Audio Vector Indexing for Reduced-Complexity Quantization

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current algorithms for indexing and de-indexing lattice points in speech and audio codecs, such as those used in the G.EV-VBR codec, have complexity that can be further reduced to enhance coding efficiency and reduce computational demands.

Innovation Solution

The method involves applying specific processing for input vectors within pre-defined groups of leader classes and general processing for others, optimizing the indexing and de-indexing procedures by simplifying calculations based on the characteristics of leader vectors, such as those with two or four different values, to reduce computational complexity.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If general indexing algorithms are applied to all leader classes, then comprehensive coverage is achieved, but computational complexity increases

Engineering Contradiction:
Improvecomprehensive coverageVSAvoidcomputational complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent divides the set of leader classes into two segments: a first group with leader vectors having a specific number of different values (e.g., two or four distinct values), and a second group with all other leader vectors. This segmentation allows the application of simplified indexing algorithms to the first group while maintaining general algorithms for the second group, thereby reducing overall computational complexity while preserving comprehensive coverage.

Inventive Principle:
Principle #1Segmentation

2Device complexity

If simplified processing is applied to specific leader classes, then computational complexity is reduced, but processing uniformity is lost

Engineering Contradiction:
Improvecomputational complexityVSAvoidprocessing uniformity
Core Design Contradiction:
Device complexityVSEase of operation

Solution Approach 1:

The patent applies different processing qualities to different parts of the data: leader classes in the first group receive simplified indexing processing tailored to their specific characteristics (having a specific number of different values), while leader classes in the second group receive general indexing processing. This local quality approach reduces computational complexity for the majority of cases while maintaining correctness for all cases.

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP3367381B1Audio quantizing by reduced-complexity vector indexing
Publication Date: 2020.09.02 NOKIA TECHNOLOGIES OY
  • EP3367381B1 patent drawingFigure 1
  • EP3367381B1 patent drawingFigure 2~3
  • EP3367381B1 patent drawingFigure 4

AI summary

This invention relates to indexing an input vector contained in a set of vectors of a plurality of sets of vectors. The indexing comprises performing, in case that the input vector is contained in a set of vectors of a pre-defined group of one or more sets of vectors of the plurality of sets of vectors, a specific processing that is adapted to a characteristic of the sets of vectors in the pre-defined group of sets of vectors and is only applicable in case of input vectors contained in sets of vectors with the characteristic. The indexing further comprises performing, in case that the input vector is not contained in a set of vectors of the pre-defined group of sets of vectors, a general processing. The invention further relates to an according determining of a target vector contained in a set of vectors of a plurality of sets of vectors based on an index associated with said target vector.