MagiCodec: Simple Masked Gaussian-Injected Codec for High-Fidelity Reconstruction and Generation

TL;DR

MagiCodec is a novel streaming Transformer-based audio codec that enhances semantic expressiveness and reconstruction quality by incorporating Gaussian noise injection and regularization, improving downstream generative tasks.

Contribution

It introduces MagiCodec, a single-layer streaming Transformer with a multistage training pipeline that improves tokenization for audio reconstruction and generation.

Findings

Outperforms state-of-the-art codecs in quality and downstream tasks.

Produces Zipf-like token distributions similar to natural language.

Analytically shown to attenuate high-frequency components through noise injection.

Abstract

Neural audio codecs have made significant strides in efficiently mapping raw audio waveforms into discrete token representations, which are foundational for contemporary audio generative models. However, most existing codecs are optimized primarily for reconstruction quality, often at the expense of the downstream modelability of the encoded tokens. Motivated by the need to overcome this bottleneck, we introduce $\textbf{MagiCodec}$, a novel single-layer, streaming Transformer-based audio codec. MagiCodec is designed with a multistage training pipeline that incorporates Gaussian noise injection and latent regularization, explicitly targeting the enhancement of semantic expressiveness in the generated codes while preserving high reconstruction fidelity. We analytically derive the effect of noise injection in the frequency domain, demonstrating its efficacy in attenuating high-frequency components and fostering robust tokenization. Extensive experimental evaluations show that MagiCodec surpasses state-of-the-art codecs in both reconstruction quality and downstream tasks. Notably, the tokens produced by MagiCodec exhibit Zipf-like distributions, as observed in natural languages, thereby improving compatibility with language-model-based generative architectures. The code and pre-trained models are available at https://github.com/Ereboas/MagiCodec.