Data
Dataset
It turns out that the Music Genre Recognition remains a compelling problem, and there are several very nice and well-formatted dataset available online. In our project, we use a dataset from Marsyas (Music Analysis, Retrieval, and Synthesis for Audio Signals), an open source software framework for audio processing with specific emphasis on Music Information Retrieval Applications.
The dataset consists of 1000 .au files (a file extension used for audio files), split in 10 different music genres (Disco, Metal, Pop, Hip Hop, Rock, Blues, Classical, Country, Jazz, Reggae), with 100 samples of 30 seconds for each genre. Each sample is a 22050Hz Mono 16-bit audio file.
Data Processing
The main tool we use to do DSP for audio files, and further implement machine learning methods is Matlab. We use Matlab and its built-in functions, including audioread(), to read in the audio file dataset, and dump it as .mat file for further processing. For any detail, please refer to the Matlab code.