Update app.py

app.py

@@ -1,16 +1,10 @@
+import spaces  # Import spaces first to avoid CUDA initialization issues
 import gradio as gr
 import torch
 from parler_tts import ParlerTTSForConditionalGeneration
 from transformers import AutoTokenizer
 import soundfile as sf
 import tempfile
-import spaces  # Import the spaces module for ZeroGPU compatibility
-
-# Load the model and tokenizers
-device = "cuda" if torch.cuda.is_available() else "cpu"
-model = ParlerTTSForConditionalGeneration.from_pretrained("ai4bharat/indic-parler-tts")
-tokenizer = AutoTokenizer.from_pretrained("ai4bharat/indic-parler-tts")
-description_tokenizer = AutoTokenizer.from_pretrained(model.config.text_encoder._name_or_path)
 
 # Supported languages and default settings
 languages = {
@@ -39,8 +33,10 @@ def generate_description(language, gender, emotion, noise, reverb, expressivity,
 # Generate audio function with GPU allocation
 @spaces.GPU  # Allocate GPU for the duration of this function
 def generate_audio(text, description):
-    # Move model to GPU
-    model.to("cuda")
+    # Load model and tokenizer
+    model = ParlerTTSForConditionalGeneration.from_pretrained("ai4bharat/indic-parler-tts").to("cuda")
+    tokenizer = AutoTokenizer.from_pretrained("ai4bharat/indic-parler-tts")
+    description_tokenizer = AutoTokenizer.from_pretrained(model.config.text_encoder._name_or_path)
 
     # Prepare model inputs
     input_ids = description_tokenizer(description, return_tensors="pt").input_ids.to("cuda")
@@ -55,9 +51,6 @@ def generate_audio(text, description):
         sf.write(f.name, audio_arr, model.config.sampling_rate)
         audio_path = f.name
 
-    # Move model back to CPU
-    model.to("cpu")
-
     return audio_path
 
 # Gradio Interface