Youtube Livestream Vocal Isolator

Livestream playback with on the fly vocal isolation

About

Allows you to listen to a stream (more specifically, the vocals from it) in the background while doing something else. The output will generally be delayed by a few seconds compared to the Youtube player (but might depending on stream settings be slightly ahead).

Notes:

• Non-vocals likely won't be affected (significantly) when mixed with vocals. Non-vocals on their own mostly disappear, though.
• When run, the program takes up almost 3800 MB of RAM and, if the GPU can be used, 10% of its VRAM (configurable).

Demo (with GPU support): To be added

Usage

0. Extract main.py and settings.py to a folder of your choice. Upon running main.py for the first time, the folders __pycache__ and pretrained_models will be created in the same directory (totalling ~76 MB).

1. In settings.py, give TEMP_FOLDER a path to a folder where temporary files will be stored (e.g. "E:\temp"). The path does not need to exist before executing the program -- folders will be created if needed.

• You might not want to use an SSD for this (as they have a limited number of write cycles). Consider using a RAM disk (using e.g. ImDisk) -- a size of 32 MB should be enough with the default settings.

2. Double click main.py, or via the command line:

  python main.py [URL] [FORMAT CODE]

 where

• URL points to an ongoing livestream (e.g. https://www.youtube.com/watch?v=21X5lGlDOfg)
• FORMAT CODE indicates the quality (see available format codes by running the script or via youtube-dl -F {URL}).

3. Press Ctrl+C to close the program (or otherwise, but you will have to remove leftover temporary files yourself). You might need to do this twice.

Dependencies

• Python 3.7+
• ffmpeg
• youtube-dl

Install via e.g. pip install {package}:

• m3u8
• pydub
• spleeter

Optional: GPU support

Settings

Notable settings in settings.py:

• TEMP_FOLDER -- where temporary files should be stored.
• DEFAULT_QUALITY -- default format code: "91"
• ASK_FOR_QUALITY -- whether to prompt for a format code if not given one as a cmd argument (uses DEFAULT_QUALITY if disabled)
• SEG_START -- indicates minimum delay depending on the stream setup (segment length).
• TF_MEMORY_FRACTION -- (GPU) limits how much VRAM TensorFlow (used by Spleeter) is allowed to allocate