use zeroGPU

app.py

@@ -204,15 +204,15 @@ demo = gr.Interface(
             [str(Path("examples/1310.4546v1.pdf"))],
             "",
             "Fun",
-            "Female",
-             "Medium (3-5 min)",
+            "Male",
+            "Medium (3-5 min)",
             "English"
         ],
         [
             [],
             "https://en.wikipedia.org/wiki/Hugging_Face",
             "Fun",
-            "Male"
+            "Male",
             "Short (1-2 min)",
             "English"
         ],
@@ -220,7 +220,7 @@ demo = gr.Interface(
             [],
             "https://simple.wikipedia.org/wiki/Taylor_Swift",
             "Fun",
-            "Female"
+            "Female",
             "Short (1-2 min)",
             "English"
         ],

requirements.txt

@@ -7,4 +7,6 @@ promptic==0.7.5
 pydantic==2.7
 pypdf==4.1
 sentry-sdk==2.5
+spaces==0.30.2
+
 tenacity==8.3

utils.py

@@ -13,6 +13,7 @@ import tempfile
 
 
 import soundfile as sf
+import spaces
 import torch
 from gradio_client import Client
 from openai import OpenAI
@@ -78,7 +79,7 @@ def parse_url(url: str) -> str:
     response = requests.get(full_url, timeout=60)
     return response.text
 
-
+@spaces.GPU
 def generate_audio(text: str, speaker: str, language: str, voice: str) -> str:
     """Generate audio using the local Parler TTS model or HuggingFace client."""