Musenet (OpenAI)

MuseNet, developed by OpenAI, is a sophisticated neural network capable of creating 4-minute musical pieces using 10 different instruments and blending styles ranging from country to Mozart to the Beatles. It operates with the same versatile unsupervised technology as GPT-2, a vast transformer model designed to forecast the next token in a sequence, applicable to both audio and text. The model learns from MIDI file data and can produce samples in a selected style by beginning with a prompt. It utilizes multiple embeddings, including positional, timing, and structural embeddings, to provide the model with additional context.