Hidden Echoes Survive Training in Audio To Audio Generative Instrument Models

TL;DR

This paper demonstrates that imperceptible echoes hidden in training data can be reproduced by various audio generative models, revealing potential for watermarking and understanding model behavior.

Contribution

It introduces a method for embedding hidden echoes in training data that persist in model outputs, aiding model attribution and watermarking in audio synthesis.

Findings

Hidden echoes are robustly reproduced across multiple architectures.

Longer echo patterns increase information capacity.

Echoes survive fine-tuning, mixing, and pitch shift augmentations.

Abstract

As generative techniques pervade the audio domain, there has been increasing interest in tracing back through these complicated models to understand how they draw on their training data to synthesize new examples, both to ensure that they use properly licensed data and also to elucidate their black box behavior. In this paper, we show that if imperceptible echoes are hidden in the training data, a wide variety of audio to audio architectures (differentiable digital signal processing (DDSP), Realtime Audio Variational autoEncoder (RAVE), and ``Dance Diffusion'') will reproduce these echoes in their outputs. Hiding a single echo is particularly robust across all architectures, but we also show promising results hiding longer time spread echo patterns for an increased information capacity. We conclude by showing that echoes make their way into fine tuned models, that they survive mixing/demixing, and that they survive pitch shift augmentation during training. Hence, this simple, classical idea in watermarking shows significant promise for tagging generative audio models.