app.py

import os
import re
import random
from scipy.io.wavfile import write
from scipy.io.wavfile import read
import numpy as np
import gradio as gr
import yt_dlp
import subprocess



mdxnet_models = [
    'UVR-MDX-NET-Inst_full_292.onnx',
    'UVR-MDX-NET_Inst_187_beta.onnx',
    'UVR-MDX-NET_Inst_82_beta.onnx',
    'UVR-MDX-NET_Inst_90_beta.onnx',
    'UVR-MDX-NET_Main_340.onnx',
    'UVR-MDX-NET_Main_390.onnx',
    'UVR-MDX-NET_Main_406.onnx',
    'UVR-MDX-NET_Main_427.onnx',
    'UVR-MDX-NET_Main_438.onnx',
    'UVR-MDX-NET-Inst_HQ_1.onnx',
    'UVR-MDX-NET-Inst_HQ_2.onnx',
    'UVR-MDX-NET-Inst_HQ_3.onnx',
    'UVR-MDX-NET-Inst_HQ_4.onnx',
    'UVR_MDXNET_Main.onnx',
    'UVR-MDX-NET-Inst_Main.onnx',
    'UVR_MDXNET_1_9703.onnx',
    'UVR_MDXNET_2_9682.onnx',
    'UVR_MDXNET_3_9662.onnx',
    'UVR-MDX-NET-Inst_1.onnx',
    'UVR-MDX-NET-Inst_2.onnx',
    'UVR-MDX-NET-Inst_3.onnx',
    'UVR_MDXNET_KARA.onnx',
    'UVR_MDXNET_KARA_2.onnx',
    'UVR_MDXNET_9482.onnx',
    'UVR-MDX-NET-Voc_FT.onnx',
    'Kim_Vocal_1.onnx',
    'Kim_Vocal_2.onnx',
    'Kim_Inst.onnx',
    'Reverb_HQ_By_FoxJoy.onnx',
    'UVR-MDX-NET_Crowd_HQ_1.onnx',
    'kuielab_a_vocals.onnx',
    'kuielab_a_other.onnx',
    'kuielab_a_bass.onnx',
    'kuielab_a_drums.onnx',
    'kuielab_b_vocals.onnx',
    'kuielab_b_other.onnx',
    'kuielab_b_bass.onnx',
    'kuielab_b_drums.onnx',
]



output_format = [
    'wav',
    'flac',
    'mp3',
]

mdxnet_overlap_values = [
    '0.25',
    '0.5',
    '0.75',
    '0.99',
]



def download_audio(url):
    ydl_opts = {
        'format': 'bestaudio/best',
        'outtmpl': 'ytdl/%(title)s.%(ext)s',
        'postprocessors': [{
            'key': 'FFmpegExtractAudio',
            'preferredcodec': 'wav',
            'preferredquality': '192',
        }],
    }

    with yt_dlp.YoutubeDL(ydl_opts) as ydl:
        info_dict = ydl.extract_info(url, download=True)
        file_path = ydl.prepare_filename(info_dict).rsplit('.', 1)[0] + '.wav'
        sample_rate, audio_data = read(file_path)
        audio_array = np.asarray(audio_data, dtype=np.int16)

        return sample_rate, audio_array


def mdxnet_separator(mdxnet_audio, mdxnet_model, mdxnet_output_format, mdxnet_segment_size, mdxnet_overlap, mdxnet_denoise):
  files_list = []
  files_list.clear()
  directory = "./outputs"
  random_id = str(random.randint(10000, 99999))
  pattern = f"{random_id}"
  os.makedirs("outputs", exist_ok=True)
  write(f'{random_id}.wav', mdxnet_audio[0], mdxnet_audio[1])
  prompt = f"audio-separator {random_id}.wav --model_filename {mdxnet_model} --output_dir=./outputs --output_format={mdxnet_output_format} --normalization=0.9 --mdx_segment_size={mdxnet_segment_size} --mdx_overlap={mdxnet_overlap}"
  
  if mdxnet_denoise:
    prompt += " --mdx_enable_denoise"
  
  os.system(prompt)

  for file in os.listdir(directory):
    if re.search(pattern, file):
      files_list.append(os.path.join(directory, file))

  stem1_file = files_list[0]
  stem2_file = files_list[1]

  return stem1_file, stem2_file


def mdxnet_batch(path_input, path_output, model, output_format, overlap, segment_size, denoise):
  found_files = []
  logs = []
  logs.clear()

  extensions = (".mp3", ".wav", ".flac")

  for audio_files in os.listdir(path_input):
    if audio_files.endswith(extensions):
      found_files.append(audio_files)
  total_files = len(found_files)

  if total_files == 0:
    logs.append("No valid audio files.")
    yield "\n".join(logs)
  else:
    logs.append(f"{total_files} audio files found")
    found_files.sort()

    for audio_files in found_files:
      file_path = os.path.join(path_input, audio_files)
      prompt = ["audio-separator", file_path, "-m", f"{model}", f"--output_dir={path_output}", f"--output_format={output_format}", "--normalization=0.9", f"--mdx_overlap={overlap}", f"--mdx_segment_size={segment_size}"]

