# Non-autoregressive audio generation

MAGNeT is a text-to-music and text-to-sound model capable of generating high-quality audio samples based on text descriptions. It is a non-autoregressive Transformer model based on masked generation, using a 32kHz EnCodec tokenizer.