Hass/xiaozhi-esp32-server
1
0
Fork 0
mirror of https://github.com/xinnan-tech/xiaozhi-esp32-server.git synced 2026-07-28 01:53:53 +08:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'main' into py_wakeup_audio

Browse Source
This commit is contained in:
hrz
2025-08-30 18:15:09 +08:00
committed by GitHub
parent 41b8fac3aa 3e1765a26d
commit 76a7ac04c9
17 changed files with 285 additions and 182 deletions
Download Patch File Download Diff File Expand all files Collapse all files
main/xiaozhi-server/core/connection.py
-4
View File
@@ -140,10 +140,6 @@ class ConnectionHandler:
self.func_handler = None
self.cmd_exit = self.config["exit_commands"]
self.max_cmd_length = 0
for cmd in self.cmd_exit:
if len(cmd) > self.max_cmd_length:
self.max_cmd_length = len(cmd)
# 是否在聊天结束后关闭连接
self.close_after_chat = False
main/xiaozhi-server/core/providers/llm/AliBL/AliBL.py
+57 -14
View File
@@ -1,8 +1,10 @@
from config.logger import setup_logging
from http import HTTPStatus
import dashscope
from dashscope import Application
from core.providers.llm.base import LLMProviderBase
from core.utils.util import check_model_key
import time
TAG = __name__
logger = setup_logging()
@@ -15,6 +17,7 @@ class LLMProvider(LLMProviderBase):
self.base_url = config.get("base_url")
self.is_No_prompt = config.get("is_no_prompt")
self.memory_id = config.get("ali_memory_id")
self.streaming_chunk_size = config.get("streaming_chunk_size", 3) # 每次流式返回的字符数
check_model_key("AliBLLLM", self.api_key)
def response(self, session_id, dialogue):
@@ -32,6 +35,8 @@ class LLMProvider(LLMProviderBase):
"app_id": self.app_id,
"session_id": session_id,
"messages": dialogue,
# 开启SDK原生流式
"stream": True,
}
if self.memory_id != False:
# 百练memory需要prompt参数
@@ -42,25 +47,63 @@ class LLMProvider(LLMProviderBase):
f"【阿里百练API服务】处理后的prompt: {prompt}"
)
# 可选地设置自定义API基地址（若配置为兼容模式URL则忽略）
if self.base_url and ("/api/" in self.base_url):
dashscope.base_http_api_url = self.base_url
responses = Application.call(**call_params)
if responses.status_code != HTTPStatus.OK:
logger.bind(tag=TAG).error(
f"code={responses.status_code}, "
f"message={responses.message}, "
f"请参考文档：https://help.aliyun.com/zh/model-studio/developer-reference/error-code"
)
yield "【阿里百练API服务响应异常】"
else:
logger.bind(tag=TAG).debug(
f"【阿里百练API服务】构造参数: {call_params}"
)
yield responses.output.text
# 流式处理（SDK在stream=True时返回可迭代对象；否则返回单次响应对象）
logger.bind(tag=TAG).debug(
f"【阿里百练API服务】构造参数: {dict(call_params, api_key='***')}"
)
last_text = ""
try:
for resp in responses:
if resp.status_code != HTTPStatus.OK:
logger.bind(tag=TAG).error(
f"code={resp.status_code}, message={resp.message}, 请参考文档：https://help.aliyun.com/zh/model-studio/developer-reference/error-code"
)
continue
current_text = getattr(getattr(resp, "output", None), "text", None)
if current_text is None:
continue
# SDK流式为增量覆盖，计算差量输出
if len(current_text) >= len(last_text):
delta = current_text[len(last_text):]
else:
# 避免偶发回退
delta = current_text
if delta:
yield delta
last_text = current_text
except TypeError:
# 非流式回落（一次性返回）
if responses.status_code != HTTPStatus.OK:
logger.bind(tag=TAG).error(
f"code={responses.status_code}, message={responses.message}, 请参考文档：https://help.aliyun.com/zh/model-studio/developer-reference/error-code"
)
yield "【阿里百练API服务响应异常】"
else:
full_text = getattr(getattr(responses, "output", None), "text", "")
logger.bind(tag=TAG).info(
f"【阿里百练API服务】完整响应长度: {len(full_text)}"
)
for i in range(0, len(full_text), self.streaming_chunk_size):
chunk = full_text[i:i + self.streaming_chunk_size]
if chunk:
yield chunk
except Exception as e:
logger.bind(tag=TAG).error(f"【阿里百练API服务】响应异常: {e}")
yield "【LLM服务响应异常】"
def response_with_functions(self, session_id, dialogue, functions=None):
logger.bind(tag=TAG).error(
f"阿里百练暂未实现完整的工具调用（function call），建议使用其他意图识别"
# 阿里百练当前未支持原生的 function call。为保持兼容，这里回退到普通文本流式输出。
# 上层会按 (content, tool_calls) 的形式消费，这里始终返回 (token, None)
logger.bind(tag=TAG).warning(
