Update voice_processing.py

voice_processing.py

@@ -23,9 +23,6 @@ from lib.infer_pack.models import (
 from rmvpe import RMVPE
 from vc_infer_pipeline import VC
 
-model_cache = {}
-
-
 # Set logging levels
 logging.getLogger("fairseq").setLevel(logging.WARNING)
 logging.getLogger("numba").setLevel(logging.WARNING)
@@ -37,11 +34,11 @@ limitation = os.getenv("SYSTEM") == "spaces"
 
 config = Config()
 
-# Edge TTS voices
+# Edge TTS
 tts_voice_list = asyncio.get_event_loop().run_until_complete(edge_tts.list_voices())
-tts_voices = ["mn-MN-BataaNeural", "mn-MN-YesuiNeural"]
+tts_voices = ["mn-MN-BataaNeural", "mn-MN-YesuiNeural"]  # Specific voices
 
-# RVC models directory
+# RVC models
 model_root = "weights"
 models = [d for d in os.listdir(model_root) if os.path.isdir(f"{model_root}/{d}")]
 models.sort()
@@ -50,7 +47,6 @@ def get_unique_filename(extension):
     return f"{uuid.uuid4()}.{extension}"
 
 def model_data(model_name):
-    # We will not modify this function to cache models
     pth_path = [
         f"{model_root}/{model_name}/{f}"
         for f in os.listdir(f"{model_root}/{model_name}")
@@ -112,18 +108,10 @@ def load_hubert():
     return hubert_model.eval()
 
 def get_model_names():
+    model_root = "weights"  # Assuming this is where your models are stored
     return [d for d in os.listdir(model_root) if os.path.isdir(f"{model_root}/{d}")]
 
-# Initialize the global models
-hubert_model = load_hubert()
-rmvpe_model = RMVPE("rmvpe.pt", config.is_half, config.device)
-
-voice_mapping = {
-    "Mongolian Male": "mn-MN-BataaNeural",
-    "Mongolian Female": "mn-MN-YesuiNeural"
-}
-
-# Function to run async functions in a new event loop within a thread
+# Add this helper function to ensure a new event loop is created if none exists
 def run_async_in_thread(fn, *args):
     loop = asyncio.new_event_loop()
     asyncio.set_event_loop(loop)
@@ -132,8 +120,7 @@ def run_async_in_thread(fn, *args):
     return result
 
 def parallel_tts(tasks):
-    # Increase max_workers to better utilize CPU and GPU resources
-    with ThreadPoolExecutor(max_workers=8) as executor:  # Adjust based on your server capacity
+    with ThreadPoolExecutor() as executor:
         futures = [executor.submit(run_async_in_thread, tts, *task) for task in tasks]
         results = [future.result() for future in futures]
     return results
@@ -146,7 +133,7 @@ async def tts(
     use_uploaded_voice,
     uploaded_voice,
 ):
-    # Default values for parameters
+    # Default values for parameters used in EdgeTTS
     speed = 0  # Default speech speed
     f0_up_key = 0  # Default pitch adjustment
     f0_method = "rmvpe"  # Default pitch extraction method
@@ -200,7 +187,6 @@ async def tts(
             )
 
         f0_up_key = int(f0_up_key)
-        # Load the model
         tgt_sr, net_g, vc, version, index_file, if_f0 = model_data(model_name)
 
         # Setup for RMVPE or other pitch extraction methods
@@ -243,11 +229,20 @@ async def tts(
 
     except EOFError:
         info = (
-            "Output not valid. This may occur when input text and speaker do not match."
+            "output not valid. This may occur when input text and speaker do not match."
         )
         print(info)
         return info, None, None
     except Exception as e:
         traceback_info = traceback.format_exc()
         print(traceback_info)
-        return str(e), None, None
+        return str(e), None, None
+
+voice_mapping = {
+    "Mongolian Male": "mn-MN-BataaNeural",
+    "Mongolian Female": "mn-MN-YesuiNeural"
+}
+
+hubert_model = load_hubert()
+
+rmvpe_model = RMVPE("rmvpe.pt", config.is_half, config.device)