vchunk 1.0.0

Audio effects classifiers for voice

Voice Of Chunk

What do we want? Clean vocal acapellas! When do we want them? Now!

Cascaded Sorting

This repository hosts several models that are meant to cascade results. The output of the final classifier is audio ready to be used in training a Singing Voice Conversion (SVC) model.

State of Research

• Voice vs. Non-voice classifier
• Reverb vs. Dry classifier
• Polyphony vs. Monophony classifier
• Multi-voice unison vs. Single-voice unison

Installation

This package was developed using Python 3.11.4, and thus that version of python is recommended.

Clone this repository and make it the current directory in the terminal.

The, open a new virtual environment and install the requirements.

pip install -r requirements.txt

Model Weights

Download model weights from my dropbox.

cd voicequalityclassifier
wget https://www.dropbox.com/scl/fi/w6yqkhzpq4l8wpw53zrjk/weights.zip?rlkey=3ju9ofcn93wdklzc3sr8nlezz&dl=1 -O weights.zip
unzip weights.zip
cd ..

Usage

For some example audio download a few files with this command:

wget https://www.dropbox.com/scl/fi/sclv9z73xobw2hbagjqob/audio.zip?rlkey=3cq04ds7ymt5q7d4v1dzkfgib&dl=1 -O audio.zip
unzip audio.zip

The voicequalityclassifier is a package that can also be run as a module. To run the whole classification scheme, run

./env/bin/python voicequalityclassifier/main.py --input path/to/audio.wav

In the console, you will see the classification results. If you see:

for file path/to/audio.mp3 detected nonvoice

This means that no voice was detected (i.e. the source is an instrumental stem or stereo mix). If you see:

for file path/to/audio.mp3 detected voice
for file path/to/audio.mp3 detected reverb

This means there was reverb detected on the track.

Finally, if you see:

for file path/to/audio.mp3 detected voice
for file path/to/audio.mp3 detected dry

Then we're in business and we have a dry acapella file.

for file path/to/audio.mp3 detected voice
for file path/to/audio.mp3 detected reverb

This means there was reverb detected on the track.

Finally, if you see:

for file path/to/audio.mp3 detected voice
for file path/to/audio.mp3 detected dry

Then we're in business and we have a dry acapella file.