'Spleeter' that can individually extract vocal drum bass sound from music data



Spleeter ”, which outputs music data composed of vocals and multiple musical instruments as files classified into each sound using machine learning, is available on GitHub . I was able to extract only the voice and accompaniment part of the vocal, so I actually used it.

deezer / spleeter: Deezer source separation library including pretrained models.
https://github.com/deezer/spleeter

To use spleeter, 'Conda' must be installed. First, access the following URL.

Conda — Conda documentation
https://docs.conda.io/en/latest/

Click “Conda-build”.


Click “Installing and updating conda-build”.


Click “install conda”.


This time, click “Windows” to use spleeter on Windows.


spleeter can be run with either 'Anaconda' or 'Miniconda'. Since Miniconda is used this time, click “Miniconda installer for Windows”.


Click the 64-bit version or the 32-bit version according to the PC you are using. This time, download 64bit version.


When Miniconda is downloaded to the desktop, an icon like an image is displayed. Click it to execute it.


Click “Run”.


Click “Next”.


Click “I Agree”.


Select the folder where Miniconda will be installed and click “Next”.


Click “Finish” to complete the installation.


Click on the installed Miniconda to launch it.


You can sort music data by instrument by typing the commands in the red frame below into Miniconda in order.


Start Miniconda, enter 'git clone https://github.com/Deezer/spleeter' to copy the spleeter file from GitHub, and execute it with the enter key.


Next, input 'conda env create -f spleeter / conda / spleeter-cpu.yaml' to set the environment of Conda and execute with the enter key.


Enter 'conda activate spleeter-cpu' to execute spleeter and execute with enter key.


When spleeter is ready to run, enter `` spleeter separate -i (folder path of music data to be sorted) -p spleeter: (classification method) -o output '' to create data extracted for each instrument. .

There were three classification methods available at the time of article creation.
2stems: Classified as vocal / accompaniment.
4stems: Vocal / Drum / Bass / Other instruments
5stems: Vocal / Drum / Bass / Piano / Other instruments

For example, if you want to classify the sample data `` audio_example.mp3 '' that comes with spleeter into two types, vocal and accompaniment, enter `` spleeter separate -i spleeter / audio_example.mp3 -p spleeter: 2stems -o output '' The


Data is output to a folder called “output” created in the same location as the spleeter folder.


A folder is created with the name of the original music data, so click on the folder ...


The vocal part was output with 'vocals.wav' and the accompaniment part was output with 'accompaniment.wav'.


As a test, try extracting the following MP3 files separately for vocals and accompaniment. The data before extraction looks like this.


Here is an extract of vocals only. The sound of breathing is well extracted, and the accompaniment part is completely inaudible.


The accompaniment part is as follows. Only a few high-pitched vocals remain, but only accompaniment.


Note that songs such as “Lururu” or “Larara” whose lyrics are not words may not be extracted well.

in Review,   Software,   Web Service, Posted by log1m_mn