Delete app.py

app.py

@@ -1,89 +0,0 @@
-import os
-import torch
-import gradio as gr
-from pathlib import Path
-from transformers import AutoFeatureExtractor
-from optimum.onnxruntime import ORTModelForFeatureExtraction
-
-# 1. Environment & Backend Management
-# We force a check for libnppicc.so.13 to prevent crashes
-try:
-    import torchcodec
-    from torchcodec.decoders import VideoDecoder # Works for audio streams too
-    HAS_CODEC = True
-except Exception as e:
-    print(f"Warning: torchcodec load failed ({e}). Reverting to standard torchaudio.")
-    os.environ["TORCHAUDIO_USE_TORCHCODEC"] = "0"
-    HAS_CODEC = False
-
-import torchaudio
-
-# Configuration
-SAVE_DIR = "fastrvc_onnx_export"
-HF_HUBERT = "facebook/hubert-base-ls960" # Default for FastRVC 3.0
-DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
-
-def process_and_compile(audio_path):
-    if not audio_path:
-        return None, "Error: Audio file required."
-
-    try:
-        # A. Faster Loading with TorchCodec (if available)
-        if HAS_CODEC:
-            # torchcodec handles the GPU decoding and resampling efficiently
-            decoder = torchcodec.decoders.create_from_file(audio_path, device=DEVICE)
-            # Fetch all audio frames from the stream
-            waveform = decoder.get_frames_by_index(range(decoder.metadata.num_frames))
-            sample_rate = decoder.metadata.sample_rate
-        else:
-            waveform, sample_rate = torchaudio.load(audio_path)
-            waveform = waveform.to(DEVICE)
-
-        # B. Resampling (HuBERT requires 16000Hz)
-        if sample_rate != 16000:
-            resampler = torchaudio.transforms.Resample(sample_rate, 16000).to(DEVICE)
-            waveform = resampler(waveform)
-
-        # C. Hugging Face to ONNX Export (using Optimum)
-        # This compiles the voice feature extraction layer
-        print(f"Starting ONNX compilation for {HF_HUBERT}...")
-        
-        onnx_model = ORTModelForFeatureExtraction.from_pretrained(
-            HF_HUBERT,
-            export=True,
-            torch_dtype=torch.float32
-        )
-
-        # Create output directory and save the compiled binary
-        Path(SAVE_DIR).mkdir(parents=True, exist_ok=True)
-        onnx_model.save_pretrained(SAVE_DIR)
-        
-        final_model_path = os.path.join(SAVE_DIR, "model.onnx")
-        
-        return final_model_path, "✅ Compilation Successful: RVC-compatible ONNX generated."
-
-    except Exception as e:
-        return None, f"❌ Status: {str(e)}"
-
-# 3. Gradio Interface Definition
-with gr.Blocks(title="FastRVC 3.0 ONNX Compiler") as demo:
-    gr.Markdown("# 🎙️ FastRVC 3.0 Voice-to-ONNX Compiler")
-    gr.Markdown("Uses **torchcodec** for high-speed decoding and **HF Optimum** for ONNX compilation.")
-    
-    with gr.Row():
-        with gr.Column():
-            input_audio = gr.Audio(label="Voice Sample (Calibration)", type="filepath")
-            compile_btn = gr.Button("Compile Model", variant="primary")
-            
-        with gr.Column():
-            status_log = gr.Textbox(label="Build Status")
-            output_file = gr.File(label="Download .onnx Binary")
-
-    compile_btn.click(
-        fn=process_and_compile,
-        inputs=[input_audio],
-        outputs=[output_file, status_log]
-    )
-
-if __name__ == "__main__":
-    demo.launch(server_name="0.0.0.0", port=7860)