Asymmetric Audio Windows for Low-Delay Transform Size Switching

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing audio coders face challenges in achieving low delay and high coding efficiency, particularly in handling transform size switching and window transitions, which often compromise on frequency separation or introduce additional delay.

Innovation Solution

The use of asymmetric transform windows with truncated overlap portions and smooth transitions allows for flexible transform size switching without increasing coder delay, maintaining perfect reconstruction and reducing memory requirements by storing only minimal window edges.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If transition windows are inserted between frames encoded using different transform lengths, then transform size switching is handled smoothly, but encoder look-ahead delay increases

Engineering Contradiction:
Improvetransform size switching capabilityVSAvoidencoder look-ahead delay
Core Design Contradiction:
Adaptability or versatilityVSLoss of time

Solution Approach 1:

The invention extracts and removes the transition window from the audio signal processing chain. Instead of inserting a separate transition window between frames of different transform lengths, the method directly switches between transform sizes using the same window function, thereby eliminating the additional delay caused by transition windows while maintaining smooth transform size switching capability

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The invention inverts the conventional approach by not adding a transition window but rather using the same window function for both transform sizes. This inversion of the traditional solution eliminates the need for increased look-ahead delay while still enabling flexible transform size switching between stationary and transient audio segments

Inventive Principle:
Principle #13The other way round (Inversion)

2Device complexity

If a fixed low window overlap is used for all transform sizes, then the need for transition windows is avoided, but frequency separation is reduced

Engineering Contradiction:
Improvewindow transition complexityVSAvoidfrequency separation
Core Design Contradiction:
Device complexityVSManufacturing precision

Solution Approach 1:

The invention applies local quality by using a long window overlap for stationary signal segments to maximize frequency separation, while using a short window overlap for transient signal segments to minimize delay. This local adaptation of window overlap length to the signal characteristics allows optimal frequency separation where needed without compromising transient response

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The invention introduces dynamics by making the window overlap length variable rather than fixed. The window overlap is dynamically adjusted based on the transform size and signal type: long overlap for stationary segments with larger transform sizes to improve frequency separation, and short overlap for transient segments with smaller transform sizes to reduce delay

Inventive Principle:
Principle #15Dynamics

3Loss of time

If asymmetric MDCT windows are used with shortened overlap, then look-ahead delay is reduced, but frequency separation is compromised

Engineering Contradiction:
Improvelook-ahead delayVSAvoidfrequency separation
Core Design Contradiction:
Loss of timeVSManufacturing precision

Solution Approach 1:

The invention applies dynamics by making the window overlap length variable based on the transform size and signal characteristics. For stationary segments with larger transform sizes, a long window overlap is used to maximize frequency separation. For transient segments with smaller transform sizes, a short window overlap is used to minimize look-ahead delay. This dynamic adjustment resolves the contradiction between delay reduction and frequency separation

Inventive Principle:
Principle #15Dynamics

Data Source

PatentEP3584792B1Processor, method and computer program for processing an audio signal using truncated analysis or synthesis window overlap portions
Publication Date: 2023.01.11 FRAUNHOFER GESELLSCHAFT ZUR FORDERUNG DER ANGEWANDTEN FORSCHUNG EV
  • EP3584792B1 patent drawingFigure 1A
  • EP3584792B1 patent drawingFigure 1B
  • EP3584792B1 patent drawingFigure 1C

AI summary

A processor for processing an audio signal (200), comprises: an analyzer (202) for deriving a window control signal (204) from the audio signal (200) indicating a change from a first asymmetric window (1400) to a second window (1402), or indicating a change from a third window (1450) to a fourth asymmetric window (1452), wherein the second window (1402) is shorter than the first window (1400), or wherein the third window (1450) is shorter than the fourth window (1452); a window constructor (206) for constructing the second window (1402) using a first overlap portion (800) of the first asymmetric window (1400), wherein the window constructor (206) is configured to determine a first overlap portion (1000) of the second window (1402) using a truncated first overlap portion of the first asymmetric window, or wherein the window constructor is configured to calculate a second overlap portion (1330) of the third window (1450) using a truncated second overlap portion (814) of the fourth asymmetric window (1452); and a windower (208) for applying the first and second windows or the third and fourth windows to obtain windowed audio signal portions (210).