修改参数

app.py

@@ -335,8 +335,8 @@ if __name__ == '__main__':
     parser.add_argument('--customvideo_config', type=str, default='', help="custom action json")
 
     parser.add_argument('--tts', type=str, default='edgetts', help="tts service type") #xtts gpt-sovits cosyvoice
-    parser.add_argument('--REF_FILE', type=str, default="zh-CN-YunxiaNeural")
+    parser.add_argument('--REF_FILE', type=str, default="input/doubao.wav")
-    parser.add_argument('--REF_TEXT', type=str, default=None)
+    parser.add_argument('--REF_TEXT', type=str, default="刚进直播间的宝子们，左上角先点个关注，点亮咱们家的粉丝灯牌！我是你们的主播陈婉婉，今天给大家准备了超级重磅的福利")
     parser.add_argument('--TTS_SERVER', type=str, default='http://127.0.0.1:9880') # http://localhost:9000
     # parser.add_argument('--CHARACTER', type=str, default='test')
     # parser.add_argument('--EMOTION', type=str, default='default')
@@ -370,9 +370,9 @@ if __name__ == '__main__':
     elif opt.model == 'wav2lip':
         from lipreal import LipReal,load_model,load_avatar,warm_up
         logger.info(opt)
-        model = load_model("./models/wav2lip.pth")
+        model = load_model("./models/wav2lip384.pth")
         avatar = load_avatar(opt.avatar_id)
-        warm_up(opt.batch_size,model,256)
+        warm_up(opt.batch_size,model,384)
     elif opt.model == 'ultralight':
         from lightreal import LightReal,load_model,load_avatar,warm_up
         logger.info(opt)

lipreal.py

@@ -36,7 +36,7 @@ import torch.multiprocessing as mp
 from lipasr import LipASR
 import asyncio
 from av import AudioFrame, VideoFrame
-from wav2lip.models import Wav2Lip
+from wav2lip.models384.wav2lip import Wav2Lip
 from basereal import BaseReal
 
 #from imgcache import ImgCache

llm.py

@@ -1,48 +1,66 @@
+import re
 import time
-import os
+import requests
+import json
 from basereal import BaseReal
 from logger import logger
 
-def llm_response(message,nerfreal:BaseReal):
+def llm_response(message: str, nerfreal: BaseReal):
     start = time.perf_counter()
-    from openai import OpenAI
+    logger.info(f"message:{message}")
-    client = OpenAI(
+    url = "http://192.168.10.70:11434/api/chat"
-        # 如果您没有配置环境变量，请在此处用您的API Key进行替换
+    data = {
-        api_key=os.getenv("DASHSCOPE_API_KEY"),
+        "model": "qwen3:30b-a3b",
-        # 填写DashScope SDK的base_url
+        "messages": [
-        base_url="https://dashscope.aliyuncs.com/compatible-mode/v1",
+            {"role": "system", "content": "你叫小苏，贵州人，带有轻微贵州口音,现在正在直播，你的人设是带货主播，你现在卖的产品是酱酒，任务是消除消费者对假酒的顾虑。"},
-    )
+            {"role": "user", "content": f"{message}\n不可以输出表情包/no_think"}
-    end = time.perf_counter()
+        ],
-    logger.info(f"llm Time init: {end-start}s")
+        "stream": True
-    completion = client.chat.completions.create(
+    }
-        model="qwen-plus",
+
-        messages=[{'role': 'system', 'content': 'You are a helpful assistant.'},
+    response = requests.post(url, json=data, stream=True)
-                  {'role': 'user', 'content': message}],
+    logger.info(f"Ollama init time: {time.perf_counter() - start:.2f}s")
-        stream=True,
+
-        # 通过以下设置，在流式输出的最后一行展示token使用信息
+    result = ""
-        stream_options={"include_usage": True}
+    full_answer = ""
-    )
-    result=""
     first = True
-    for chunk in completion:
+    think_tag_pattern = re.compile(r"</?think>", re.IGNORECASE)
-        if len(chunk.choices)>0:
+    punctuation = ",.!;:，。！？：；"
-            #print(chunk.choices[0].delta.content)
+
-            if first:
+    for line in response.iter_lines():
-                end = time.perf_counter()
+        if not line:
-                logger.info(f"llm Time to first chunk: {end-start}s")
+            continue
-                first = False
+
-            msg = chunk.choices[0].delta.content
+        try:
-            lastpos=0
+            json_data = json.loads(line.decode("utf-8"))
-            #msglist = re.split('[,.!;:，。！?]',msg)
+        except json.JSONDecodeError:
-            for i, char in enumerate(msg):
+            continue
-                if char in ",.!;:，。！？：；" :
+
-                    result = result+msg[lastpos:i+1]
+        # 如果有结束标志，可以在此处判断并 break
-                    lastpos = i+1
+        # if json_data.get("done"):
-                    if len(result)>10:
+        #     break
-                        logger.info(result)
+
-                        nerfreal.put_msg_txt(result)
+        msg = json_data.get("message", {}).get("content", "")
-                        result=""
+        msg = think_tag_pattern.sub("", msg)
-            result = result+msg[lastpos:]
+        if not msg:
-    end = time.perf_counter()
+            continue
-    logger.info(f"llm Time to last chunk: {end-start}s")
+
-    nerfreal.put_msg_txt(result)    
+        full_answer += msg
+
+        if first:
+            logger.info(f"Ollama time to first chunk: {time.perf_counter() - start:.2f}s")
+            first = False
+
+        lastpos = 0
+        for i, char in enumerate(msg):
+            if char in punctuation:
+                result += msg[lastpos:i+1]
+                lastpos = i+1
+                if len(result) > 10:
+                    logger.info(result)
+                    nerfreal.put_msg_txt(result)
+                    result = ""
+        result += msg[lastpos:]
+
+    if result:
+        nerfreal.put_msg_txt(result)