bdim/Canto
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

feat: enhance emotion handling in chapter segment parsing with clearer guidelines and intensity recommendations

Browse Source
This commit is contained in:
bdim404
2026-03-12 23:38:44 +08:00
parent a91aff13a4
commit 233c4a9a98
1 changed files with 8 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
qwen3-tts-backend/core/llm_service.py
View File

@@ -218,13 +218,15 @@ class LLMService:
"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。" "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}。" f"已知角色列表（必须从中选择）：{names_str}。"
"所有非对话的叙述文字归属于narrator角色。\n" "所有非对话的叙述文字归属于narrator角色。\n"
"同时根据语境为每个片段判断情绪，可选情绪及对应强度如下（必须严格使用以下值）：\n" "同时根据语境为每个片段判断情绪类型和强度（emo_alpha）。\n"
"开心(emo_alpha=0.6)、愤怒(emo_alpha=0.15)、悲伤(emo_alpha=0.4)、恐惧(emo_alpha=0.4)、" "可选情绪：开心、愤怒、悲伤、恐惧、厌恶、低沉、惊讶、中性。\n"
"厌恶(emo_alpha=0.6)、低沉(emo_alpha=0.6)、惊讶(emo_alpha=0.3)、中性(emo_alpha=0.5)。\n" "强度参考值（仅供参考，请根据实际语境自行判断）：开心≈0.4、愤怒≈0.08、悲伤≈0.1、恐惧≈0.1、厌恶≈0.4、低沉≈0.4、惊讶≈0.3、中性≈0.3。\n"
"narrator旁白及情绪不明显的片段，emo_text设为\"\"，emo_alpha设为0.5。\n" "强度建议控制在0.5以下，情绪越激烈越应克制，强烈情绪不超过0.45。\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，避免情绪在相邻片段间跳跃。\n"
"narrator旁白及情绪不明显的片段，emo_text设为\"\"，emo_alpha设为0.3。\n"
"只输出JSON数组，不要有其他文字，格式如下：\n" "只输出JSON数组，不要有其他文字，格式如下：\n"
'[{"character": "narrator", "text": "叙述文字", "emo_text": "", "emo_alpha": 0.5}, ' '[{"character": "narrator", "text": "叙述文字", "emo_text": "", "emo_alpha": 0.3}, '
'{"character": "角色名", "text": "对话内容", "emo_text": "开心", "emo_alpha": 0.6}, ...]' '{"character": "角色名", "text": "对话内容", "emo_text": "开心", "emo_alpha": 0.4}, ...]'
) )
user_message = f"请解析以下章节文本：\n\n{chapter_text}" user_message = f"请解析以下章节文本：\n\n{chapter_text}"
result = await self.stream_chat_json(system_prompt, user_message, on_token, max_tokens=16384, usage_callback=usage_callback) result = await self.stream_chat_json(system_prompt, user_message, on_token, max_tokens=16384, usage_callback=usage_callback)