AEROMamba: An efficient architecture for audio super-resolution using generative adversarial networks and state space models

TL;DR

AEROMamba introduces a novel architecture for audio super-resolution that replaces attention and LSTM modules with a state space model, significantly reducing memory usage and increasing speed while improving subjective audio quality.

Contribution

The paper presents AEROMamba, a new model that replaces traditional modules with a state space model, achieving faster inference, lower memory consumption, and better audio quality in super-resolution tasks.

Findings

14x faster inference compared to AERO

5x less GPU memory during training

Superior subjective listening scores on MUSDB and PianoEval datasets

Abstract

Audio super-resolution aims to enhance low-resolution signals by creating high-frequency content. In this work, we modify the architecture of AERO (a state-of-the-art system for this task) for music super-resolution. SPecifically, we replace its original Attention and LSTM layers with Mamba, a State Space Model (SSM), across all network layers. Mamba is capable of effectively substituting the mentioned modules, as it offers a mechanism similar to that of Attention while also functioning as a recurrent network. With the proposed AEROMamba, training requires 2-4x less GPU memory, since Mamba exploits the convolutional formulation and leverages GPU memory hierarchy. Additionally, during inference, Mamba operates in constant memory due to recurrence, avoiding memory growth associated with Attention. This results in a 14x speed improvement using 5x less GPU. Subjective listening tests (0 to 100 scale) show that the proposed model surpasses the AERO model. In the MUSDB dataset, degraded signals scored 38.22, while AERO and AEROMamba scored 60.03 and 66.74, respectively. For the PianoEval dataset, scores were 72.92 for degraded signals, 76.89 for AERO, and 84.41 for AEROMamba.