feat: 后端封装tts，可以通过配置文件切换openvoice和vits

app/controllers/chat_controller.py

@@ -11,6 +11,7 @@ from datetime import datetime
 from utils.xf_asr_utils import generate_xf_asr_url
 from config import get_config
 import numpy as np
+import struct
 import uuid
 import json
 import asyncio
@@ -319,20 +320,11 @@ async def sct_llm_handler(ws,session_id,response_type,llm_info,tts_info,db,redis
                             response_message = {"type": "text", "code":200, "msg": sentence}
                             await ws.send_text(json.dumps(response_message, ensure_ascii=False)) #返回文本信息
                         elif response_type == RESPONSE_AUDIO:
-                            if target_se.size == 0:
-                                audio,sr = tts._base_tts(text=sentence,
-                                                        noise_scale=tts_info["noise_scale"],
-                                                        noise_scale_w=tts_info["noise_scale_w"],
-                                                        speed=tts_info["speed"])
-                            else:
-                                audio,sr = tts.synthesize(text=sentence,
-                                                        noise_scale=tts_info["noise_scale"],
-                                                        noise_scale_w=tts_info["noise_scale_w"],
-                                                        speed=tts_info["speed"],
-                                                        target_se=target_se)
+                            audio,sr = tts.synthesize(text=sentence,tts_info=tts_info,target_se=target_se)
                             response_message = {"type": "text", "code":200, "msg": sentence}
-                            await ws.send_bytes(audio)  #返回音频数据
-                            await ws.send_text(json.dumps(response_message, ensure_ascii=False))    #返回文本信息
+                            response_bytes = json.dumps(response_message, ensure_ascii=False).encode('utf-8')
+                            header = struct.pack('!II',len(response_bytes),len(audio))
+                            message_bytes = header + response_bytes + audio
                         logger.debug(f"websocket返回: {sentence}")     
                     if is_end:
                         logger.debug(f"llm返回结果: {llm_response}")
@@ -500,17 +492,7 @@ async def scl_llm_handler(ws,session_id,response_type,llm_info,tts_info,db,redis
                                 response_message = {"type": "text", "code":200, "msg": sentence}
                                 await ws.send_text(json.dumps(response_message, ensure_ascii=False))
                             elif response_type == RESPONSE_AUDIO:
-                                if target_se.size == 0:
-                                    audio,sr = tts._base_tts(text=sentence,
-                                                            noise_scale=tts_info["noise_scale"],
-                                                            noise_scale_w=tts_info["noise_scale_w"],
-                                                            speed=tts_info["speed"])
-                                else:
-                                    audio,sr = tts.synthesize(text=sentence,
-                                                            noise_scale=tts_info["noise_scale"],
-                                                            noise_scale_w=tts_info["noise_scale_w"],
-                                                            speed=tts_info["speed"],
-                                                            target_se=target_se)
+                                audio,sr = tts.synthesize(text=sentence,tts_info=tts_info,target_se=target_se)
                                 response_message = {"type": "text", "code":200, "msg": sentence}
                                 await ws.send_bytes(audio)
                                 await ws.send_text(json.dumps(response_message, ensure_ascii=False))
@@ -684,17 +666,7 @@ async def voice_call_llm_handler(ws,session_id,llm_info,tts_info,db,redis,asr_re
                             llm_response += chunk_data
                         sentences,current_sentence,is_first = split_string_with_punctuation(current_sentence,chunk_data,is_first,is_end)
                         for sentence in sentences:
-                            if target_se.size == 0:
-                                audio,sr = tts._base_tts(text=sentence,
-                                                        noise_scale=tts_info["noise_scale"],
-                                                        noise_scale_w=tts_info["noise_scale_w"],
-                                                        speed=tts_info["speed"])
-                            else:
-                                audio,sr = tts.synthesize(text=sentence,
-                                                        noise_scale=tts_info["noise_scale"],
-                                                        noise_scale_w=tts_info["noise_scale_w"],
-                                                        speed=tts_info["speed"],
-                                                        target_se=target_se)
+                            audio,sr = tts.synthesize(text=sentence,tts_info=tts_info,target_se=target_se)
                             text_response = {"type": "llm_text", "code": 200, "msg": sentence}
                             await ws.send_bytes(audio) #返回音频二进制流数据
                             await ws.send_text(json.dumps(text_response, ensure_ascii=False)) #返回文本数据

app/controllers/user_controller.py

@@ -14,7 +14,7 @@ import io
 logger = get_logger()
 
 #依赖注入获取tts
-tts = get_tts()
+tts = get_tts("OPENVOICE")
 
 #创建用户
 async def create_user_handler(user:UserCrateRequest, db: Session):

app/dependencies/tts.py

@@ -1,11 +1,23 @@
-from utils.tts.openvoice_utils import TextToSpeech
+
 from app.dependencies.logger import get_logger
+from config import get_config
 
 logger = get_logger()
+Config = get_config()
+
+from utils.tts.openvoice_utils import TextToSpeech
+openvoice_tts = TextToSpeech(use_tone_convert=True,device='cuda')
+logger.info("TTS_OPENVOICE 初始化成功")
+
+from utils.tts.vits_utils import TextToSpeech
+vits_tts = TextToSpeech()
+logger.info("TTS_VITS 初始化成功")
+
+    
 
 #初始化全局tts对象
-tts = TextToSpeech(use_tone_convert=True,device='cuda')
-logger.info("TTS初始化成功")
-
-def get_tts():
-    return tts
+def get_tts(tts_type=Config.TTS_UTILS):
+    if tts_type == "OPENVOICE":
+        return openvoice_tts
+    elif tts_type == "VITS":
+        return vits_tts