app.py

import gradio as gr
import os
import torch



def build_codec():
 ...

def build_model():
 ...

def ns2_inference(
 prmopt_audio_path,
 text,
 diffusion_steps=100,
):
 device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

demo_inputs = ...
demo_outputs = ...

demo = gr.Interface(
 fn=ns2_inference,
 inputs=demo_inputs,
 outputs=demo_outputs,
 title="Amphion Zero-Shot TTS NaturalSpeech2"
)

if __name__ == "__main__":
 demo.launch()