Super-wideband Noise Suppression via Band-splitting

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current methods lack effectiveness in noise suppression for super-wideband audio signals, particularly in the high-frequency band, as they struggle to distinguish noise from speech energy, leading to over- or under-suppression, resulting in low-quality noise suppression.

Innovation Solution

A noise suppression system that splits super-wideband input signals into low-frequency and high-frequency bands, performing noise suppression in the frequency domain on the low-frequency band to generate data used for time-domain processing in the high-frequency band, using a time-domain filter bank and frame-based gains to effectively suppress noise without affecting the richer speech sound.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Object-affected harmful factors

If noise suppression is applied to the high-frequency band, then noise is reduced, but speech quality deteriorates due to over-suppression or under-suppression

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

Solution Approach 1:

The patent divides the super-wideband input signal into two separate frequency bands: low-frequency band (L-band) and high-frequency band (H-band). The L-band is processed in the frequency domain while the H-band is processed in the time domain. This segmentation allows different processing strategies to be applied to each band, enabling effective noise suppression in the H-band without degrading speech quality, as the L-band provides reliable speech energy reference that prevents over-suppression.

Inventive Principle:
Principle #1Segmentation

2Object-affected harmful factors

If noise suppression is applied to super-wideband signals, then noise is reduced, but computational complexity increases

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

Solution Approach 1:

The patent segments the processing into two distinct paths: frequency domain processing for L-band and time domain processing for H-band. This segmentation allows the system to avoid the high computational complexity of full frequency-domain processing for the entire super-wideband signal, while still achieving effective noise suppression.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent uses an intermediary approach where noise suppression data is generated from the L-band in the frequency domain, then this data serves as a reference for guiding noise suppression in the H-band in the time domain. This intermediary mechanism enables efficient cross-band noise suppression without requiring complex full-band frequency-domain processing.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Object-affected harmful factors

If full frequency-domain processing is applied to super-wideband signals, then noise suppression is achieved, but computational load increases significantly

Engineering Contradiction:
Improvenoise suppressionVSAvoidcomputational energy
Core Design Contradiction:
Object-affected harmful factorsVSUse of energy by moving object

Solution Approach 1:

The patent segments the super-wideband signal processing into frequency domain (L-band) and time domain (H-band) components. This segmentation enables the system to perform computationally efficient time-domain processing for the H-band while using frequency-domain processing only for the L-band to generate noise suppression data, significantly reducing overall computational energy requirements compared to full frequency-domain processing.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP2710590B1Super-wideband noise supression
Publication Date: 2015.10.07 GOOGLE LLC
  • EP2710590B1 patent drawingFigure 1
  • EP2710590B1 patent drawingFigure 2
  • EP2710590B1 patent drawingFigure 3

AI summary

A time-domain filter bank splits a super-wideband input signal into two signal streams, a low-frequency band (L-band) stream and a high-frequency band (H-band) stream. The L-band stream contains the lower frequency components of the received signal, and the H-band stream contains the higher frequency components of the signal. Noise suppression processing performed on the L-band stream in the frequency-domain generates speech/noise probability and gain filter data that is used for noise suppression processing of the H-band stream, which remains in the time-domain.