"阿里百练未实现原生 function call，已回退为纯文本流式输出"
)
for token in self.response(session_id, dialogue):
yield token, None
main/xiaozhi-server/core/providers/tools/server_mcp/mcp_executor.py
+1 -1
View File
@@ -18,8 +18,8 @@ class ServerMCPExecutor(ToolExecutor):
"""初始化MCP管理器"""
if not self._initialized:
self.mcp_manager = ServerMCPManager(self.conn)
await self.mcp_manager.initialize_servers()
self._initialized = True
await self.mcp_manager.initialize_servers()
async def execute(
self, conn, tool_name: str, arguments: Dict[str, Any]
main/xiaozhi-server/core/providers/tools/server_mcp/mcp_manager.py
+3
View File
@@ -68,6 +68,9 @@ class ServerMCPManager:
# 输出当前支持的服务端MCP工具列表
if hasattr(self.conn, "func_handler") and self.conn.func_handler:
# 刷新工具缓存以确保服务端MCP工具被正确加载
if hasattr(self.conn.func_handler, "tool_manager"):
self.conn.func_handler.tool_manager.refresh_tools()
self.conn.func_handler.current_support_functions()
def get_all_tools(self) -> List[Dict[str, Any]]:
main/xiaozhi-server/core/providers/tts/base.py
-2
View File
@@ -33,7 +33,6 @@ class TTSProviderBase(ABC):
def __init__(self, config, delete_audio_file):
self.interface_type = InterfaceType.NON_STREAM
self.conn = None
self.tts_timeout = 10
self.delete_audio_file = delete_audio_file
self.audio_file_type = "wav"
self.output_file = config.get("output_dir", "tmp/")
@@ -192,7 +191,6 @@ class TTSProviderBase(ABC):
async def open_audio_channels(self, conn):
self.conn = conn
self.tts_timeout = conn.config.get("tts_timeout", 10)
# tts 消化线程
self.tts_priority_thread = threading.Thread(
target=self.tts_text_priority_thread, daemon=True
main/xiaozhi-server/core/providers/tts/paddle_speech.py
+26 -4
View File
@@ -8,6 +8,7 @@ import wave
import websockets
from core.providers.tts.base import TTSProviderBase
from config.logger import setup_logging
from datetime import datetime
TAG = __name__
logger = setup_logging()
@@ -18,11 +19,12 @@ class TTSProvider(TTSProviderBase):
super().__init__(config, delete_audio_file)
self.url = config.get("url", "ws://192.168.1.10:8092/paddlespeech/tts/streaming")
self.protocol = config.get("protocol", "websocket")
if config.get("private_voice"):
self.spk_id = int(config.get("private_voice"))
else:
self.spk_id = int(config.get("spk_id", "0"))
self.spk_id = int(config.get("spk_id", "0"))
sample_rate = config.get("sample_rate", 24000)
self.sample_rate = float(sample_rate) if sample_rate else 24000
@@ -32,7 +34,21 @@ class TTSProvider(TTSProviderBase):
volume = config.get("volume", 1.0)
self.volume = float(volume) if volume else 1.0
self.save_path = config.get("save_path", "./streaming_tts.wav")
self.delete_audio_file = config.get("delete_audio", True)
if not self.delete_audio_file:
timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
save_path = config.get("save_path")
if save_path:
if not save_path.endswith('.wav'):
save_path = f"{save_path}_{timestamp}.wav"
else:
other_path = save_path[:-4]
save_path = f"{other_path}_{timestamp}.wav"
self.save_path = save_path
else:
self.save_path = f"./streaming_tts_{timestamp}.wav"
else:
self.save_path = None
async def pcm_to_wav(self, pcm_data: bytes, sample_rate: int = 24000, num_channels: int = 1,
bits_per_sample: int = 16) -> bytes:
@@ -151,6 +167,12 @@ class TTSProvider(TTSProviderBase):
# 接收结束响应避免服务抛出异常
await ws.recv()
# 根据配置决定是否保存文件
if not self.delete_audio_file and self.save_path:
with open(self.save_path, "wb") as f:
f.write(wav_data)
logger.bind(tag=TAG).info(f"音频文件已保存到: {self.save_path}")
# 返回或保存音频数据
if output_file:
with open(output_file, "wb") as file_to_save:
@@ -159,4 +181,4 @@ class TTSProvider(TTSProviderBase):
return wav_data
except Exception as e:
raise Exception(f"Error during TTS WebSocket request: {e} while processing text: {text}")
raise Exception(f"Error during TTS WebSocket request: {e} while processing text: {text}")