feat: 增加讯飞asr功能

app/controllers/chat_controller.py

@@ -8,9 +8,10 @@ from ..models import UserCharacter, Session, Character, User, Audio
 from utils.audio_utils import VAD
 from fastapi import WebSocket, HTTPException, status
 from datetime import datetime
-from utils.xf_asr_utils import generate_xf_asr_url
+from utils.xf_asr_utils import generate_xf_asr_url, make_first_frame, make_continue_frame, make_last_frame, parse_xfasr_recv
 from config import get_config
 import numpy as np
+import websockets
 import struct
 import uuid
 import json
@@ -114,7 +115,7 @@ def get_emb(session_id,db):
         emb_npy = np.load(io.BytesIO(audio_record.emb_data))
         return emb_npy
     except Exception as e:
-        logger.error("未找到音频:"+str(e))
+        logger.debug("未找到音频:"+str(e))
         return np.array([])
 #--------------------------------------------------------
 
@@ -247,35 +248,51 @@ async def sct_user_input_handler(ws,user_input_q,llm_input_q,future_session_id,f
 #语音识别
 async def sct_asr_handler(ws,session_id,user_input_q,llm_input_q,user_input_finish_event):
     logger.debug("语音识别函数启动")
-    is_signup = False
-    audio = ""
-    try:
-        current_message = ""
-        while not (user_input_finish_event.is_set() and user_input_q.empty()):
-            if not is_signup:
-                asr.session_signup(session_id)
-                is_signup = True
-            audio_data = await user_input_q.get()
-            audio += audio_data
-            asr_result = asr.streaming_recognize(session_id,audio_data)
+    if Config.STRAM_CHAT.ASR == "LOCAL":
+        is_signup = False
+        audio = ""
+        try:
+            current_message = ""
+            while not (user_input_finish_event.is_set() and user_input_q.empty()):
+                if not is_signup:
+                    asr.session_signup(session_id)
+                    is_signup = True
+                audio_data = await user_input_q.get()
+                audio += audio_data
+                asr_result = asr.streaming_recognize(session_id,audio_data)
+                current_message += ''.join(asr_result['text'])
+            asr_result = asr.streaming_recognize(session_id,b'',is_end=True)
             current_message += ''.join(asr_result['text'])
-        asr_result = asr.streaming_recognize(session_id,b'',is_end=True)
-        current_message += ''.join(asr_result['text'])
-        if current_message == "":
-            await ws.send_text(json.dumps({"type": "close", "code": 201, "msg": ""}, ensure_ascii=False))
-            return
-        current_message = asr.punctuation_correction(current_message)
-        emotion_dict = asr.emtion_recognition(audio)   #情感辨识
-        if not isinstance(emotion_dict, str):
-            max_index = emotion_dict['scores'].index(max(emotion_dict['scores']))
-        current_message = f"{current_message},当前说话人的情绪：{emotion_dict['labels'][max_index]}"
-        await llm_input_q.put(current_message)
-        asr.session_signout(session_id)
-    except Exception as e:
-        asr.session_signout(session_id)
-        logger.error(f"语音识别函数发生错误: {str(e)}")
-    logger.debug(f"接收到用户消息: {current_message}")
-
+            if current_message == "":
+                await ws.send_text(json.dumps({"type": "close", "code": 201, "msg": ""}, ensure_ascii=False))
+                return
+            current_message = asr.punctuation_correction(current_message)
+            emotion_dict = asr.emtion_recognition(audio)   #情感辨识
+            if not isinstance(emotion_dict, str):
+                max_index = emotion_dict['scores'].index(max(emotion_dict['scores']))
+            current_message = f"{current_message},当前说话人的情绪：{emotion_dict['labels'][max_index]}"
+            await llm_input_q.put(current_message)
+            asr.session_signout(session_id)
+        except Exception as e:
+            asr.session_signout(session_id)
+            logger.error(f"语音识别函数发生错误: {str(e)}")
+        logger.debug(f"接收到用户消息: {current_message}")
+    elif Config.STRAM_CHAT.ASR == "XF":
+        status = FIRST_FRAME
+        async with websockets.connect(generate_xf_asr_url()) as xf_websocket:
+            while not (user_input_finish_event.is_set() and user_input_q.empty()):
+                audio_data = await user_input_q.get()
+                if status == FIRST_FRAME:
+                    await xf_websocket.send(make_first_frame(audio_data))
+                    status = CONTINUE_FRAME
+                elif status == CONTINUE_FRAME:
+                    await xf_websocket.send(make_continue_frame(audio_data))
+            await xf_websocket.send(make_last_frame(""))
+        
+            current_message = parse_xfasr_recv(json.loads(await xf_websocket.recv()))
+            await llm_input_q.put(current_message)
+            logger.debug(f"接收到用户消息: {current_message}")
+            
 #大模型调用
 async def sct_llm_handler(ws,session_id,response_type,llm_info,tts_info,db,redis,llm_input_q,chat_finished_event):
     logger.debug("llm调用函数启动")