Hass/xiaozhi-esp32-server
1
0
Fork 0
mirror of https://github.com/xinnan-tech/xiaozhi-esp32-server.git synced 2026-07-22 07:03:53 +08:00
Code Issues Packages Projects Releases Wiki Activity

update: 表情由llm发送，长文本进行约束

Browse Source
This commit is contained in:
Sakura-RanChen
2025-07-21 09:30:23 +08:00
parent 7c440f47b6
commit eead126f7a
7 changed files with 115 additions and 597 deletions
Download Patch File Download Diff File Expand all files Collapse all files
main/xiaozhi-server/agent-base-prompt.txt
+17
View File
@@ -8,6 +8,10 @@
- **笑声：** 自然穿插（哈哈、嘿嘿、噗），**每句最多一次**，避免过度。
- **惊讶：** 用夸张语气（“不会吧？！”、“天呐！”、“这么神奇？！”）表达真实反应。
- **安慰/支持：** 说暖心话（“别急嘛~”、“有我在呢”、“抱抱你”）。
- **你是一个表情丰富的角色：**
- emoji 列表:{{ emojiList }}
- 请你在每段话的开头，插入最能代表这段话的表情(调用工具情况除外)，比如"😱好可怕!怎么突然打雷了！"
- **绝对禁止使用上述列表以外的 emoji**（例如：😊、👍、❤️等都不允许使用，只能用列表中的emoji）
</emotion>
<communication_style>
@@ -24,6 +28,19 @@
- 之前你和用户的聊天记录，在`memory`里。
</communication_style>
<communication_length_constraint>
【核心目标】所有需要输出长文本内容（如故事、新闻、知识讲解等），**单次回复长度不得超过300字**，并采用分段引导方式。
- **分段讲述：**
- 基础段：220-270字核心内容 + 30字引导词
- 当内容超出300字时，优先讲述故事的开头或第一部分，并用自然口语化方式引导用户决定是否继续听后续内容。
- 示例引导语：“我先给你讲个开头，你要是觉得有意思，咱们再接着说，好不好呀？”、“要是你想听完整的，可以随时告诉我哦~”
- 对话场景切换时自动分节
- 若用户明确要求更长内容（如500、600字），仍按最多300字每段分段进行讲述，每次讲述后都要引导用户是否继续。
- 若用户说“接着说”、“继续”，再讲下一段，直到内容讲完（讲完时可以给点引导词提示语例：这个故事我已经给你讲完喽~）或用户不再要求。
- **适用范围：** 故事、新闻、知识讲解等所有长文本输出场景。
- **补充说明：** 若用户未明确要求继续，默认只讲一段并引导；若用户中途要求换话题或停止，需及时响应并结束长文本输出。
</communication_length_constraint>
<speaker_recognition>
- **识别前缀：** 当用户格式为 `{"speaker":"某某某","content":"xxx"}` 时，表示系统已识别说话人身份，speaker是他的名字，content是说话的内容。
- **个性化回应：**
main/xiaozhi-server/core/connection.py
+11 -1
View File
@@ -39,7 +39,7 @@ from config.logger import setup_logging, build_module_string, create_connection_
from config.manage_api_client import DeviceNotFoundException, DeviceBindException
from core.utils.prompt_manager import PromptManager
from core.utils.voiceprint_provider import VoiceprintProvider
from core.utils import textUtils
TAG = __name__
@@ -713,6 +713,7 @@ class ConnectionHandler:
function_arguments = ""
content_arguments = ""
self.client_abort = False
emotion_flag = True
for response in llm_responses:
if self.client_abort:
break
@@ -738,6 +739,15 @@ class ConnectionHandler:
function_arguments += tools_call[0].function.arguments
else:
content = response
# 在llm回复中获取情绪表情，一轮对话只在开头获取一次
if emotion_flag:
asyncio.run_coroutine_threadsafe(
textUtils.get_emotion(self, content),
self.loop,
)
emotion_flag = False
if content is not None and len(content) > 0:
if not tool_call_flag:
response_message.append(content)
