app.py

import gradio as gr
import requests
import soundfile as sf
import io
import torch
import torchaudio

API_URL = "https://hf.space/embed/akhaliq/denoise-audio/+/api/predict"

def enhance_vo(file, denoise_strength):
    # Step 1: Read audio and send to remote model
    with open(file, "rb") as f:
        response = requests.post(API_URL, files={"data": f})

    if response.status_code != 200:
        raise Exception("Denoising model failed to process audio.")

    # Step 2: Get denoised audio back
    response_data = response.json()
    url = response_data["data"][0]["url"]
    audio_response = requests.get(url)
    denoised_bytes = io.BytesIO(audio_response.content)

    # Step 3: Load both original and denoised audio
    orig_waveform, sr = torchaudio.load(file)
    denoised_waveform, _ = torchaudio.load(denoised_bytes)

    # Step 4: Blend based on slider value
    blend_ratio = denoise_strength / 100.0
    output = (1 - blend_ratio) * orig_waveform + blend_ratio * denoised_waveform
    output = output / output.abs().max()

    # Step 5: Save to file
    output_path = "enhanced_output.wav"
    torchaudio.save(output_path, output, sr)

    return output_path

# Gradio app UI
interface = gr.Interface(
    fn=enhance_vo,
    inputs=[
        gr.Audio(type="filepath", label="Upload MP3 or WAV"),
        gr.Slider(0, 100, value=100, label="Noise Reduction Strength (%)")
    ],
    outputs=gr.Audio(type="filepath", label="Enhanced Audio (WAV)"),
    title="Adobe-style VO Enhancer (Online Model)",
    description="Upload VO audio (MP3/WAV), adjust slider, and download cleaned WAV file."
)

interface.launch()