Update app.py

app.py

@@ -50,6 +50,9 @@ def predict(prompt, upload_reference_audio, microphone_reference_audio, referenc
     )
     return gr.Audio(value=output_file_path, type='filepath')
 
+def create_inferencer():
+    return Inferencer()
+
 @spaces.GPU
 def main():
     if torch.cuda.is_available():
@@ -108,7 +111,7 @@ def main():
                         smoothed_t = gr.Slider(minimum=0, maximum=1, step=0.1, label="T", value=0.8)
                         kd_submit = gr.Button("Generate", elem_id="kdtalker_generate", variant="primary")
                         kd_submit.click(
-                                fn=Inferencer().generate_with_audio_img,
+                                fn=create_inferencer().generate_with_audio_img,
                                 inputs=[
                                     upload_driven_audio, tts_driven_audio, driven_audio_type, source_image,
                                     smoothed_pitch, smoothed_yaw, smoothed_roll, smoothed_t