Two-Channel Audio Decoding With IGF Spectral Gap Reconstruction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current audio codecs face limitations in coding efficiency and quality at low bitrates due to restricted bandwidth extension techniques, which fail to accurately align tonal harmonics and introduce computational complexity, memory requirements, and artifacts, especially in mobile devices.

Innovation Solution

The proposed Intelligent Gap Filling (IGF) technology analyzes correlation situations in both source and target ranges to identify optimal two-channel representations, allowing for waveform-preserving encoding and parametric encoding with frequency tile filling, enabling efficient reconstruction of spectral portions without transforming data into secondary domains.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If bandwidth extension techniques are used to code wide-band signals at low bitrates, then coding efficiency is improved, but manufacturing precision deteriorates due to inaccurate alignment of tonal harmonics

Engineering Contradiction:
Improvecoding efficiencyVSAvoidalignment accuracy of tonal harmonics
Core Design Contradiction:
ProductivityVSManufacturing precision

Solution Approach 1:

The patent replaces traditional mechanical spectral patching methods with a neural network-based system. The neural network automatically identifies and aligns tonal harmonics across frequency bands, substituting the manual or rule-based mechanical alignment process with an intelligent system that learns optimal alignment from data, thereby maintaining high precision at low bitrates

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

Solution Approach 2:

The patent changes the parameter representation from fixed spectral patches to dynamic neural network parameters. The neural network adjusts alignment parameters adaptively based on the input signal characteristics, allowing precise tonal harmonic alignment while maintaining coding efficiency at low bitrates

Inventive Principle:
Principle #35Parameter changes

2Reliability

If traditional bandwidth extension methods are applied, then audio quality is maintained, but device complexity increases due to additional computational requirements

Engineering Contradiction:
Improveaudio qualityVSAvoidcomputational complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent uses a neural network model that has been trained to copy and reconstruct high-frequency content from low-frequency inputs. Instead of complex real-time processing, the pre-trained neural network efficiently generates high-frequency components by learning from training data, reducing computational complexity while maintaining audio quality

Inventive Principle:
Principle #26Copying

Solution Approach 2:

The neural network is pre-trained offline on large audio datasets before deployment. This preliminary action transfers complex learning to the training phase, allowing the deployed system to perform simple inference operations that maintain audio quality without requiring complex computational resources in the actual audio coding process

Inventive Principle:
Principle #10Preliminary action

3Productivity

If spectral patching is used to fill high-frequency regions, then bandwidth extension is achieved, but harmful factors increase due to introduction of artifacts

Engineering Contradiction:
Improvebandwidth extension capabilityVSAvoidaudio artifacts
Core Design Contradiction:
ProductivityVSObject-generated harmful factors

Solution Approach 1:

The patent replaces traditional mechanical spectral patching with neural network-based synthesis. The neural network learns natural spectral transitions and harmonics from training data, generating smooth continuations that avoid the discontinuities and artifacts inherent in rigid spectral copying methods

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

Solution Approach 2:

The patent transitions from fixed spectral patch parameters to dynamic neural network-generated parameters. The neural network adjusts spectral parameters adaptively to match the local signal characteristics, ensuring smooth transitions and natural-sounding high-frequency extensions without artificial artifacts

Inventive Principle:
Principle #35Parameter changes

4Adaptability or versatility

If data is transformed into secondary domains for bandwidth extension, then processing flexibility is improved, but device complexity increases due to additional transformation stages

Engineering Contradiction:
Improveprocessing flexibilityVSAvoidtransformation complexity
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent implements a universal neural network architecture that handles multiple audio processing functions within a single framework. The same neural network performs both bandwidth extension and tonal harmonic alignment, eliminating the need for separate transformation stages and reducing overall system complexity while maintaining processing flexibility

Inventive Principle:
Principle #6Universality (Multi-functionality)

Data Source

PatentEP3025328B1Audio decoder and related method using two-channel processing within an intelligent gap filling framework
Publication Date: 2018.08.01 FRAUNHOFER GESELLSCHAFT ZUR FORDERUNG DER ANGEWANDTEN FORSCHUNG EV
  • EP3025328B1 patent drawingFigure 1A~1B
  • EP3025328B1 patent drawingFigure 2A
  • EP3025328B1 patent drawingFigure 2B

AI summary

An apparatus for generating a decoded two-channel signal, comprises: an audio processor (802) for decoding an encoded two-channel signal to obtain a first set of first spectral portions; a parametric decoder (804) for providing parametric data for a second set of second spectral portions and a two-channel identification identifying either a first or a second different two-channel representation for the second spectral portions; and a frequency regenerator (806) for regenerating a second spectral portion depending on a first spectral portion of the first set of first spectral portions, the parametric data for the second portion and the two-channel identification for the second portion.