Re-Bottleneck: Latent Re-Structuring for Neural Audio Autoencoders
Abstract
A Re-Bottleneck framework modifies pre-trained autoencoders to enforce specific latent structures, improving performance in diverse downstream applications.
Neural audio codecs and autoencoders have emerged as versatile models for audio compression, transmission, feature-extraction, and latent-space generation. However, a key limitation is that most are trained to maximize reconstruction fidelity, often neglecting the specific latent structure necessary for optimal performance in diverse downstream applications. We propose a simple, post-hoc framework to address this by modifying the bottleneck of a pre-trained autoencoder. Our method introduces a "Re-Bottleneck", an inner bottleneck trained exclusively through latent space losses to instill user-defined structure. We demonstrate the framework's effectiveness in three experiments. First, we enforce an ordering on latent channels without sacrificing reconstruction quality. Second, we align latents with semantic embeddings, analyzing the impact on downstream diffusion modeling. Third, we introduce equivariance, ensuring that a filtering operation on the input waveform directly corresponds to a specific transformation in the latent space. Ultimately, our Re-Bottleneck framework offers a flexible and efficient way to tailor representations of neural audio models, enabling them to seamlessly meet the varied demands of different applications with minimal additional training.
Community
Re-Bottleneck is a framework that efficiently restructures the latent spaces of frozen pre-trained neural audio autoencoders. This allows for the post-hoc imposition of desired properties like channel ordering, semantic alignment, and transformation equivariance, improving downstream tasks such as text-to-audio generation while maintaining reconstruction quality.
Models citing this paper 0
No model linking this paper
Datasets citing this paper 0
No dataset linking this paper
Spaces citing this paper 0
No Space linking this paper
Collections including this paper 0
No Collection including this paper