MSSE is a content-based similarity search engine and web application for audio similarity retrieval. It was written as a semestral project at the Czech Technical University in Prague. The application's main purpose is to allow quick prototyping and testing of different similarity measures and methods of retrieval.
Python 3 is required. First, install the requirements:
pip install -r requirements.txt
Libsamplerate and its Python wrapper are recommended for efficient processing. Unfortunately, the official wrapper doesn't support Python 3: please install this fork.
To set up the application SQLite database and preprocess data files for retrieval, use the following commands:
python create_database.py
python preprocess.py engine_name dataset_name
The engine name can be any engine class name from the engine.py file. A reasonably fast and reliable engine is the MandelEllisEngine. Your reference files (the ones being searched through) should be placed in webapp/data/[dataset_name]. A good reference dataset is GTZAN. If unzipped into webapp/data/genres, the preprocessing command for the MandelEllis engine would then look like this:
python preprocess.py genres MandelEllisEngine
To launch the application, run python webapp.py and connect to it on port 8000.