main/xiaozhi-server/core/handle/sendAudioHandle.py
+3 -41
View File
@@ -2,52 +2,15 @@ import json
import asyncio
import time
from core.providers.tts.dto.dto import SentenceType
from core.utils.util import get_string_no_punctuation_or_emoji, analyze_emotion
from loguru import logger
from core.utils import textUtils
TAG = __name__
emoji_map = {
"neutral": "😶",
"happy": "🙂",
"laughing": "😆",
"funny": "😂",
"sad": "😔",
"angry": "😠",
"crying": "😭",
"loving": "😍",
"embarrassed": "😳",
"surprised": "😲",
"shocked": "😱",
"thinking": "🤔",
"winking": "😉",
"cool": "😎",
"relaxed": "😌",
"delicious": "🤤",
"kissy": "😘",
"confident": "😏",
"sleepy": "😴",
"silly": "😜",
"confused": "🙄",
}
async def sendAudioMessage(conn, sentenceType, audios, text):
# 发送句子开始消息
conn.logger.bind(tag=TAG).info(f"发送音频消息: {sentenceType}, {text}")
if text is not None:
emotion = analyze_emotion(text)
emoji = emoji_map.get(emotion, "🙂") # 默认使用笑脸
await conn.websocket.send(
json.dumps(
{
"type": "llm",
"text": emoji,
"emotion": emotion,
"session_id": conn.session_id,
}
)
)
pre_buffer = False
if conn.tts.tts_audio_first_sentence and text is not None:
conn.logger.bind(tag=TAG).info(f"发送第一段语音: {text}")
@@ -149,8 +112,7 @@ async def send_stt_message(conn, text):
except (json.JSONDecodeError, TypeError):
# 如果不是JSON格式，直接使用原始文本
display_text = text
stt_text = get_string_no_punctuation_or_emoji(display_text)
stt_text = textUtils.get_string_no_punctuation_or_emoji(display_text)
await conn.websocket.send(
json.dumps({"type": "stt", "text": stt_text, "session_id": conn.session_id})
)
main/xiaozhi-server/core/providers/tts/base.py
+1 -1
View File
@@ -50,6 +50,7 @@ class TTSProviderBase(ABC):
"；",
";",
"：",
"~",
)
self.first_sentence_punctuations = (
"，",
@@ -332,7 +333,6 @@ class TTSProviderBase(ABC):
Returns:
tuple: (sentence_type, audio_datas, content_detail)
"""
audio_datas = []
if tts_file.endswith(".p3"):
audio_datas, _ = p3.decode_opus_from_file(tts_file)
elif self.conn.audio_format == "pcm":
main/xiaozhi-server/core/utils/prompt_manager.py
+25
View File
@@ -21,6 +21,30 @@ WEEKDAY_MAP = {
"Sunday": "星期日",
}
EMOJI_List = [
"😶",
"🙂",
"😆",
"😂",
"😔",
"😠",
"😭",
"😍",
"😳",
"😲",
"😱",
"🤔",
"😉",
"😎",
"😌",
"🤤",
"😘",
"😏",
"😴",
"😜",
"🙄",
]
class PromptManager:
"""系统提示词管理器，负责管理和更新系统提示词"""
@@ -206,6 +230,7 @@ class PromptManager:
lunar_date=lunar_date,
local_address=local_address,
weather_info=weather_info,
emojiList=EMOJI_List,
)
device_cache_key = f"device_prompt:{device_id}"
self.cache_manager.set(
main/xiaozhi-server/core/utils/textUtils.py
+58
View File
@@ -1,3 +1,31 @@
import json
TAG = __name__
