From d1503b08cb4088f69f596c162a1a28fab2cdcf68 Mon Sep 17 00:00:00 2001 From: bdim404 Date: Fri, 13 Mar 2026 15:56:56 +0800 Subject: [PATCH] feat: enhance character data handling in chapter parsing and LLM service --- qwen3-tts-backend/core/audiobook_service.py | 7 +++++-- qwen3-tts-backend/core/llm_service.py | 18 +++++++++++++++--- 2 files changed, 20 insertions(+), 5 deletions(-) diff --git a/qwen3-tts-backend/core/audiobook_service.py b/qwen3-tts-backend/core/audiobook_service.py index 51d3f00..a236ffe 100644 --- a/qwen3-tts-backend/core/audiobook_service.py +++ b/qwen3-tts-backend/core/audiobook_service.py @@ -965,7 +965,10 @@ async def parse_one_chapter(project_id: int, chapter_id: int, user: User, db) -> raise ValueError("No characters found. Please analyze the project first.") char_map: dict[str, AudiobookCharacter] = {c.name: c for c in characters} - character_names = list(char_map.keys()) + characters_data = [ + {"name": c.name, "gender": c.gender or "未知", "description": c.description or ""} + for c in characters + ] label = chapter.title or f"第 {chapter.chapter_index + 1} 章" ps.append_line(key, f"[{label}] 开始解析 ({len(chapter.source_text)} 字)") @@ -998,7 +1001,7 @@ async def parse_one_chapter(project_id: int, chapter_id: int, user: User, db) -> ps.append_token(key, token) try: - segments_data = await llm.parse_chapter_segments(chunk, character_names, on_token=on_token, usage_callback=_log_parse_usage) + segments_data = await llm.parse_chapter_segments(chunk, characters_data, on_token=on_token, usage_callback=_log_parse_usage) except Exception as e: logger.warning(f"Chapter {chapter_id} chunk {i} failed: {e}") ps.append_line(key, f"\n[回退] {e}") diff --git a/qwen3-tts-backend/core/llm_service.py b/qwen3-tts-backend/core/llm_service.py index 81ca9ba..28d40ba 100644 --- a/qwen3-tts-backend/core/llm_service.py +++ b/qwen3-tts-backend/core/llm_service.py @@ -357,6 +357,11 @@ class LLMService: ) -> str: char_names = [c.get("name", "") for c in characters if c.get("name") not in ("narrator", "旁白")] names_str = "、".join(char_names) + char_personality_lines = "\n".join( + f" - {c['name']}（{c.get('gender', '')}）：{c.get('description', '').strip()}" + for c in characters if c.get("name") not in ("narrator", "旁白") and c.get("description", "").strip() + ) + char_personality_str = f"\n角色性格（据此调整情绪幅度，外向/激动者可偏高，内敛/沉稳者应偏低）：\n{char_personality_lines}\n" if char_personality_lines else "" limits_str, emo_guidance = self._emotion_limits(violence_level, eroticism_level) emo_guidance_line = f"- {emo_guidance}\n" if emo_guidance else "" max_level = max(violence_level, eroticism_level) @@ -384,6 +389,7 @@ class LLMService: f"- 各情感比重上限（严格不超过）：{limits_str}\n" "- 鼓励使用低值（0.05–0.10）表达微弱、内敛或一闪而过的情绪，无需非强即无\n" "- 确实没有任何情绪色彩时可省略整个括号\n" + + char_personality_str + narrator_rule + emo_guidance_line + "\n其他规则：\n" @@ -449,8 +455,13 @@ class LLMService: result = await self.stream_chat_json(system_prompt, user_message, max_tokens=4096, usage_callback=usage_callback) return result.get("chapters", []) - async def parse_chapter_segments(self, chapter_text: str, character_names: list[str], on_token=None, usage_callback: Optional[Callable[[int, int], None]] = None) -> list[Dict]: - names_str = "、".join(character_names) + async def parse_chapter_segments(self, chapter_text: str, characters: list[Dict], on_token=None, usage_callback: Optional[Callable[[int, int], None]] = None) -> list[Dict]: + names_str = "、".join(c.get("name", "") for c in characters) + personality_lines = "\n".join( + f" - {c['name']}（{c.get('gender', '')}）：{c.get('description', '').strip()}" + for c in characters if c.get("name") not in ("narrator", "旁白") and c.get("description", "").strip() + ) + personality_str = f"\n角色性格（据此调整情绪幅度，外向/激动者可偏高，内敛/沉稳者应偏低）：\n{personality_lines}\n" if personality_lines else "" system_prompt = ( "你是一个专业的有声书制作助手。请将给定的章节文本解析为对话片段列表。" f"已知角色列表（必须从中选择）：{names_str}。" @@ -462,7 +473,8 @@ class LLMService: " 混合情感：用 情感词:比重 格式拼接，emo_alpha 设为 1.0，如 emo_text=\"开心:0.6+悲伤:0.2\", emo_alpha=1.0\n" "各情感比重上限（严格不超过）：开心=0.20、愤怒=0.15、悲伤=0.1、恐惧=0.1、厌恶=0.35、低沉=0.35、惊讶=0.10。\n" "鼓励用低值（0.05–0.10）表达微弱或内敛的情绪，不要非强即无；完全无情绪色彩时 emo_text 置空。\n" - "同一角色的连续台词，情绪应尽量保持一致或仅有微弱变化，避免相邻片段间情绪跳跃。\n" + + personality_str + + "同一角色的连续台词，情绪应尽量保持一致或仅有微弱变化，避免相邻片段间情绪跳跃。\n" "只输出JSON数组，不要有其他文字，格式如下：\n" '[{"character": "旁白", "text": "叙述文字", "emo_text": "", "emo_alpha": 0}, ' '{"character": "角色名", "text": "淡淡的问候", "emo_text": "开心", "emo_alpha": 0.08}, '