Adaptive LPC Windowing for Better Voice Prediction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current Linear Prediction Coding (LPC) analysis techniques face suboptimal performance due to the use of fixed window functions and increased complexity from multiple rounds of analysis with different window sizes, which affects the efficiency and accuracy of voice signal compression.

Innovation Solution

An adaptive windowing method is introduced, where the window function is dynamically selected based on the amplitude values of the input voice signal's sample points, allowing for improved linear prediction performance with reduced complexity by using specific cosine-based window functions for different segments of the signal.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If a fixed window function is applied in the windowing process, then the device complexity is reduced, but the linear prediction performance deteriorates

Engineering Contradiction:
Improvecoding complexityVSAvoidlinear prediction performance
Core Design Contradiction:
Device complexityVSMeasurement precision

Solution Approach 1:

The patent applies dynamics by making the window function adaptive rather than fixed. The system dynamically selects between different window functions (first window function when amplitude exceeds threshold, second window function when amplitude does not exceed threshold) based on the input signal characteristics, thereby optimizing linear prediction performance without significantly increasing complexity

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent changes the parameter of the window function based on signal amplitude. By comparing the amplitude of the input signal with a threshold value and selecting different window functions accordingly, the system adapts the windowing parameters to match the signal characteristics, improving prediction accuracy while maintaining manageable complexity

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If two rounds of LPC analysis are performed with different window sizes, then the linear prediction performance is improved, but the device complexity increases

Engineering Contradiction:
Improvelinear prediction performanceVSAvoidanalysis complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent reduces complexity by making the analysis process dynamic and adaptive rather than performing fixed multiple rounds. The system dynamically determines whether to apply a first or second window function based on signal amplitude characteristics, achieving optimal performance with a single adaptive analysis pass instead of multiple fixed rounds

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent changes the window function parameter based on signal characteristics (amplitude comparison with threshold) to optimize the single analysis round, eliminating the need for multiple analysis passes with different window sizes while maintaining improved prediction performance

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP2407963B1Linear prediction analysis method, apparatus and system
Publication Date: 2015.05.13 HUAWEI TECH CO LTD
  • EP2407963B1 patent drawingFigure 1~2
  • EP2407963B1 patent drawingFigure 3~4
  • EP2407963B1 patent drawingFigure 5~6

AI summary

The present invention relates to communication technologies and discloses a method, an apparatus and a system for Linear Prediction Coding (LPC) analysis to improve LPC prediction performance and simplify analysis operation. The method includes: obtaining signal feature information of at least one sample point of input signals; comparing and analyzing the signal feature information to obtain an analysis result; selecting a window function according to the analysis result to perform adaptive windowing for the input signals and obtain windowed signals; and processing the windowed signals to obtain an LPC coefficient for linear prediction. The embodiments of the present invention are applicable to LPC.