Audio Signal Processor Using Autoregressive Reverberation Modeling

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current audio signal processing technologies face challenges in effectively reducing noise and reverberation in real-time, particularly in distant speech communication scenarios, where existing methods often degrade audio quality and are computationally complex, and struggle with integrating additive noise without affecting neighboring time-frames.

Innovation Solution

A signal processor that estimates coefficients of an autoregressive reverberation model using delayed noise-reduced reverberant signals, allowing for separate and alternating estimation of coefficients and noise-reduced signals, thereby maintaining good audio quality while reducing computational complexity.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Object-affected harmful factors

If conventional noise reduction and dereverberation methods are applied, then noise and reverberation are reduced, but audio quality degrades and computational complexity increases

Engineering Contradiction:
Improvenoise and reverberationVSAvoidaudio quality
Core Design Contradiction:
Object-affected harmful factorsVSManufacturing precision

Solution Approach 1:

The patent segments the audio signal processing into distinct frequency bands using Short-Time Fourier Transform (STFT), applying separate autoregressive models to each band. This segmentation allows targeted noise and reverberation reduction in each frequency range while preserving the unique characteristics of different bands, thereby maintaining overall audio quality while reducing harmful factors.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent changes the parameter representation by using autoregressive coefficients to model the reverberation impulse response instead of directly processing the time-domain signal. By estimating and manipulating these coefficients in the frequency domain, the system achieves effective dereverberation with controlled computational complexity and preserved audio fidelity.

Inventive Principle:
Principle #35Parameter changes

2Object-affected harmful factors

If conventional noise reduction and dereverberation methods are applied, then noise and reverberation are reduced, but computational complexity increases

Engineering Contradiction:
Improvenoise and reverberationVSAvoidcomputational complexity
Core Design Contradiction:
Object-affected harmful factorsVSDevice complexity

Solution Approach 1:

The patent divides the computational task into independent frequency band processing using STFT segmentation. Each band is processed separately with its own autoregressive model, allowing parallel computation and reducing the overall computational burden compared to processing the entire spectrum simultaneously, while still achieving comprehensive noise and reverberation reduction.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent replaces direct time-domain signal processing with frequency-domain autoregressive modeling. Instead of applying complex time-domain filters and iterative optimization algorithms, the system uses efficient frequency-domain convolution and autoregressive coefficient estimation, significantly reducing computational complexity while maintaining effective dereverberation performance.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

3Object-affected harmful factors

If additive noise is removed in advance, then noise reduction is achieved, but the relations between neighboring time-frames of the reverberant signal are destroyed

Engineering Contradiction:
Improveadditive noiseVSAvoidrelations between neighboring time-frames
Core Design Contradiction:
Object-affected harmful factorsVSStability of the object's composition

Solution Approach 1:

The patent applies preliminary noise reduction to obtain an initial noise-reduced signal, then uses this signal to estimate autoregressive coefficients that capture the temporal relationships. By performing coefficient estimation on the noise-reduced signal rather than directly filtering the original signal, the method preserves the underlying temporal structure and inter-frame relationships while removing noise.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent introduces autoregressive coefficients as an intermediary representation that mediates between noise reduction and temporal relationship preservation. These coefficients are estimated from the noise-reduced signal and then used to reconstruct the reverberant signal, ensuring that noise removal does not destroy the temporal correlations between neighboring time-frames.

Inventive Principle:
Principle #24Intermediary (Mediator)

4Manufacturing precision

If batch algorithms are used for multichannel linear prediction, then accurate dereverberation is achieved, but real-time processing capability is lost

Engineering Contradiction:
Improvedereverberation accuracyVSAvoidreal-time processing speed
Core Design Contradiction:
Manufacturing precisionVSProductivity

Solution Approach 1:

The patent transforms the static batch processing approach into a dynamic online algorithm that processes audio frames sequentially in real-time. The autoregressive coefficients are updated adaptively for each time frame based on incoming data, allowing the system to maintain high dereverberation accuracy while achieving real-time processing capability through incremental updates rather than batch computation.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent changes the computational approach by using recursive least squares or similar efficient algorithms to update autoregressive coefficients incrementally frame-by-frame. This parameter update strategy maintains the accuracy benefits of batch processing while achieving real-time performance, as each frame's coefficients are computed based on previous estimates without requiring complete reprocessing of all data.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP3685378B1Signal processor and method for providing a processed audio signal reducing noise and reverberation
Publication Date: 2021.10.13 FRAUNHOFER GESELLSCHAFT ZUR FORDERUNG DER ANGEWANDTEN FORSCHUNG EV
  • EP3685378B1 patent drawingFigure 1
  • EP3685378B1 patent drawingFigure 2
  • EP3685378B1 patent drawingFigure 3

AI summary

A signal processor for providing one or more processed audio signals on the basis of one or more input audio signals is configured to estimate coefficients of an autoregressive reverberation model using the input audio signals and the delayed noise-reduced reverberant signals obtained using a noise reduction. The signal processor is configured to provide noise-reduced reverberant signals using the input audio signals and the estimated coefficients of the autoregressive reverberation model. The signal processor is configured to derive noise-reduced and reverberation-reduced output signals using the noise-reduced reverberant signals and the estimated coefficients of the autoregressive reverberation model. A method and a computer program comprise a similar functionality.