Audio Signal Denoising Chain That Preserves Phase Information

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional audio recording methods suffer from mediocre audio quality and require substantial manual effort for denoising, especially in noisy environments, often distorting phase information during noise reduction.

Innovation Solution

A computer-implemented method involving two denoising processes: a machine-learning based process that separates amplitude and phase information using partial convolutional layers, and a non-machine-learning based process that removes noise by setting thresholds based on quantiles, with resampling to match a specified sampling rate for improved denoising.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Object-affected harmful factors

If conventional noise reduction algorithms are used, then noise is reduced, but phase information is distorted

Engineering Contradiction:
ImprovenoiseVSAvoidphase information
Core Design Contradiction:
Object-affected harmful factorsVSLoss of information

Solution Approach 1:

The patent segments the noise reduction process into separate amplitude and phase processing paths. The amplitude envelope is extracted and processed independently from the phase information, allowing noise reduction in the amplitude domain while preserving phase integrity. This is achieved through Hilbert transform-based envelope extraction and separate filtering operations.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces an intermediary approach by using the Hilbert transform to create an analytic signal, which separates amplitude and phase components. This intermediary representation allows independent processing of amplitude (for noise reduction) and phase (for preservation), resolving the contradiction between noise removal and phase integrity.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Loss of information

If higher sampling rate is used, then phase information and harmonics are preserved, but computational power requirement increases significantly

Engineering Contradiction:
Improvephase informationVSAvoidcomputational power
Core Design Contradiction:
Loss of informationVSUse of energy by moving object

Solution Approach 1:

The patent changes the parameter of sampling rate adaptively. Instead of using a fixed high sampling rate, the system adjusts the sampling rate based on the specific audio signal characteristics and processing requirements. This allows optimization between computational cost and information preservation by selecting appropriate sampling rates for different scenarios.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent applies partial action by processing only the necessary portions of the signal at high resolution. Rather than uniformly processing the entire audio signal at maximum sampling rate, the system applies high-resolution processing selectively where needed, reducing overall computational burden while maintaining phase information integrity in critical frequency ranges.

Inventive Principle:
Principle #16Partial or excessive action

3Manufacturing precision

If manual denoising parameters are adjusted, then denoising quality is optimized, but processing time is consumed

Engineering Contradiction:
Improvedenoising qualityVSAvoidprocessing time
Core Design Contradiction:
Manufacturing precisionVSLoss of time

Solution Approach 1:

The patent implements self-service by enabling the system to automatically optimize denoising parameters without manual intervention. The algorithm autonomously adjusts filtering characteristics, threshold values, and processing parameters based on the input signal analysis, eliminating the need for manual parameter tuning while maintaining high denoising quality.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The patent applies preliminary action by pre-configuring optimal processing parameters and performing rapid signal analysis before the main denoising operation. The system prepares processing templates and characteristic analyses in advance, allowing the actual denoising to proceed more quickly with fewer iterative adjustments, thus reducing overall processing time.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4099324B1Methods for processing an audio signal and computer system
Publication Date: 2025.12.10 NOMONO AS
  • EP4099324B1 patent drawingFigure 1~2
  • EP4099324B1 patent drawingFigure 3~5
  • EP4099324B1 patent drawingFigure 6~7

AI summary

The invention relates to a computer-implemented method for processing an audio signal, in particular including speech. The method comprises obtaining a recorded audio signal having a dedicated sampling rate, in particular above 22 kHz, whereas the recorded audio signal comprises a speech portion and a noise portion. A spectrogram is generated from the recorded audio signal and then applied to a processing chain of two subsequent denoiser processes. A first of the two subsequent denoiser processes is a machine-learning based denoiser process to provide a denoised spectrogram. The second of the two subsequent denoiser process is a denoiser process configured to remove stationary noise from a spectrogram applied to it.