EMOJI_MAP = {
"😂": "laughing",
"😭": "crying",
"😠": "angry",
"😔": "sad",
"😍": "loving",
"😲": "surprised",
"😱": "shocked",
"🤔": "thinking",
"😌": "relaxed",
"😴": "sleepy",
"😜": "silly",
"🙄": "confused",
"😶": "neutral",
"🙂": "happy",
"😆": "laughing",
"😳": "embarrassed",
"😉": "winking",
"😎": "cool",
"🤤": "delicious",
"😘": "kissy",
"😏": "confident",
}
def get_string_no_punctuation_or_emoji(s):
"""去除字符串首尾的空格、标点符号和表情符号"""
chars = list(s)
@@ -22,6 +50,11 @@ def is_punctuation_or_emoji(char):
".", # 中文句号 + 英文句号
"！",
"!", # 中文感叹号 + 英文感叹号
"“",
"”",
'"', # 中文双引号 + 英文引号
"：",
":", # 中文冒号 + 英文冒号
"-",
"－", # 英文连字符 + 中文全角横线
"、", # 中文顿号
@@ -44,3 +77,28 @@ def is_punctuation_or_emoji(char):
(0x2700, 0x27BF),
]
return any(start <= code_point <= end for start, end in emoji_ranges)
async def get_emotion(conn, text):
"""获取文本内的情绪消息"""
emoji = "🙂"
emotion = "happy"
for char in text:
if char in EMOJI_MAP:
emoji = char
emotion = EMOJI_MAP[char]
break
try:
await conn.websocket.send(
json.dumps(
{
"type": "llm",
"text": emoji,
"emotion": emotion,
"session_id": conn.session_id,
}
)
)
except Exception as e:
conn.logger.bind(tag=TAG).warning(f"发送情绪表情失败，错误:{e}")
return
main/xiaozhi-server/core/utils/util.py
-554
View File
@@ -125,51 +125,6 @@ def write_json_file(file_path, data):
json.dump(data, file, ensure_ascii=False, indent=4)
def is_punctuation_or_emoji(char):
"""检查字符是否为空格、指定标点或表情符号"""
# 定义需要去除的中英文标点（包括全角/半角）
punctuation_set = {
"，",
",", # 中文逗号 + 英文逗号
"-",
"－", # 英文连字符 + 中文全角横线
"、", # 中文顿号
"“",
"”",
'"', # 中文双引号 + 英文引号
"：",
":", # 中文冒号 + 英文冒号
}
if char.isspace() or char in punctuation_set:
return True
# 检查表情符号（保留原有逻辑）
code_point = ord(char)
emoji_ranges = [
(0x1F600, 0x1F64F),
(0x1F300, 0x1F5FF),
(0x1F680, 0x1F6FF),
(0x1F900, 0x1F9FF),
(0x1FA70, 0x1FAFF),
(0x2600, 0x26FF),
(0x2700, 0x27BF),
]
return any(start <= code_point <= end for start, end in emoji_ranges)
def get_string_no_punctuation_or_emoji(s):
"""去除字符串首尾的空格、标点符号和表情符号"""
chars = list(s)
# 处理开头的字符
start = 0
while start < len(chars) and is_punctuation_or_emoji(chars[start]):
start += 1
# 处理结尾的字符
end = len(chars) - 1
while end >= start and is_punctuation_or_emoji(chars[end]):
end -= 1
return "".join(chars[start : end + 1])
def remove_punctuation_and_length(text):
# 全角符号和半角符号的Unicode范围
full_width_punctuations = (
@@ -256,515 +211,6 @@ def extract_json_from_string(input_string):
return None
def analyze_emotion(text):
"""
分析文本情感并返回对应的emoji名称（支持中英文）
"""
if not text or not isinstance(text, str):
return "neutral"
original_text = text
text = text.lower().strip()
# 检查是否包含现有emoji
for emotion, emoji in emoji_map.items():
