Deep Audio Generation Model Disruption Using Imperceptible Audio Modification

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional generative AI models lack effective mechanisms to protect copyrighted audio works from being exploited, leading to unauthorized reproduction and infringement.

Innovation Solution

An audio modification system that applies micro-level alterations to audio samples, using adaptive segmentation and machine learning models to generate modified audio that interferes with the encoding and diffusion mechanisms of generative models, ensuring the modified audio sounds similar to the original but produces unexpected and inferior results when used for new content generation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If generative AI models are trained on copyrighted audio works, then the models can generate realistic content that mimics existing works, but this leads to copyright infringement and unauthorized reproduction

Engineering Contradiction:

Improvegenerative AI model's ability to create realistic audio contentVSAvoidcopyright infringement and unauthorized reproduction

Core Design Contradiction:

Adaptability or versatilityVSObject-affected harmful factors

Solution Approach 1:

The system applies preliminary anti-action by generating modified audio samples before they can be used for training generative AI models. The audio modification system creates perturbed versions of copyrighted audio that, when used for training, prevent the models from learning accurate representations of the original works, thereby blocking copyright infringement before it occurs.

Inventive Principle:

Principle #9Preliminary anti-action

Solution Approach 2:

The modified audio samples serve as an intermediary between the copyrighted audio and the generative AI model training process. These modified samples appear to be legitimate training data to the model, but they contain embedded modifications that prevent the model from generating accurate reproductions of the original copyrighted works.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Reliability

If audio modification is applied to prevent generative AI mimicry, then copyright protection is achieved, but the modified audio may be detected as altered

Engineering Contradiction:

Improvecopyright protection effectivenessVSAvoiddetection of audio alteration

Core Design Contradiction:

ReliabilityVSObject-affected harmful factors

Solution Approach 1:

The audio modification system applies local quality changes by introducing modifications at specific, localized regions of the audio signal rather than altering the entire audio uniformly. This allows the modified audio to maintain its overall quality and sound natural to human listeners while containing specific perturbations that prevent generative AI models from accurately reproducing the original works.

Inventive Principle:

Principle #3Local quality

Solution Approach 2:

The system changes parameters of the audio signal in a controlled manner, adjusting characteristics such as amplitude, frequency, or temporal properties at specific locations. These parameter changes are subtle and localized, making them imperceptible to human listeners while effectively preventing the audio from being used to train accurate generative AI models.

Inventive Principle:

Principle #35Parameter changes

3Reliability

If existing legal recourse is used to protect copyrighted works, then copyright infringement can be addressed, but the process is time-consuming, difficult, and expensive

Engineering Contradiction:

Improvecopyright protectionVSAvoidtime and resources required for legal action

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The system replaces the mechanical/legal system of copyright protection with an automated technical solution. Instead of relying on legal recourse which is time-consuming and expensive, the audio modification system automatically processes copyrighted audio files, generates modified versions, and provides them as training data, thereby substituting human legal action with automated computational processes.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The audio modification system enables copyright holders to protect their own works automatically without requiring external legal intervention. The system can process copyrighted audio, generate protected modified versions, and provide them for training purposes, allowing the copyright holder to serve their own protection needs through an automated system rather than relying on legal institutions.

Inventive Principle:

Principle #25Self-service

Data Source

PatentUS20250329322A1Method to disrupt generation quality of deep audio generation models

Publication Date: 2025.10.23 SOURCEAUDIO HOLDINGS LLC

AI summary

An audio signal is segmented into a plurality of audio signal segments. A plurality of modified audio signal segments are generated based on processing data from the plurality of audio signal segments using a trained machine learning model. The plurality of modified audio signal segments are indistinguishable from the plurality of audio signal segments to the average human listener. A reconstructed audio signal corresponding to the audio signal is generated by combining the plurality of modified audio signal segments. The reconstructed audio signal is indistinguishable from the audio signal to the average human listener but, when used to train a generative machine learning model, constrains an ability of the trained generative machine learning model to generate new audio signals similar to the audio signal.