Multi-band Frequency Reconstruction for Neural Psychoacoustic Coding

TL;DR

MUFFIN is a novel neural psychoacoustic coding framework that uses multi-band spectral residual vector quantization and transformer-inspired architecture to achieve high-fidelity, efficient audio compression and effective downstream task performance.

Contribution

Introduces MUFFIN, a fully convolutional neural psychoacoustic coding framework with multi-band spectral residual vector quantization and a transformer-inspired backbone for improved audio compression.

Findings

Outperforms existing methods in reconstruction quality on benchmarks.

Achieves a state-of-the-art 12.5 Hz compression rate with minimal quality loss.

Effective as a token representation for downstream generative tasks.

Abstract

Achieving high-fidelity audio compression while preserving perceptual quality across diverse content remains a key challenge in Neural Audio Coding (NAC). We introduce MUFFIN, a fully convolutional Neural Psychoacoustic Coding (NPC) framework that leverages psychoacoustically guided multi-band frequency reconstruction. At its core is a Multi-Band Spectral Residual Vector Quantization (MBS-RVQ) module that allocates bitrate across frequency bands based on perceptual salience. This design enables efficient compression while disentangling speaker identity from content using distinct codebooks. MUFFIN incorporates a transformer-inspired convolutional backbone and a modified snake activation to enhance resolution in fine-grained spectral regions. Experimental results on multiple benchmarks demonstrate that MUFFIN consistently outperforms existing approaches in reconstruction quality. A high-compression variant achieves a state-of-the-art 12.5 Hz rate with minimal loss. MUFFIN also proves effective in downstream generative tasks, highlighting its promise as a token representation for integration with language models. Audio samples and code are available.