Music Classifier

How it works

1. We gather a dataset of various genre music.
2. Calculate various metrics for each and save them to a database.
3. By using the same metrics and some cost functions we find most comparable music tracks

Basic Usage

Init Project

1. Create venv

python -m venv venv

2. Activate venv

./venv/Scripts/activate

3. Install requirements

pip install -r requirements.txt

Setup

1. To start the database localy you need docker running and then use

docker compose up -d

2. Once the database is up initialize the schema with

docker cp  ./dumps/schema.sql music-db:/schema.sql
docker exec music-db bash -c "psql -U root music < schema.sql"

3. Fill the databases

python ./utils/init_db.py

Example

• The script asks for a relative path to the track to classify and an integer for the number nearest tracks to find.
• These values can also be manually set by changing the variables in the find_nearest.py file.
• You can also set the desired label of the input track, in that case it will also pring the precission of each function used.

python find_nearest.py

The output will be in the out folder and the precision will be logged.

Configuring

Add a cost function

To add another cost function just add the implementation as a py file in the ./distance_calc/functions.py file and then add it to the enum and dictionary in the ./distance_calc/distance_funcs.py file.

Add music tracks to dataset

Music tracks to be used for filling up the database must be placed in the ./dataset directory in the appropriate directory based on its genre.

This must be done before running ./utils/init_db.py and if we need another genre, it needs to be added in the labels list in init_db.

The tracks used for testing were found here

Attributes for distance

To change the attributes used for distance calculation, change the `fetchable_attributes` in the  `./utils/fetch_entries.py`