Self-Supervised VQ-VAE for One-Shot Music Style Transfer
It addresses the problem of one-shot style transfer for music audio, which is incremental as it builds on existing VQ-VAE and style transfer techniques.
The paper tackles one-shot timbre transfer in music audio by proposing a novel method based on an extended VQ-VAE with self-supervised learning for disentangled representations, showing it outperforms selected baselines in objective metrics.
Neural style transfer, allowing to apply the artistic style of one image to another, has become one of the most widely showcased computer vision applications shortly after its introduction. In contrast, related tasks in the music audio domain remained, until recently, largely untackled. While several style conversion methods tailored to musical signals have been proposed, most lack the 'one-shot' capability of classical image style transfer algorithms. On the other hand, the results of existing one-shot audio style transfer methods on musical inputs are not as compelling. In this work, we are specifically interested in the problem of one-shot timbre transfer. We present a novel method for this task, based on an extension of the vector-quantized variational autoencoder (VQ-VAE), along with a simple self-supervised learning strategy designed to obtain disentangled representations of timbre and pitch. We evaluate the method using a set of objective metrics and show that it is able to outperform selected baselines.