Create app.py

app.py

@@ -0,0 +1,63 @@
+import os
+import torch
+import gradio as gr
+from pathlib import Path
+from transformers import AutoFeatureExtractor
+from optimum.onnxruntime import ORTModelForFeatureExtraction
+
+# Backend Safety Check
+try:
+    import torchcodec
+    HAS_CODEC = True
+except Exception:
+    os.environ["TORCHAUDIO_USE_TORCHCODEC"] = "0"
+    HAS_CODEC = False
+
+import torchaudio
+
+SAVE_DIR = "fastrvc_export"
+HF_MODEL = "facebook/hubert-base-ls960"
+
+def run_compilation(audio_path):
+    if not audio_path:
+        return None, "Please upload a voice sample."
+    
+    try:
+        # Step 1: Load Audio
+        waveform, sr = torchaudio.load(audio_path)
+        if sr != 16000:
+            waveform = torchaudio.transforms.Resample(sr, 16000)(waveform)
+        
+        # Step 2: Compile to ONNX
+        # Using Optimum to handle the complex HF -> ONNX mapping
+        model = ORTModelForFeatureExtraction.from_pretrained(
+            HF_MODEL, 
+            export=True, 
+            torch_dtype=torch.float32
+        )
+        
+        Path(SAVE_DIR).mkdir(parents=True, exist_ok=True)
+        model.save_pretrained(SAVE_DIR)
+        
+        onnx_file = os.path.join(SAVE_DIR, "model.onnx")
+        return onnx_file, "Success! Model compiled for FastRVC 3.0."
+        
+    except Exception as e:
+        return None, f"Error: {str(e)}"
+
+# Gradio 6.10.0 UI
+with gr.Blocks(title="FastRVC 3.0 Compiler") as demo:
+    gr.Markdown("# 🎙️ FastRVC 3.0 Voice-to-ONNX")
+    
+    with gr.Row():
+        with gr.Column():
+            audio_in = gr.Audio(label="Input Voice", type="filepath")
+            btn = gr.Button("Compile Now", variant="primary")
+        with gr.Column():
+            status = gr.Textbox(label="System Log")
+            file_out = gr.File(label="Download ONNX")
+
+    btn.click(run_compilation, inputs=audio_in, outputs=[file_out, status])
+
+if __name__ == "__main__":
+    demo.launch()