Sound Signal Shape Parameter Estimation for Easier Classification

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current techniques for sound signal processing, such as those using LSP parameters, face difficulties in direct application for sound classification and section estimation due to the complexity of handling multiple values, making it challenging for efficient encoding and decoding processes.

Innovation Solution

A parameter determination device and method that estimates the shape parameter η for a generalized Gaussian distribution, which simplifies the encoding and decoding process by transforming the frequency domain sample sequence into a whitened spectral sequence, allowing for easier sound classification and section estimation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If LSP parameters are used for sound signal processing, then sound characteristics can be represented, but the complexity of handling multiple values makes it difficult to apply directly for sound classification and section estimation

Engineering Contradiction:
Improvesound characteristic representationVSAvoidprocess complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent extracts a single shape parameter η from the frequency domain signal that characterizes the distribution of spectral coefficients. This extracted parameter replaces the need to process multiple LSP values directly, simplifying the classification and section estimation processes while retaining the essential sound characteristic information.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent transforms the representation from multiple LSP parameters to a single shape parameter η that describes the generalized Gaussian distribution of spectral coefficients. This parameter change simplifies the data structure and makes threshold-based processing feasible for sound classification applications.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If multiple LSP values are used, then sound characteristics are captured, but it is not easy to perform threshold-based processes

Engineering Contradiction:
Improvesound characteristic captureVSAvoidthreshold process ease
Core Design Contradiction:
Measurement precisionVSEase of operation

Solution Approach 1:

The patent changes the parameter representation from multiple discrete LSP values to a single continuous shape parameter η that characterizes the distribution shape. This enables straightforward threshold-based classification operations while maintaining the ability to capture essential sound characteristics through the distribution shape.

Inventive Principle:
Principle #35Parameter changes

3Ease of operation

If a single value parameter is used, then threshold-based processes become easier, but the ability to represent complex sound characteristics may be reduced

Engineering Contradiction:
Improvethreshold process easeVSAvoidsound characteristic representation
Core Design Contradiction:
Ease of operationVSMeasurement precision

Solution Approach 1:

The patent uses the shape parameter η of the generalized Gaussian distribution to represent the distribution shape of spectral coefficients. This single parameter effectively captures the essential characteristics of the sound signal's spectral distribution, enabling both easy threshold-based processing and accurate sound characteristic representation.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentUS10276186B2Parameter determination device, method, program and recording medium for determining a parameter indicating a characteristic of sound signal
Publication Date: 2019.04.30 NIPPON TELEGRAPH & TELEPHONE CORP
  • US10276186B2 patent drawing
  • US10276186B2 patent drawing
  • US10276186B2 patent drawing

AI summary

A parameter determination device includes: a spectral envelope estimating portion performing estimation of a spectral envelope using a parameter η0 specified in a predetermined method, regarding the η0-th power of absolute values of a frequency domain sample sequence corresponding to a time-series signal as a power spectrum on the assumption that the parameter η0 and a parameter η are positive numbers; a whitened spectral sequence generating portion obtaining a whitened spectral sequence which is a sequence obtained by dividing the frequency domain sample sequence by the spectral envelope; and a parameter acquiring portion determining such a parameter η that generalized Gaussian distribution with the parameter η as a shape parameter approximates a histogram of the whitened spectral sequence.