      if denoise:
        prompt.append("--mdx_enable_denoise")

      logs.append(f"Processing file: {audio_files}")
      yield "\n".join(logs)
      subprocess.run(prompt)
      logs.append(f"File: {audio_files} processed!")
      yield "\n".join(logs)


with gr.Blocks(theme="Blane187/fuchsia", title="🎵 UVR5 MDX 🎵") as app:
    gr.Markdown("<h1> 🎵 UVR MDX 🎵 </h1>")
    gr.Markdown("If you liked this HF Space you can give me a ❤️")
    gr.Markdown("Try UVR5 UI using Colab [here](https://colab.research.google.com/github/Eddycrack864/UVR5-UI/blob/main/UVR_UI.ipynb)")
    with gr.Tabs():       
        with gr.TabItem("MDX-NET"):
            with gr.Row():
                mdxnet_model = gr.Dropdown(
                    label = "Select the Model",
                    choices = mdxnet_models,
                    interactive = True
                )
            with gr.Row():
                mdxnet_output_format = gr.Dropdown(
                    label = "Select the Output Format",
                    choices = output_format,
                    interactive = True
                )
            with gr.Row():
                mdxnet_segment_size = gr.Slider(
                    minimum = 32,
                    maximum = 4000,
                    step = 32,
                    label = "Segment Size",
                    info = "Larger consumes more resources, but may give better results.",
                    value = 256,
                    interactive = True
                )
            with gr.Row():
                mdxnet_overlap = gr.Dropdown(
                        label = "Overlap",
                        choices = mdxnet_overlap_values,
                        value = mdxnet_overlap_values[0],
                        interactive = True
                )
                mdxnet_denoise = gr.Checkbox(
                    label = "Denoise",
                    info = "Enable denoising during separation.",
                    value = True,
                    interactive = True
                )
            with gr.Row():
                mdxnet_audio = gr.Audio(
                    label = "Input Audio",
                    type = "numpy",
                    interactive = True
                )
            with gr.Accordion("Separation by Link", open = False):
                with gr.Row():
                    mdxnet_link = gr.Textbox(
                    label = "Link",
                    placeholder = "Paste the link here",
                    interactive = True
                )
                with gr.Row():
                   gr.Markdown("You can paste the link to the video/audio from many sites, check the complete list [here](https://github.com/yt-dlp/yt-dlp/blob/master/supportedsites.md)")
                with gr.Row():
                    mdxnet_download_button = gr.Button(
                    "Download!",
                    variant = "primary"
                )

            mdxnet_download_button.click(download_audio, [mdxnet_link], [mdxnet_audio])

            with gr.Accordion("Batch Separation", open = False):
                with gr.Row():
                    mdxnet_input_path = gr.Textbox(
                        label = "Input Path",
                        placeholder = "Place the input path here",
                        interactive = True
                    )
                    mdxnet_output_path = gr.Textbox(
                        label = "Output Path",
                        placeholder = "Place the output path here",
                        interactive = True
                    )
                with gr.Row():
                    mdxnet_bath_button = gr.Button("Separate!", variant = "primary")
                with gr.Row():
                    mdxnet_info = gr.Textbox(
                        label = "Output Information",
                        interactive = False
                    )

            mdxnet_bath_button.click(mdxnet_batch, [mdxnet_input_path, mdxnet_output_path, mdxnet_model, mdxnet_output_format, mdxnet_overlap, mdxnet_segment_size, mdxnet_denoise], [mdxnet_info])

            with gr.Row():
                mdxnet_button = gr.Button("Separate!", variant = "primary")
            with gr.Row():
                mdxnet_stem1 = gr.Audio(
                    show_download_button = True,
                    interactive = False,
                    label = "instrumental",
                    type = "filepath"
                )
                mdxnet_stem2 = gr.Audio(
                    show_download_button = True,
                    interactive = False,
                    label = "vocal",
                    type = "filepath"
                )

            mdxnet_button.click(mdxnet_separator, [mdxnet_audio, mdxnet_model, mdxnet_output_format, mdxnet_segment_size, mdxnet_overlap, mdxnet_denoise], [mdxnet_stem1, mdxnet_stem2])

        

        with gr.TabItem("Credits"):
           gr.Markdown(
              """
              UVR5 UI created by **[Eddycrack 864](https://github.com/Eddycrack864).** Join **[AI HUB](https://discord.gg/aihub)** community.
              * python-audio-separator by [beveradb](https://github.com/beveradb).
              * Special thanks to [Ilaria](https://github.com/TheStingerX) for hosting this space and help.
              * Thanks to [Mikus](https://github.com/cappuch) for the help with the code.
              * Thanks to [Nick088](https://huggingface.co./Nick088) for the help to fix roformers.
              * Thanks to [yt_dlp](https://github.com/yt-dlp/yt-dlp) devs.
              * Separation by link source code and improvements by [Blane187](https://huggingface.co./Blane187).

              You can donate to the original UVR5 project here:
              [!["Buy Me A Coffee"](https://www.buymeacoffee.com/assets/img/custom_images/orange_img.png)](https://www.buymeacoffee.com/uvr5)
              """
           )


app.launch(share=True, debug=True)