Re-Bottleneck: Latent Re-Structuring for Neural Audio Autoencoders
This work addresses a key limitation in neural audio models for applications like compression and generation, offering a flexible solution to tailor representations, though it is incremental as it builds on existing autoencoder frameworks.
The paper tackles the problem of neural audio autoencoders lacking specific latent structures for diverse downstream applications by proposing a post-hoc Re-Bottleneck framework that modifies pre-trained autoencoders to instill user-defined structures, demonstrating effectiveness in experiments such as enforcing latent ordering, aligning with semantic embeddings, and introducing equivariance without sacrificing reconstruction quality.
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.