Hierarchical Timbre-Painting and Articulation Generation

TL;DR

This paper introduces a fast, high-fidelity music generation method that synthesizes instrument-specific audio from specified pitch and loudness, utilizing learned source-filtering networks with multi-resolution spectral and adversarial losses.

Contribution

The authors propose a novel hierarchical source-filtering architecture for timbre and articulation generation, achieving state-of-the-art timbre transfer with minimal sample data.

Findings

High-quality instrument fitting with only minutes of sample data

State-of-the-art timbre transfer capabilities

Efficient multi-resolution spectral optimization

Abstract

We present a fast and high-fidelity method for music generation, based on specified f0 and loudness, such that the synthesized audio mimics the timbre and articulation of a target instrument. The generation process consists of learned source-filtering networks, which reconstruct the signal at increasing resolutions. The model optimizes a multi-resolution spectral loss as the reconstruction loss, an adversarial loss to make the audio sound more realistic, and a perceptual f0 loss to align the output to the desired input pitch contour. The proposed architecture enables high-quality fitting of an instrument, given a sample that can be as short as a few minutes, and the method demonstrates state-of-the-art timbre transfer capabilities. Code and audio samples are shared at https://github.com/mosheman5/timbre_painting.