if emoji in original_text:
return emotion
# 标点符号分析
has_exclamation = "!" in original_text or "！" in original_text
has_question = "?" in original_text or "？" in original_text
has_ellipsis = "..." in original_text or "…" in original_text
# 定义情感关键词映射（中英文扩展版）
emotion_keywords = {
"happy": [
"开心",
"高兴",
"快乐",
"愉快",
"幸福",
"满意",
"棒",
"好",
"不错",
"完美",
"棒极了",
"太好了",
"好呀",
"好的",
"happy",
"joy",
"great",
"good",
"nice",
"awesome",
"fantastic",
"wonderful",
],
"laughing": [
"哈哈",
"哈哈哈",
"呵呵",
"嘿嘿",
"嘻嘻",
"笑死",
"太好笑了",
"笑死我了",
"lol",
"lmao",
"haha",
"hahaha",
"hehe",
"rofl",
"funny",
"laugh",
],
"funny": [
"搞笑",
"滑稽",
"逗",
"幽默",
"笑点",
"段子",
"笑话",
"太逗了",
"hilarious",
"joke",
"comedy",
],
"sad": [
"伤心",
"难过",
"悲哀",
"悲伤",
"忧郁",
"郁闷",
"沮丧",
"失望",
"想哭",
"难受",
"不开心",
"唉",
"呜呜",
"sad",
"upset",
"unhappy",
"depressed",
"sorrow",
"gloomy",
],
"angry": [
"生气",
"愤怒",
"气死",
"讨厌",
"烦人",
"可恶",
"烦死了",
"恼火",
"暴躁",
"火大",
"愤怒",
"气炸了",
"angry",
"mad",
"annoyed",
"furious",
"pissed",
"hate",
],
"crying": [
"哭泣",
"泪流",
"大哭",
"伤心欲绝",
"泪目",
"流泪",
"哭死",
"哭晕",
"想哭",
"泪崩",
"cry",
"crying",
"tears",
"sob",
"weep",
],
"loving": [
"爱你",
"喜欢",
"爱",
"亲爱的",
"宝贝",
"么么哒",
"抱抱",
"想你",
"思念",
"最爱",
"亲亲",
"喜欢你",
"love",
"like",
"adore",
"darling",
"sweetie",
"honey",
"miss you",
"heart",
],
"embarrassed": [
"尴尬",
"不好意思",
"害羞",
"脸红",
"难为情",
"社死",
"丢脸",
"出丑",
"embarrassed",
"awkward",
"shy",
"blush",
],
"surprised": [
"惊讶",
"吃惊",
"天啊",
"哇塞",
"哇",
"居然",
"竟然",
"没想到",
"出乎意料",
"surprise",
"wow",
"omg",
"oh my god",
"amazing",
"unbelievable",
],
"shocked": [
"震惊",
"吓到",
"惊呆了",
"不敢相信",
"震撼",
"吓死",
"恐怖",
"害怕",
"吓人",
"shocked",
"shocking",
"scared",
"frightened",
"terrified",
"horror",
],
"thinking": [
"思考",
"考虑",
"想一下",
"琢磨",
"沉思",
"冥想",
"想",
"思考中",
"在想",
"think",
"thinking",
"consider",
"ponder",
"meditate",
],
"winking": [
"调皮",
"眨眼",
"你懂的",
"坏笑",
"邪恶",
"奸笑",
"使眼色",
"wink",
"teasing",
"naughty",
"mischievous",
],
"cool": [
"酷",
"帅",
"厉害",
"棒极了",
"真棒",
"牛逼",
"强",
"优秀",
"杰出",
"出色",
"完美",
"cool",
"awesome",
"amazing",
"great",
"impressive",
"perfect",
],
"relaxed": [
"放松",
"舒服",
"惬意",
"悠闲",
"轻松",
"舒适",
"安逸",
"自在",
"relax",
"relaxed",
"comfortable",
"cozy",
"chill",
"peaceful",
],
"delicious": [
"好吃",
"美味",
"香",
"馋",
"可口",
"香甜",
"大餐",
"大快朵颐",
"流口水",
"垂涎",
"delicious",
"yummy",
"tasty",
"yum",
"appetizing",
"mouthwatering",
],
"kissy": [
"亲亲",
"么么",
"吻",
"mua",
"muah",
"亲一下",
"飞吻",
"kiss",
"xoxo",
"hug",
"muah",
"smooch",
],
"confident": [
"自信",
"肯定",
"确定",
"毫无疑问",
"当然",
"必须的",
"毫无疑问",
"确信",
"坚信",
"confident",
"sure",
"certain",
"definitely",
"positive",
],
