106 lines
3.8 KiB
Python
106 lines
3.8 KiB
Python
import json
|
||
import aiohttp
|
||
import asyncio
|
||
from fastapi.encoders import jsonable_encoder
|
||
from starlette.websockets import WebSocket
|
||
|
||
from . import tts
|
||
from app.constants.model_data import tip_message, base_url, headers
|
||
|
||
|
||
async def process_voice_conversation(websocket: WebSocket, asr_text: str, message_id: str):
|
||
try:
|
||
print(f"开始处理语音对话 [{message_id}]: {asr_text}")
|
||
|
||
# 1. 发送ASR识别结果到前端
|
||
await websocket.send_json({
|
||
"type": "asr_result",
|
||
"messageId": message_id,
|
||
"result": asr_text
|
||
})
|
||
|
||
# 2. 构建LLM请求
|
||
messages = [
|
||
tip_message,
|
||
{"role": "user", "content": asr_text}
|
||
]
|
||
payload = {
|
||
"model": "gpt-4o",
|
||
"messages": messages,
|
||
"stream": True
|
||
}
|
||
|
||
print(f"发送LLM请求 [{message_id}]: {json.dumps(payload, ensure_ascii=False)}")
|
||
|
||
# 3. 流式处理LLM响应
|
||
full_response = ""
|
||
llm_completed = False
|
||
|
||
async with aiohttp.ClientSession() as session:
|
||
async with session.post(
|
||
base_url,
|
||
headers=headers,
|
||
json=jsonable_encoder(payload)
|
||
) as resp:
|
||
if resp.status != 200:
|
||
error_text = await resp.text()
|
||
raise Exception(f"LLM API请求失败: {resp.status} - {error_text}")
|
||
|
||
# 读取流式响应
|
||
async for line in resp.content:
|
||
if line:
|
||
line = line.decode('utf-8').strip()
|
||
if line.startswith('data: '):
|
||
data = line[6:].strip()
|
||
if data == '[DONE]':
|
||
llm_completed = True
|
||
print(f"LLM响应完成 [{message_id}]")
|
||
break
|
||
|
||
try:
|
||
result = json.loads(data)
|
||
# 提取内容
|
||
choices = result.get("choices", [])
|
||
if not choices:
|
||
# 跳过空choices数据包
|
||
continue
|
||
|
||
delta = choices[0].get("delta", {})
|
||
content = delta.get("content")
|
||
|
||
if content:
|
||
full_response += content
|
||
|
||
except json.JSONDecodeError as e:
|
||
print(f"JSON解析错误 [{message_id}]: {e}, 数据: {data}")
|
||
continue
|
||
except Exception as e:
|
||
print(f"处理数据包异常 [{message_id}]: {e}, 数据: {data}")
|
||
continue
|
||
|
||
# 4. LLM生成完成后，启动完整的TTS处理
|
||
if llm_completed and full_response:
|
||
print(f"LLM生成完成 [{message_id}], 总内容长度: {len(full_response)}")
|
||
print(f"完整内容: {full_response}")
|
||
|
||
# 发送完成消息
|
||
await websocket.send_json({
|
||
"type": "llm_complete_response",
|
||
"messageId": message_id,
|
||
"content": full_response
|
||
})
|
||
|
||
# 启动TTS处理完整内容
|
||
print(f"启动完整TTS处理 [{message_id}]: {full_response}")
|
||
await tts.handle_tts_text(websocket, message_id, full_response)
|
||
|
||
except Exception as e:
|
||
print(f"语音对话处理异常 [{message_id}]: {e}")
|
||
import traceback
|
||
traceback.print_exc()
|
||
await websocket.send_json({
|
||
"type": "voice_conversation_error",
|
||
"messageId": message_id,
|
||
"message": f"处理失败: {str(e)}"
|
||
})
|