High Quality Audio Coding with MDCTNet

TL;DR

This paper introduces MDCTNet, a neural audio generative model that operates in the perceptually weighted MDCT domain, capturing time-frequency correlations with RNNs, achieving high-quality audio coding at low bitrates.

Contribution

The paper presents MDCTNet, a novel neural model for audio coding that uses perceptual weighting and RNNs to improve compression quality at low bitrates.

Findings

Achieves similar subjective quality to Opus at half the bitrate.

Operates effectively on diverse fullband monophonic audio signals.

Utilizes a perceptual encoder conditioned generative model.

Abstract

We propose a neural audio generative model, MDCTNet, operating in the perceptually weighted domain of an adaptive modified discrete cosine transform (MDCT). The architecture of the model captures correlations in both time and frequency directions with recurrent layers (RNNs). An audio coding system is obtained by training MDCTNet on a diverse set of fullband monophonic audio signals at 48 kHz sampling, conditioned by a perceptual audio encoder. In a subjective listening test with ten excerpts chosen to be balanced across content types, yet stressful for both codecs, the mean performance of the proposed system for 24 kb/s variable bitrate (VBR) is similar to that of Opus at twice the bitrate.