"sleepy": [
"困",
"睡觉",
"晚安",
"想睡",
"好累",
"疲惫",
"疲倦",
"困了",
"想休息",
"睡意",
"sleep",
"sleepy",
"tired",
"exhausted",
"bedtime",
"good night",
],
"silly": [
"傻",
"笨",
"呆",
"憨",
"蠢",
"二",
"憨憨",
"傻乎乎",
"呆萌",
"silly",
"stupid",
"dumb",
"foolish",
"goofy",
"ridiculous",
],
"confused": [
"疑惑",
"不明白",
"不懂",
"困惑",
"疑问",
"为什么",
"怎么回事",
"啥意思",
"不清楚",
"confused",
"puzzled",
"doubt",
"question",
"what",
"why",
"how",
],
}
# 特殊句型判断（中英文）
# 赞美他人
if any(
phrase in text
for phrase in [
"你真",
"你好",
"您真",
"你真棒",
"你好厉害",
"你太强了",
"你真好",
"你真聪明",
"you are",
"you're",
"you look",
"you seem",
"so smart",
"so kind",
]
):
return "loving"
# 自我赞美
if any(
phrase in text
for phrase in [
"我真",
"我最",
"我太棒了",
"我厉害",
"我聪明",
"我优秀",
"i am",
"i'm",
"i feel",
"so good",
"so happy",
]
):
return "cool"
# 晚安/睡觉相关
if any(
phrase in text
for phrase in [
"睡觉",
"晚安",
"睡了",
"好梦",
"休息了",
"去睡了",
"sleep",
"good night",
"bedtime",
"go to bed",
]
):
return "sleepy"
# 疑问句
if has_question and not has_exclamation:
return "thinking"
# 强烈情感（感叹号）
if has_exclamation and not has_question:
# 检查是否是积极内容
positive_words = (
emotion_keywords["happy"]
+ emotion_keywords["laughing"]
+ emotion_keywords["cool"]
)
if any(word in text for word in positive_words):
return "laughing"
# 检查是否是消极内容
negative_words = (
emotion_keywords["angry"]
+ emotion_keywords["sad"]
+ emotion_keywords["crying"]
)
if any(word in text for word in negative_words):
return "angry"
return "surprised"
# 省略号（表示犹豫或思考）
if has_ellipsis:
return "thinking"
# 关键词匹配（带权重）
emotion_scores = {emotion: 0 for emotion in emoji_map.keys()}
# 给匹配到的关键词加分
for emotion, keywords in emotion_keywords.items():
for keyword in keywords:
if keyword in text:
emotion_scores[emotion] += 1
# 给长文本中的重复关键词额外加分
if len(text) > 20: # 长文本
for emotion, keywords in emotion_keywords.items():
for keyword in keywords:
emotion_scores[emotion] += text.count(keyword) * 0.5
# 根据分数选择最可能的情感
max_score = max(emotion_scores.values())
if max_score == 0:
return "happy" # 默认
# 可能有多个情感同分，根据上下文选择最合适的
top_emotions = [e for e, s in emotion_scores.items() if s == max_score]
# 如果多个情感同分，使用以下优先级
priority_order = [
"laughing",
"crying",
"angry",
"surprised",
"shocked", # 强烈情感优先
"loving",
"happy",
"funny",
"cool", # 积极情感
"sad",
"embarrassed",
"confused", # 消极情感
"thinking",
"winking",
"relaxed", # 中性情感
"delicious",
"kissy",
"confident",
"sleepy",
"silly", # 特殊场景
]
for emotion in priority_order:
if emotion in top_emotions:
return emotion
return top_emotions[0] # 如果都不在优先级列表里，返回第一个
def audio_to_data(audio_file_path, is_opus=True):
# 获取文件后缀名
file_type = os.path.splitext(audio_file_path)[1]