Update xtts.py

xtts.py

@@ -5,7 +5,6 @@ import torch
 import traceback
 import numpy as np
 import scipy
-from importlib import import_module
 from TTS.api import TTS
 from flask import Flask, Blueprint,  request, jsonify, send_file
 
@@ -28,6 +27,8 @@ from qili import upload_bytes
 # else:
 #     app = Blueprint("xtts", __name__)
 
+tts=None
+model=None 
 
 sample_root= os.environ.get('XTTS_SAMPLE_DIR') 
 if(sample_root==None):
@@ -38,8 +39,21 @@ if not os.path.exists(sample_root):
 default_sample=f'{os.path.dirname(os.path.abspath(__file__))}/sample.wav', f'{sample_root}/sample.pt'
 ffmpeg=f'{os.path.dirname(os.path.abspath(__file__))}/ffmpeg'
 
-tts=None
-model=None 
+if tts is None:
+    model_name="tts_models/multilingual/multi-dataset/xtts_v2"
+    logging.info(f"loading model {model_name} ...")
+    tts = TTS(
+        model_path=model_path, 
+        config_path=config_path, 
+        vocoder_config_path=vocoder_config_path,  
+        progress_bar=False
+    )
+    model=tts.synthesizer.tts_model
+    #hack to use cache
+    model.__get_conditioning_latents=model.get_conditioning_latents
+    model.get_conditioning_latents=get_conditioning_latents
+    logging.info("model is ready")
+
 @app.route("/convert")
 def predict():
     global tts 
@@ -55,22 +69,6 @@ def predict():
     text= re.sub("([^\x00-\x7F]|\w)(\.|\。|\?)",r"\1 \2\2",text)
     
     try:
-        if tts is None:
-            TTS=import_module("TTS.api").TTS
-            model_name="tts_models/multilingual/multi-dataset/xtts_v2"
-            logging.info(f"loading model {model_name} ...")
-            tts = TTS(
-                model_path=model_path, 
-                config_path=config_path, 
-                vocoder_config_path=vocoder_config_path,  
-                progress_bar=False
-            )
-            model=tts.synthesizer.tts_model
-            #hack to use cache
-            model.__get_conditioning_latents=model.get_conditioning_latents
-            model.get_conditioning_latents=get_conditioning_latents
-            logging.info("model is ready")
-
         wav = tts.tts(
             text, 
             language=language if language is not None else "zh",
@@ -157,6 +155,4 @@ def trim_sample_audio(speaker_wav):
         traceback.print_exc()
         return speaker_wav
 
-for key, value in os.environ.items():
-    logging.info(f"{key}: {value}")
-loggin.info("xtts is ready")
+logging.info("xtts is ready")