
                 "gender": "未知",
                 "description": "第三人称旁白叙述者",
                 "instruct": (
@@ -266,10 +603,12 @@ async def analyze_project(project_id: int, user: User, db: Session, turbo: bool
         backend_type = user.user_preferences.get("default_backend", "aliyun") if user.user_preferences else "aliyun"
 
         for char_data in characters_data:
-            name = char_data.get("name", "narrator")
+            name = char_data.get("name", "旁白")
+            if name == "narrator":
+                name = "旁白"
             instruct = char_data.get("instruct", "")
             description = char_data.get("description", "")
-            gender = char_data.get("gender") or ("未知" if name == "narrator" else None)
+            gender = char_data.get("gender") or ("未知" if name == "旁白" else None)
             try:
                 voice_design = crud.create_voice_design(
                     db=db,
@@ -443,7 +782,7 @@ async def parse_one_chapter(project_id: int, chapter_id: int, user: User, db) ->
                 ps.append_line(key, f"\n[回退] {e}")
                 failed_chunks += 1
                 last_error = str(e)
-                narrator = char_map.get("narrator")
+                narrator = char_map.get("旁白") or char_map.get("narrator")
                 if narrator:
                     crud.create_audiobook_segment(
                         db, project_id, narrator.id, chunk.strip(),
@@ -457,7 +796,7 @@ async def parse_one_chapter(project_id: int, chapter_id: int, user: User, db) ->
                 seg_text = seg.get("text", "").strip()
                 if not seg_text:
                     continue
-                char = char_map.get(seg.get("character", "narrator")) or char_map.get("narrator")
+                char = char_map.get(seg.get("character", "旁白")) or char_map.get("旁白") or char_map.get("narrator")
                 if not char:
                     continue
                 seg_emo_text = seg.get("emo_text", "") or None
@@ -834,8 +1173,6 @@ async def generate_character_preview(project_id: int, char_id: int, user: User,
     audio_path = output_base / f"char_{char_id}.wav"
 
     preview_name = char.name
-    if preview_name == "narrator":
-        preview_name = "旁白"
         
     preview_desc = ""
     if char.description: