Hybrid Auditory Filterbanks for Stable Adaptive Audio Encoding

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing audio processing techniques face challenges in achieving a balance between interpretability and adaptability, particularly in encoder-mask-decoder settings, with fixed transforms being inflexible and data-driven methods being unstable and hard to interpret.

Innovation Solution

A hybrid auditory filterbank is introduced, combining fixed and trainable filters to enhance stability and adaptability, using an auditory filterbank structure with trainable convolutional layers and a penalty term to maintain tightness, ensuring numerical stability and efficient reconstruction.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If fixed time-frequency transforms (STFT, CQT) are used as encoders, then interpretability and control are improved, but adaptability to different tasks deteriorates

Engineering Contradiction:
ImproveinterpretabilityVSAvoidtask adaptability
Core Design Contradiction:
Ease of operationVSAdaptability or versatility

Solution Approach 1:

The patent merges fixed time-frequency transforms with trainable neural network components to create a hybrid encoder architecture. The fixed transform provides interpretable frequency decomposition while trainable layers adapt to specific tasks, combining the advantages of both fixed and data-driven approaches.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The patent introduces dynamic adaptability by making parts of the encoder trainable while keeping other parts fixed. This allows the system to adapt its parameters dynamically during training for different tasks while maintaining the structured interpretability of the fixed transform components.

Inventive Principle:
Principle #15Dynamics

2Adaptability or versatility

If data-driven feature extraction methods are used, then flexibility and adaptability are improved, but stability and interpretability deteriorate

Engineering Contradiction:
ImproveflexibilityVSAvoidstability
Core Design Contradiction:
Adaptability or versatilityVSReliability

Solution Approach 1:

The patent combines data-driven trainable components with fixed, stable transform operations. The trainable parts provide flexibility and adaptability while the fixed transform components ensure numerical stability and interpretability, resolving the contradiction between flexibility and stability.

Inventive Principle:
Principle #5Merging (Combining)

3Adaptability or versatility

If purely trainable filterbanks are used, then adaptability to tasks is improved, but numerical stability and reconstruction quality deteriorate

Engineering Contradiction:
Improvetask optimizationVSAvoidnumerical stability
Core Design Contradiction:
Adaptability or versatilityVSReliability

Solution Approach 1:

The patent applies different properties to different parts of the filterbank: some filters are fully trainable for task optimization while others are fixed to ensure numerical stability. This local differentiation allows simultaneous achievement of adaptability and stability.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The filterbank is constructed as a composite structure combining trainable and fixed components, analogous to composite materials. This hybrid construction allows the system to benefit from both the adaptability of trainable filters and the stability of fixed transforms.

Inventive Principle:
Principle #40Composite materials

Data Source

PatentEP4700769A1Stable hybrid auditory filterbanks
Publication Date: 2026.02.25 OESTERRISCHE ACAD DER WISSENSCHAFTEN
  • EP4700769A1 patent drawingFigure 1
  • EP4700769A1 patent drawingFigure 2~4
  • EP4700769A1 patent drawingFigure 3

AI summary

Disclosed is a hybrid auditory filterbank (100) configured for audio processing implemented on a data processing apparatus. The hybrid auditory filterbank (100) comprises a plurality of filters (102) configured to decompose an input audio signal (104) into a plurality of sub-bands (106). The filters (102) are each based on a fixed filter (108) and a trainable filter (110). The fixed filters (108) each comprise a filter of an auditory filterbank. The trainable filters (110) have been trained to perform an audio processing task and to improve the stability of the hybrid auditory filterbank (100).