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

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Song, Yakun, Chen, Jiawei, Zhuang, Xiaobin, Du, Chenpeng, Ma, Ziyang, Wu, Jian, Cong, Jian, Jia, Dongya, Chen, Zhuo, Wang, Yuping, Wang, Yuxuan, Chen, Xie
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915315754991616
author Song, Yakun
Chen, Jiawei
Zhuang, Xiaobin
Du, Chenpeng
Ma, Ziyang
Wu, Jian
Cong, Jian
Jia, Dongya
Chen, Zhuo
Wang, Yuping
Wang, Yuxuan
Chen, Xie
author_facet Song, Yakun
Chen, Jiawei
Zhuang, Xiaobin
Du, Chenpeng
Ma, Ziyang
Wu, Jian
Cong, Jian
Jia, Dongya
Chen, Zhuo
Wang, Yuping
Wang, Yuxuan
Chen, Xie
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2506_00385
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle MagiCodec: Simple Masked Gaussian-Injected Codec for High-Fidelity Reconstruction and Generation
Song, Yakun
Chen, Jiawei
Zhuang, Xiaobin
Du, Chenpeng
Ma, Ziyang
Wu, Jian
Cong, Jian
Jia, Dongya
Chen, Zhuo
Wang, Yuping
Wang, Yuxuan
Chen, Xie
Sound
Artificial Intelligence
Machine Learning
Audio and Speech Processing
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.
title MagiCodec: Simple Masked Gaussian-Injected Codec for High-Fidelity Reconstruction and Generation
topic Sound
Artificial Intelligence
Machine Learning
Audio and Speech Processing
url https://arxiv.org/abs/2506.00385